Preserve UTF-8 encoding in JSON files (ensure_ascii=False)

Signed-off-by: bigcat88 <bigcat88@icloud.com>

app/node_replace_manager.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+import logging
+
 from aiohttp import web
 
 from typing import TYPE_CHECKING, TypedDict
@@ -31,8 +33,22 @@ class NodeReplaceManager:
         self._replacements: dict[str, list[NodeReplace]] = {}
 
     def register(self, node_replace: NodeReplace):
-        """Register a node replacement mapping."""
-        self._replacements.setdefault(node_replace.old_node_id, []).append(node_replace)
+        """Register a node replacement mapping.
+
+        Idempotent: if a replacement with the same (old_node_id, new_node_id)
+        is already registered, the duplicate is ignored. This prevents stale
+        entries from accumulating when custom nodes are reloaded in the same
+        process (e.g. via ComfyUI-Manager).
+        """
+        existing = self._replacements.setdefault(node_replace.old_node_id, [])
+        for entry in existing:
+            if entry.new_node_id == node_replace.new_node_id:
+                logging.debug(
+                    "Node replacement %s -> %s already registered, ignoring duplicate.",
+                    node_replace.old_node_id, node_replace.new_node_id,
+                )
+                return
+        existing.append(node_replace)
 
     def get_replacement(self, old_node_id: str) -> list[NodeReplace] | None:
         """Get replacements for an old node ID."""

blueprints/Canny to Image (Z-Image-Turbo).json

@@ -162,7 +162,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Canny to Image (Z-Image-Turbo)",
+        "name": "Canny to Image (Z-Image-Turbo)",
         "inputNode": {
           "id": -10,
           "bounding": [
@@ -1554,7 +1554,7 @@
           "VHS_KeepIntermediate": true
         },
         "category": "Image generation and editing/Canny to image",
-        "description": "Generates an image from a Canny edge map using Z-Image-Turbo, with text conditioning and fast inference."
+        "description": "Generates an image from a Canny edge map using Z-Image-Turbo, with text conditioning."
       }
     ]
   },

blueprints/Canny to Video (LTX 2.0).json

@@ -192,7 +192,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Canny to Video (LTX 2.0)",
+        "name": "Canny to Video (LTX 2.0)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Chromatic Aberration.json

@@ -232,7 +232,7 @@
               "Node name for S&R": "GLSLShader"
             },
             "widgets_values": [
-              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform int u_int0;      // Mode\nuniform float u_float0;  // Amount (0 to 100)\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst int MODE_LINEAR   = 0;\nconst int MODE_RADIAL   = 1;\nconst int MODE_BARREL   = 2;\nconst int MODE_SWIRL    = 3;\nconst int MODE_DIAGONAL = 4;\n\nconst float AMOUNT_SCALE = 0.0005;\nconst float RADIAL_MULT = 4.0;\nconst float BARREL_MULT = 8.0;\nconst float INV_SQRT2 = 0.70710678118;\n\nvoid main() {\n    vec2 uv = v_texCoord;\n    vec4 original = texture(u_image0, uv);\n\n    float amount = u_float0 * AMOUNT_SCALE;\n\n    if (amount < 0.000001) {\n        fragColor = original;\n        return;\n    }\n\n    // Aspect-corrected coordinates for circular effects\n    float aspect = u_resolution.x / u_resolution.y;\n    vec2 centered = uv - 0.5;\n    vec2 corrected = vec2(centered.x * aspect, centered.y);\n    float r = length(corrected);\n    vec2 dir = r > 0.0001 ? corrected / r : vec2(0.0);\n    vec2 offset = vec2(0.0);\n\n    if (u_int0 == MODE_LINEAR) {\n        // Horizontal shift (no aspect correction needed)\n        offset = vec2(amount, 0.0);\n    }\n    else if (u_int0 == MODE_RADIAL) {\n        // Outward from center, stronger at edges\n        offset = dir * r * amount * RADIAL_MULT;\n        offset.x /= aspect;  // Convert back to UV space\n    }\n    else if (u_int0 == MODE_BARREL) {\n        // Lens distortion simulation (r\u00b2 falloff)\n        offset = dir * r * r * amount * BARREL_MULT;\n        offset.x /= aspect;  // Convert back to UV space\n    }\n    else if (u_int0 == MODE_SWIRL) {\n        // Perpendicular to radial (rotational aberration)\n        vec2 perp = vec2(-dir.y, dir.x);\n        offset = perp * r * amount * RADIAL_MULT;\n        offset.x /= aspect;  // Convert back to UV space\n    }\n    else if (u_int0 == MODE_DIAGONAL) {\n        // 45\u00b0 offset (no aspect correction needed)\n        offset = vec2(amount, amount) * INV_SQRT2;\n    }\n    \n    float red = texture(u_image0, uv + offset).r;\n    float green = original.g;\n    float blue = texture(u_image0, uv - offset).b;\n    \n    fragColor = vec4(red, green, blue, original.a);\n}",
+              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform int u_int0;      // Mode\nuniform float u_float0;  // Amount (0 to 100)\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst int MODE_LINEAR   = 0;\nconst int MODE_RADIAL   = 1;\nconst int MODE_BARREL   = 2;\nconst int MODE_SWIRL    = 3;\nconst int MODE_DIAGONAL = 4;\n\nconst float AMOUNT_SCALE = 0.0005;\nconst float RADIAL_MULT = 4.0;\nconst float BARREL_MULT = 8.0;\nconst float INV_SQRT2 = 0.70710678118;\n\nvoid main() {\n    vec2 uv = v_texCoord;\n    vec4 original = texture(u_image0, uv);\n\n    float amount = u_float0 * AMOUNT_SCALE;\n\n    if (amount < 0.000001) {\n        fragColor = original;\n        return;\n    }\n\n    // Aspect-corrected coordinates for circular effects\n    float aspect = u_resolution.x / u_resolution.y;\n    vec2 centered = uv - 0.5;\n    vec2 corrected = vec2(centered.x * aspect, centered.y);\n    float r = length(corrected);\n    vec2 dir = r > 0.0001 ? corrected / r : vec2(0.0);\n    vec2 offset = vec2(0.0);\n\n    if (u_int0 == MODE_LINEAR) {\n        // Horizontal shift (no aspect correction needed)\n        offset = vec2(amount, 0.0);\n    }\n    else if (u_int0 == MODE_RADIAL) {\n        // Outward from center, stronger at edges\n        offset = dir * r * amount * RADIAL_MULT;\n        offset.x /= aspect;  // Convert back to UV space\n    }\n    else if (u_int0 == MODE_BARREL) {\n        // Lens distortion simulation (r² falloff)\n        offset = dir * r * r * amount * BARREL_MULT;\n        offset.x /= aspect;  // Convert back to UV space\n    }\n    else if (u_int0 == MODE_SWIRL) {\n        // Perpendicular to radial (rotational aberration)\n        vec2 perp = vec2(-dir.y, dir.x);\n        offset = perp * r * amount * RADIAL_MULT;\n        offset.x /= aspect;  // Convert back to UV space\n    }\n    else if (u_int0 == MODE_DIAGONAL) {\n        // 45° offset (no aspect correction needed)\n        offset = vec2(amount, amount) * INV_SQRT2;\n    }\n    \n    float red = texture(u_image0, uv + offset).r;\n    float green = original.g;\n    float blue = texture(u_image0, uv - offset).b;\n    \n    fragColor = vec4(red, green, blue, original.a);\n}",
               "from_input"
             ]
           },

blueprints/Color Adjustment.json

@@ -256,7 +256,7 @@
               "Node name for S&R": "GLSLShader"
             },
             "widgets_values": [
-              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0; // temperature (-100 to 100)\nuniform float u_float1; // tint (-100 to 100)\nuniform float u_float2; // vibrance (-100 to 100)\nuniform float u_float3; // saturation (-100 to 100)\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst float INPUT_SCALE = 0.01;\nconst float TEMP_TINT_PRIMARY = 0.3;\nconst float TEMP_TINT_SECONDARY = 0.15;\nconst float VIBRANCE_BOOST = 2.0;\nconst float SATURATION_BOOST = 2.0;\nconst float SKIN_PROTECTION = 0.5;\nconst float EPSILON = 0.001;\nconst vec3 LUMA_WEIGHTS = vec3(0.299, 0.587, 0.114);\n\nvoid main() {\n    vec4 tex = texture(u_image0, v_texCoord);\n    vec3 color = tex.rgb;\n    \n    // Scale inputs: -100/100 \u2192 -1/1\n    float temperature = u_float0 * INPUT_SCALE;\n    float tint = u_float1 * INPUT_SCALE;\n    float vibrance = u_float2 * INPUT_SCALE;\n    float saturation = u_float3 * INPUT_SCALE;\n    \n    // Temperature (warm/cool): positive = warm, negative = cool\n    color.r += temperature * TEMP_TINT_PRIMARY;\n    color.b -= temperature * TEMP_TINT_PRIMARY;\n    \n    // Tint (green/magenta): positive = green, negative = magenta\n    color.g += tint * TEMP_TINT_PRIMARY;\n    color.r -= tint * TEMP_TINT_SECONDARY;\n    color.b -= tint * TEMP_TINT_SECONDARY;\n    \n    // Single clamp after temperature/tint\n    color = clamp(color, 0.0, 1.0);\n    \n    // Vibrance with skin protection\n    if (vibrance != 0.0) {\n        float maxC = max(color.r, max(color.g, color.b));\n        float minC = min(color.r, min(color.g, color.b));\n        float sat = maxC - minC;\n        float gray = dot(color, LUMA_WEIGHTS);\n        \n        if (vibrance < 0.0) {\n            // Desaturate: -100 \u2192 gray\n            color = mix(vec3(gray), color, 1.0 + vibrance);\n        } else {\n            // Boost less saturated colors more\n            float vibranceAmt = vibrance * (1.0 - sat);\n            \n            // Branchless skin tone protection\n            float isWarmTone = step(color.b, color.g) * step(color.g, color.r);\n            float warmth = (color.r - color.b) / max(maxC, EPSILON);\n            float skinTone = isWarmTone * warmth * sat * (1.0 - sat);\n            vibranceAmt *= (1.0 - skinTone * SKIN_PROTECTION);\n            \n            color = mix(vec3(gray), color, 1.0 + vibranceAmt * VIBRANCE_BOOST);\n        }\n    }\n    \n    // Saturation\n    if (saturation != 0.0) {\n        float gray = dot(color, LUMA_WEIGHTS);\n        float satMix = saturation < 0.0\n            ? 1.0 + saturation                      // -100 \u2192 gray\n            : 1.0 + saturation * SATURATION_BOOST;  // +100 \u2192 3x boost\n        color = mix(vec3(gray), color, satMix);\n    }\n    \n    fragColor = vec4(clamp(color, 0.0, 1.0), tex.a);\n}",
+              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0; // temperature (-100 to 100)\nuniform float u_float1; // tint (-100 to 100)\nuniform float u_float2; // vibrance (-100 to 100)\nuniform float u_float3; // saturation (-100 to 100)\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst float INPUT_SCALE = 0.01;\nconst float TEMP_TINT_PRIMARY = 0.3;\nconst float TEMP_TINT_SECONDARY = 0.15;\nconst float VIBRANCE_BOOST = 2.0;\nconst float SATURATION_BOOST = 2.0;\nconst float SKIN_PROTECTION = 0.5;\nconst float EPSILON = 0.001;\nconst vec3 LUMA_WEIGHTS = vec3(0.299, 0.587, 0.114);\n\nvoid main() {\n    vec4 tex = texture(u_image0, v_texCoord);\n    vec3 color = tex.rgb;\n    \n    // Scale inputs: -100/100 → -1/1\n    float temperature = u_float0 * INPUT_SCALE;\n    float tint = u_float1 * INPUT_SCALE;\n    float vibrance = u_float2 * INPUT_SCALE;\n    float saturation = u_float3 * INPUT_SCALE;\n    \n    // Temperature (warm/cool): positive = warm, negative = cool\n    color.r += temperature * TEMP_TINT_PRIMARY;\n    color.b -= temperature * TEMP_TINT_PRIMARY;\n    \n    // Tint (green/magenta): positive = green, negative = magenta\n    color.g += tint * TEMP_TINT_PRIMARY;\n    color.r -= tint * TEMP_TINT_SECONDARY;\n    color.b -= tint * TEMP_TINT_SECONDARY;\n    \n    // Single clamp after temperature/tint\n    color = clamp(color, 0.0, 1.0);\n    \n    // Vibrance with skin protection\n    if (vibrance != 0.0) {\n        float maxC = max(color.r, max(color.g, color.b));\n        float minC = min(color.r, min(color.g, color.b));\n        float sat = maxC - minC;\n        float gray = dot(color, LUMA_WEIGHTS);\n        \n        if (vibrance < 0.0) {\n            // Desaturate: -100 → gray\n            color = mix(vec3(gray), color, 1.0 + vibrance);\n        } else {\n            // Boost less saturated colors more\n            float vibranceAmt = vibrance * (1.0 - sat);\n            \n            // Branchless skin tone protection\n            float isWarmTone = step(color.b, color.g) * step(color.g, color.r);\n            float warmth = (color.r - color.b) / max(maxC, EPSILON);\n            float skinTone = isWarmTone * warmth * sat * (1.0 - sat);\n            vibranceAmt *= (1.0 - skinTone * SKIN_PROTECTION);\n            \n            color = mix(vec3(gray), color, 1.0 + vibranceAmt * VIBRANCE_BOOST);\n        }\n    }\n    \n    // Saturation\n    if (saturation != 0.0) {\n        float gray = dot(color, LUMA_WEIGHTS);\n        float satMix = saturation < 0.0\n            ? 1.0 + saturation                      // -100 → gray\n            : 1.0 + saturation * SATURATION_BOOST;  // +100 → 3x boost\n        color = mix(vec3(gray), color, satMix);\n    }\n    \n    fragColor = vec4(clamp(color, 0.0, 1.0), tex.a);\n}",
               "from_input"
             ]
           },

blueprints/Crop Images 2x2.json

@@ -1610,7 +1610,7 @@
         ],
         "extra": {},
         "category": "Image Tools/Crop",
-        "description": "Splits an image into a 2\u00d72 grid of four equal tiles."
+        "description": "Splits an image into a 2×2 grid of four equal tiles."
       }
     ]
   },

blueprints/Crop Images 3x3.json

@@ -368,7 +368,7 @@
                 ]
               }
             ],
-            "title": "Math Expression \uff08Width\uff09",
+            "title": "Math Expression （Width）",
             "properties": {
               "ue_properties": {
                 "widget_ue_connectable": {},
@@ -2947,7 +2947,7 @@
         ],
         "extra": {},
         "category": "Image Tools/Crop",
-        "description": "Splits an image into a 3\u00d73 grid of nine equal tiles."
+        "description": "Splits an image into a 3×3 grid of nine equal tiles."
       }
     ]
   },

blueprints/Edge-Preserving Blur.json

@@ -252,7 +252,7 @@
               "Node name for S&R": "GLSLShader"
             },
             "widgets_values": [
-              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0;   // Blur radius (0\u201320, default ~5)\nuniform float u_float1;   // Edge threshold (0\u2013100, default ~30)\nuniform int u_int0;       // Step size (0/1 = every pixel, 2+ = skip pixels)\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst int MAX_RADIUS = 20;\nconst float EPSILON = 0.0001;\n\n// Perceptual luminance\nfloat getLuminance(vec3 rgb) {\n    return dot(rgb, vec3(0.299, 0.587, 0.114));\n}\n\nvec4 bilateralFilter(vec2 uv, vec2 texelSize, int radius,\n                     float sigmaSpatial, float sigmaColor)\n{\n    vec4 center = texture(u_image0, uv);\n    vec3 centerRGB = center.rgb;\n\n    float invSpatial2 = -0.5 / (sigmaSpatial * sigmaSpatial);\n    float invColor2   = -0.5 / (sigmaColor * sigmaColor + EPSILON);\n\n    vec3 sumRGB = vec3(0.0);\n    float sumWeight = 0.0;\n\n    int step = max(u_int0, 1);\n    float radius2 = float(radius * radius);\n\n    for (int dy = -MAX_RADIUS; dy <= MAX_RADIUS; dy++) {\n        if (dy < -radius || dy > radius) continue;\n        if (abs(dy) % step != 0) continue;\n\n        for (int dx = -MAX_RADIUS; dx <= MAX_RADIUS; dx++) {\n            if (dx < -radius || dx > radius) continue;\n            if (abs(dx) % step != 0) continue;\n\n            vec2 offset = vec2(float(dx), float(dy));\n            float dist2 = dot(offset, offset);\n            if (dist2 > radius2) continue;\n\n            vec3 sampleRGB = texture(u_image0, uv + offset * texelSize).rgb;\n\n            // Spatial Gaussian\n            float spatialWeight = exp(dist2 * invSpatial2);\n\n            // Perceptual color distance (weighted RGB)\n            vec3 diff = sampleRGB - centerRGB;\n            float colorDist = dot(diff * diff, vec3(0.299, 0.587, 0.114));\n            float colorWeight = exp(colorDist * invColor2);\n\n            float w = spatialWeight * colorWeight;\n            sumRGB += sampleRGB * w;\n            sumWeight += w;\n        }\n    }\n\n    vec3 resultRGB = sumRGB / max(sumWeight, EPSILON);\n    return vec4(resultRGB, center.a); // preserve center alpha\n}\n\nvoid main() {\n    vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));\n\n    float radiusF = clamp(u_float0, 0.0, float(MAX_RADIUS));\n    int radius = int(radiusF + 0.5);\n\n    if (radius == 0) {\n        fragColor = texture(u_image0, v_texCoord);\n        return;\n    }\n\n    // Edge threshold \u2192 color sigma\n    // Squared curve for better low-end control\n    float t = clamp(u_float1, 0.0, 100.0) / 100.0;\n    t *= t;\n    float sigmaColor = mix(0.01, 0.5, t);\n\n    // Spatial sigma tied to radius\n    float sigmaSpatial = max(radiusF * 0.75, 0.5);\n\n    fragColor = bilateralFilter(\n        v_texCoord,\n        texelSize,\n        radius,\n        sigmaSpatial,\n        sigmaColor\n    );\n}",
+              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0;   // Blur radius (0–20, default ~5)\nuniform float u_float1;   // Edge threshold (0–100, default ~30)\nuniform int u_int0;       // Step size (0/1 = every pixel, 2+ = skip pixels)\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst int MAX_RADIUS = 20;\nconst float EPSILON = 0.0001;\n\n// Perceptual luminance\nfloat getLuminance(vec3 rgb) {\n    return dot(rgb, vec3(0.299, 0.587, 0.114));\n}\n\nvec4 bilateralFilter(vec2 uv, vec2 texelSize, int radius,\n                     float sigmaSpatial, float sigmaColor)\n{\n    vec4 center = texture(u_image0, uv);\n    vec3 centerRGB = center.rgb;\n\n    float invSpatial2 = -0.5 / (sigmaSpatial * sigmaSpatial);\n    float invColor2   = -0.5 / (sigmaColor * sigmaColor + EPSILON);\n\n    vec3 sumRGB = vec3(0.0);\n    float sumWeight = 0.0;\n\n    int step = max(u_int0, 1);\n    float radius2 = float(radius * radius);\n\n    for (int dy = -MAX_RADIUS; dy <= MAX_RADIUS; dy++) {\n        if (dy < -radius || dy > radius) continue;\n        if (abs(dy) % step != 0) continue;\n\n        for (int dx = -MAX_RADIUS; dx <= MAX_RADIUS; dx++) {\n            if (dx < -radius || dx > radius) continue;\n            if (abs(dx) % step != 0) continue;\n\n            vec2 offset = vec2(float(dx), float(dy));\n            float dist2 = dot(offset, offset);\n            if (dist2 > radius2) continue;\n\n            vec3 sampleRGB = texture(u_image0, uv + offset * texelSize).rgb;\n\n            // Spatial Gaussian\n            float spatialWeight = exp(dist2 * invSpatial2);\n\n            // Perceptual color distance (weighted RGB)\n            vec3 diff = sampleRGB - centerRGB;\n            float colorDist = dot(diff * diff, vec3(0.299, 0.587, 0.114));\n            float colorWeight = exp(colorDist * invColor2);\n\n            float w = spatialWeight * colorWeight;\n            sumRGB += sampleRGB * w;\n            sumWeight += w;\n        }\n    }\n\n    vec3 resultRGB = sumRGB / max(sumWeight, EPSILON);\n    return vec4(resultRGB, center.a); // preserve center alpha\n}\n\nvoid main() {\n    vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));\n\n    float radiusF = clamp(u_float0, 0.0, float(MAX_RADIUS));\n    int radius = int(radiusF + 0.5);\n\n    if (radius == 0) {\n        fragColor = texture(u_image0, v_texCoord);\n        return;\n    }\n\n    // Edge threshold → color sigma\n    // Squared curve for better low-end control\n    float t = clamp(u_float1, 0.0, 100.0) / 100.0;\n    t *= t;\n    float sigmaColor = mix(0.01, 0.5, t);\n\n    // Spatial sigma tied to radius\n    float sigmaSpatial = max(radiusF * 0.75, 0.5);\n\n    fragColor = bilateralFilter(\n        v_texCoord,\n        texelSize,\n        radius,\n        sigmaSpatial,\n        sigmaColor\n    );\n}",
               "from_input"
             ]
           },

blueprints/Film Grain.json

@@ -268,7 +268,7 @@
               "Node name for S&R": "GLSLShader"
             },
             "widgets_values": [
-              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0; // grain amount      [0.0 \u2013 1.0]   typical: 0.2\u20130.8\nuniform float u_float1; // grain size        [0.3 \u2013 3.0]   lower = finer grain\nuniform float u_float2; // color amount      [0.0 \u2013 1.0]   0 = monochrome, 1 = RGB grain\nuniform float u_float3; // luminance bias    [0.0 \u2013 1.0]   0 = uniform, 1 = shadows only\nuniform int   u_int0;   // noise mode        [0 or 1]      0 = smooth, 1 = grainy\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\n// High-quality integer hash (pcg-like)\nuint pcg(uint v) {\n    uint state = v * 747796405u + 2891336453u;\n    uint word = ((state >> ((state >> 28u) + 4u)) ^ state) * 277803737u;\n    return (word >> 22u) ^ word;\n}\n\n// 2D -> 1D hash input\nuint hash2d(uvec2 p) {\n    return pcg(p.x + pcg(p.y));\n}\n\n// Hash to float [0, 1]\nfloat hashf(uvec2 p) {\n    return float(hash2d(p)) / float(0xffffffffu);\n}\n\n// Hash to float with offset (for RGB channels)\nfloat hashf(uvec2 p, uint offset) {\n    return float(pcg(hash2d(p) + offset)) / float(0xffffffffu);\n}\n\n// Convert uniform [0,1] to roughly Gaussian distribution\n// Using simple approximation: average of multiple samples\nfloat toGaussian(uvec2 p) {\n    float sum = hashf(p, 0u) + hashf(p, 1u) + hashf(p, 2u) + hashf(p, 3u);\n    return (sum - 2.0) * 0.7;  // Centered, scaled\n}\n\nfloat toGaussian(uvec2 p, uint offset) {\n    float sum = hashf(p, offset) + hashf(p, offset + 1u) \n              + hashf(p, offset + 2u) + hashf(p, offset + 3u);\n    return (sum - 2.0) * 0.7;\n}\n\n// Smooth noise with better interpolation\nfloat smoothNoise(vec2 p) {\n    vec2 i = floor(p);\n    vec2 f = fract(p);\n    \n    // Quintic interpolation (less banding than cubic)\n    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);\n    \n    uvec2 ui = uvec2(i);\n    float a = toGaussian(ui);\n    float b = toGaussian(ui + uvec2(1u, 0u));\n    float c = toGaussian(ui + uvec2(0u, 1u));\n    float d = toGaussian(ui + uvec2(1u, 1u));\n    \n    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);\n}\n\nfloat smoothNoise(vec2 p, uint offset) {\n    vec2 i = floor(p);\n    vec2 f = fract(p);\n    \n    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);\n    \n    uvec2 ui = uvec2(i);\n    float a = toGaussian(ui, offset);\n    float b = toGaussian(ui + uvec2(1u, 0u), offset);\n    float c = toGaussian(ui + uvec2(0u, 1u), offset);\n    float d = toGaussian(ui + uvec2(1u, 1u), offset);\n    \n    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);\n}\n\nvoid main() {\n    vec4 color = texture(u_image0, v_texCoord);\n    \n    // Luminance (Rec.709)\n    float luma = dot(color.rgb, vec3(0.2126, 0.7152, 0.0722));\n    \n    // Grain UV (resolution-independent)\n    vec2 grainUV = v_texCoord * u_resolution / max(u_float1, 0.01);\n    uvec2 grainPixel = uvec2(grainUV);\n    \n    float g;\n    vec3 grainRGB;\n    \n    if (u_int0 == 1) {\n        // Grainy mode: pure hash noise (no interpolation = no banding)\n        g = toGaussian(grainPixel);\n        grainRGB = vec3(\n            toGaussian(grainPixel, 100u),\n            toGaussian(grainPixel, 200u),\n            toGaussian(grainPixel, 300u)\n        );\n    } else {\n        // Smooth mode: interpolated with quintic curve\n        g = smoothNoise(grainUV);\n        grainRGB = vec3(\n            smoothNoise(grainUV, 100u),\n            smoothNoise(grainUV, 200u),\n            smoothNoise(grainUV, 300u)\n        );\n    }\n    \n    // Luminance weighting (less grain in highlights)\n    float lumWeight = mix(1.0, 1.0 - luma, clamp(u_float3, 0.0, 1.0));\n    \n    // Strength\n    float strength = u_float0 * 0.15;\n    \n    // Color vs monochrome grain\n    vec3 grainColor = mix(vec3(g), grainRGB, clamp(u_float2, 0.0, 1.0));\n    \n    color.rgb += grainColor * strength * lumWeight;\n    fragColor0 = vec4(clamp(color.rgb, 0.0, 1.0), color.a);\n}\n",
+              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0; // grain amount      [0.0 – 1.0]   typical: 0.2–0.8\nuniform float u_float1; // grain size        [0.3 – 3.0]   lower = finer grain\nuniform float u_float2; // color amount      [0.0 – 1.0]   0 = monochrome, 1 = RGB grain\nuniform float u_float3; // luminance bias    [0.0 – 1.0]   0 = uniform, 1 = shadows only\nuniform int   u_int0;   // noise mode        [0 or 1]      0 = smooth, 1 = grainy\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\n// High-quality integer hash (pcg-like)\nuint pcg(uint v) {\n    uint state = v * 747796405u + 2891336453u;\n    uint word = ((state >> ((state >> 28u) + 4u)) ^ state) * 277803737u;\n    return (word >> 22u) ^ word;\n}\n\n// 2D -> 1D hash input\nuint hash2d(uvec2 p) {\n    return pcg(p.x + pcg(p.y));\n}\n\n// Hash to float [0, 1]\nfloat hashf(uvec2 p) {\n    return float(hash2d(p)) / float(0xffffffffu);\n}\n\n// Hash to float with offset (for RGB channels)\nfloat hashf(uvec2 p, uint offset) {\n    return float(pcg(hash2d(p) + offset)) / float(0xffffffffu);\n}\n\n// Convert uniform [0,1] to roughly Gaussian distribution\n// Using simple approximation: average of multiple samples\nfloat toGaussian(uvec2 p) {\n    float sum = hashf(p, 0u) + hashf(p, 1u) + hashf(p, 2u) + hashf(p, 3u);\n    return (sum - 2.0) * 0.7;  // Centered, scaled\n}\n\nfloat toGaussian(uvec2 p, uint offset) {\n    float sum = hashf(p, offset) + hashf(p, offset + 1u) \n              + hashf(p, offset + 2u) + hashf(p, offset + 3u);\n    return (sum - 2.0) * 0.7;\n}\n\n// Smooth noise with better interpolation\nfloat smoothNoise(vec2 p) {\n    vec2 i = floor(p);\n    vec2 f = fract(p);\n    \n    // Quintic interpolation (less banding than cubic)\n    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);\n    \n    uvec2 ui = uvec2(i);\n    float a = toGaussian(ui);\n    float b = toGaussian(ui + uvec2(1u, 0u));\n    float c = toGaussian(ui + uvec2(0u, 1u));\n    float d = toGaussian(ui + uvec2(1u, 1u));\n    \n    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);\n}\n\nfloat smoothNoise(vec2 p, uint offset) {\n    vec2 i = floor(p);\n    vec2 f = fract(p);\n    \n    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);\n    \n    uvec2 ui = uvec2(i);\n    float a = toGaussian(ui, offset);\n    float b = toGaussian(ui + uvec2(1u, 0u), offset);\n    float c = toGaussian(ui + uvec2(0u, 1u), offset);\n    float d = toGaussian(ui + uvec2(1u, 1u), offset);\n    \n    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);\n}\n\nvoid main() {\n    vec4 color = texture(u_image0, v_texCoord);\n    \n    // Luminance (Rec.709)\n    float luma = dot(color.rgb, vec3(0.2126, 0.7152, 0.0722));\n    \n    // Grain UV (resolution-independent)\n    vec2 grainUV = v_texCoord * u_resolution / max(u_float1, 0.01);\n    uvec2 grainPixel = uvec2(grainUV);\n    \n    float g;\n    vec3 grainRGB;\n    \n    if (u_int0 == 1) {\n        // Grainy mode: pure hash noise (no interpolation = no banding)\n        g = toGaussian(grainPixel);\n        grainRGB = vec3(\n            toGaussian(grainPixel, 100u),\n            toGaussian(grainPixel, 200u),\n            toGaussian(grainPixel, 300u)\n        );\n    } else {\n        // Smooth mode: interpolated with quintic curve\n        g = smoothNoise(grainUV);\n        grainRGB = vec3(\n            smoothNoise(grainUV, 100u),\n            smoothNoise(grainUV, 200u),\n            smoothNoise(grainUV, 300u)\n        );\n    }\n    \n    // Luminance weighting (less grain in highlights)\n    float lumWeight = mix(1.0, 1.0 - luma, clamp(u_float3, 0.0, 1.0));\n    \n    // Strength\n    float strength = u_float0 * 0.15;\n    \n    // Color vs monochrome grain\n    vec3 grainColor = mix(vec3(g), grainRGB, clamp(u_float2, 0.0, 1.0));\n    \n    color.rgb += grainColor * strength * lumWeight;\n    fragColor0 = vec4(clamp(color.rgb, 0.0, 1.0), color.a);\n}\n",
               "from_input"
             ]
           },

blueprints/Image Edit (Qwen 2511).json

@@ -132,7 +132,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Image Edit (Qwen 2511)",
+        "name": "Image Edit (Qwen 2511)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Image Upscale(Z-image-Turbo).json

@@ -141,7 +141,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Image Upscale(Z-image-Turbo)",
+        "name": "Image Upscale(Z-image-Turbo)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Image to Depth Map (Lotus).json

@@ -99,7 +99,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Image to Depth Map (Lotus)",
+        "name": "Image to Depth Map (Lotus)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Image to Model (Hunyuan3d 2.1).json

@@ -72,7 +72,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Image to Model (Hunyuan3d 2.1)",
+        "name": "Image to Model (Hunyuan3d 2.1)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Image to Video (Wan 2.2).json

@@ -206,7 +206,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Image to Video (Wan 2.2)",
+        "name": "Image to Video (Wan 2.2)",
         "inputNode": {
           "id": -10,
           "bounding": [
@@ -825,7 +825,7 @@
               }
             },
             "widgets_values": [
-              "\u8272\u8c03\u8273\u4e3d\uff0c\u8fc7\u66dd\uff0c\u9759\u6001\uff0c\u7ec6\u8282\u6a21\u7cca\u4e0d\u6e05\uff0c\u5b57\u5e55\uff0c\u98ce\u683c\uff0c\u4f5c\u54c1\uff0c\u753b\u4f5c\uff0c\u753b\u9762\uff0c\u9759\u6b62\uff0c\u6574\u4f53\u53d1\u7070\uff0c\u6700\u5dee\u8d28\u91cf\uff0c\u4f4e\u8d28\u91cf\uff0cJPEG\u538b\u7f29\u6b8b\u7559\uff0c\u4e11\u964b\u7684\uff0c\u6b8b\u7f3a\u7684\uff0c\u591a\u4f59\u7684\u624b\u6307\uff0c\u753b\u5f97\u4e0d\u597d\u7684\u624b\u90e8\uff0c\u753b\u5f97\u4e0d\u597d\u7684\u8138\u90e8\uff0c\u7578\u5f62\u7684\uff0c\u6bc1\u5bb9\u7684\uff0c\u5f62\u6001\u7578\u5f62\u7684\u80a2\u4f53\uff0c\u624b\u6307\u878d\u5408\uff0c\u9759\u6b62\u4e0d\u52a8\u7684\u753b\u9762\uff0c\u6742\u4e71\u7684\u80cc\u666f\uff0c\u4e09\u6761\u817f\uff0c\u80cc\u666f\u4eba\u5f88\u591a\uff0c\u5012\u7740\u8d70"
+              "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走"
             ],
             "color": "#322",
             "bgcolor": "#533"
@@ -1525,7 +1525,7 @@
               }
             },
             "widgets_values": [
-              "## GPU:RTX4090D 24GB\n\n| Model            | Size |VRAM Usage | 1st Generation | 2nd Generation |\n|---------------------|-------|-----------|---------------|-----------------|\n| fp8_scaled               |640*640| 84%               | \u2248  536s              | \u2248 513s                   |\n| fp8_scaled +  4steps LoRA  | 640*640  | 83%                | \u2248 97s               | \u2248 71s                   |"
+              "## GPU:RTX4090D 24GB\n\n| Model            | Size |VRAM Usage | 1st Generation | 2nd Generation |\n|---------------------|-------|-----------|---------------|-----------------|\n| fp8_scaled               |640*640| 84%               | ≈  536s              | ≈ 513s                   |\n| fp8_scaled +  4steps LoRA  | 640*640  | 83%                | ≈ 97s               | ≈ 71s                   |"
             ],
             "color": "#222",
             "bgcolor": "#000"
@@ -1555,7 +1555,7 @@
               }
             },
             "widgets_values": [
-              "[Tutorial](https://docs.comfy.org/tutorials/video/wan/wan2_2\n)\n\n**Diffusion Model**\n- [wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors)\n- [wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors)\n\n**LoRA**\n- [wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors)\n- [wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise.safetensors)\n\n**VAE**\n- [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors)\n\n**Text Encoder**   \n- [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors)\n\n\nFile save location\n\n```\nComfyUI/\n\u251c\u2500\u2500\u2500\ud83d\udcc2 models/\n\u2502   \u251c\u2500\u2500\u2500\ud83d\udcc2 diffusion_models/\n\u2502   \u2502   \u251c\u2500\u2500\u2500 wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors\n\u2502   \u2502   \u2514\u2500\u2500\u2500 wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors\n\u2502   \u251c\u2500\u2500\u2500\ud83d\udcc2 loras/\n\u2502   \u2502   \u251c\u2500\u2500\u2500 wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors\n\u2502   \u2502   \u2514\u2500\u2500\u2500 wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors\n\u2502   \u251c\u2500\u2500\u2500\ud83d\udcc2 text_encoders/\n\u2502   \u2502   \u2514\u2500\u2500\u2500 umt5_xxl_fp8_e4m3fn_scaled.safetensors \n\u2502   \u2514\u2500\u2500\u2500\ud83d\udcc2 vae/\n\u2502       \u2514\u2500\u2500 wan_2.1_vae.safetensors\n```\n"
+              "[Tutorial](https://docs.comfy.org/tutorials/video/wan/wan2_2\n)\n\n**Diffusion Model**\n- [wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors)\n- [wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors)\n\n**LoRA**\n- [wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors)\n- [wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_i2v_lightx2v_4steps_lora_v1_high_noise.safetensors)\n\n**VAE**\n- [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors)\n\n**Text Encoder**   \n- [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors)\n\n\nFile save location\n\n```\nComfyUI/\n├───📂 models/\n│   ├───📂 diffusion_models/\n│   │   ├─── wan2.2_i2v_low_noise_14B_fp8_scaled.safetensors\n│   │   └─── wan2.2_i2v_high_noise_14B_fp8_scaled.safetensors\n│   ├───📂 loras/\n│   │   ├─── wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors\n│   │   └─── wan2.2_i2v_lightx2v_4steps_lora_v1_low_noise.safetensors\n│   ├───📂 text_encoders/\n│   │   └─── umt5_xxl_fp8_e4m3fn_scaled.safetensors \n│   └───📂 vae/\n│       └── wan_2.1_vae.safetensors\n```\n"
             ],
             "color": "#222",
             "bgcolor": "#000"

blueprints/Pose to Image (Z-Image-Turbo).json

@@ -134,7 +134,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Pose to Image (Z-Image-Turbo)",
+        "name": "Pose to Image (Z-Image-Turbo)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Sharpen.json

@@ -267,7 +267,7 @@
               "Node name for S&R": "GLSLShader"
             },
             "widgets_values": [
-              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0;  // strength [0.0 \u2013 2.0] typical: 0.3\u20131.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n    vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));\n    \n    // Sample center and neighbors\n    vec4 center = texture(u_image0, v_texCoord);\n    vec4 top    = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n    vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0,  texel.y));\n    vec4 left   = texture(u_image0, v_texCoord + vec2(-texel.x,  0.0));\n    vec4 right  = texture(u_image0, v_texCoord + vec2( texel.x,  0.0));\n    \n    // Edge enhancement (Laplacian)\n    vec4 edges = center * 4.0 - top - bottom - left - right;\n    \n    // Add edges back scaled by strength\n    vec4 sharpened = center + edges * u_float0;\n    \n    fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}",
+              "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0;  // strength [0.0 – 2.0] typical: 0.3–1.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n    vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));\n    \n    // Sample center and neighbors\n    vec4 center = texture(u_image0, v_texCoord);\n    vec4 top    = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n    vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0,  texel.y));\n    vec4 left   = texture(u_image0, v_texCoord + vec2(-texel.x,  0.0));\n    vec4 right  = texture(u_image0, v_texCoord + vec2( texel.x,  0.0));\n    \n    // Edge enhancement (Laplacian)\n    vec4 edges = center * 4.0 - top - bottom - left - right;\n    \n    // Add edges back scaled by strength\n    vec4 sharpened = center + edges * u_float0;\n    \n    fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}",
               "from_input"
             ]
           }

blueprints/Text to Audio (ACE-Step 1.5).json

@@ -222,7 +222,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Text to Audio (ACE-Step 1.5)",
+        "name": "Text to Audio (ACE-Step 1.5)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Text to Image (Flux.1 Krea Dev).json

@@ -1024,7 +1024,7 @@
           "workflowRendererVersion": "LG"
         },
         "category": "Image generation and editing/Text to image",
-        "description": "Generates images from text prompts using Flux.1 Krea Dev, a BFL \u00d7 Krea collaboration variant."
+        "description": "Generates images from text prompts using Flux.1 Krea Dev, a BFL × Krea collaboration variant."
       }
     ]
   },

blueprints/Text to Image (Qwen-Image 2512).json

@@ -751,7 +751,7 @@
               "secondTabWidth": 65
             },
             "widgets_values": [
-              "\u4f4e\u5206\u8fa8\u7387\uff0c\u4f4e\u753b\u8d28\uff0c\u80a2\u4f53\u7578\u5f62\uff0c\u624b\u6307\u7578\u5f62\uff0c\u753b\u9762\u8fc7\u9971\u548c\uff0c\u8721\u50cf\u611f\uff0c\u4eba\u8138\u65e0\u7ec6\u8282\uff0c\u8fc7\u5ea6\u5149\u6ed1\uff0c\u753b\u9762\u5177\u6709AI\u611f\u3002\u6784\u56fe\u6df7\u4e71\u3002\u6587\u5b57\u6a21\u7cca\uff0c\u626d\u66f2"
+              "低分辨率，低画质，肢体畸形，手指畸形，画面过饱和，蜡像感，人脸无细节，过度光滑，画面具有AI感。构图混乱。文字模糊，扭曲"
             ],
             "color": "#322",
             "bgcolor": "#533"

blueprints/Text to Image (Z-Image-Turbo).json

@@ -149,7 +149,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Text to Image (Z-Image-Turbo)",
+        "name": "Text to Image (Z-Image-Turbo)",
         "inputNode": {
           "id": -10,
           "bounding": [

blueprints/Text to Video (Wan 2.2).json

@@ -1160,7 +1160,7 @@
             "title": "Model Links",
             "properties": {},
             "widgets_values": [
-              "[Tutorial](https://docs.comfy.org/tutorials/video/wan/wan2_2\n) | [\u6559\u7a0b](https://docs.comfy.org/zh-CN/tutorials/video/wan/wan2_2\n)\n\n**Diffusion Model**       \n- [wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors)\n- [wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors)\n\n**LoRA**\n\n- [wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors)\n- [wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors)\n\n**VAE**\n- [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors)\n\n**Text Encoder**   \n- [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors)\n\n\nFile save location\n\n```\nComfyUI/\n\u251c\u2500\u2500\u2500\ud83d\udcc2 models/\n\u2502   \u251c\u2500\u2500\u2500\ud83d\udcc2 diffusion_models/\n\u2502   \u2502   \u251c\u2500\u2500\u2500 wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors\n\u2502   \u2502   \u2514\u2500\u2500\u2500 wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors\n\u2502   \u251c\u2500\u2500\u2500\ud83d\udcc2 loras/\n\u2502   \u2502   \u251c\u2500\u2500\u2500wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors\n\u2502   \u2502   \u2514\u2500\u2500\u2500wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors\n\u2502   \u251c\u2500\u2500\u2500\ud83d\udcc2 text_encoders/\n\u2502   \u2502   \u2514\u2500\u2500\u2500 umt5_xxl_fp8_e4m3fn_scaled.safetensors \n\u2502   \u2514\u2500\u2500\u2500\ud83d\udcc2 vae/\n\u2502       \u2514\u2500\u2500 wan_2.1_vae.safetensors\n```\n"
+              "[Tutorial](https://docs.comfy.org/tutorials/video/wan/wan2_2\n) | [教程](https://docs.comfy.org/zh-CN/tutorials/video/wan/wan2_2\n)\n\n**Diffusion Model**       \n- [wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors)\n- [wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors)\n\n**LoRA**\n\n- [wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors)\n- [wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors)\n\n**VAE**\n- [wan_2.1_vae.safetensors](https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors)\n\n**Text Encoder**   \n- [umt5_xxl_fp8_e4m3fn_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors)\n\n\nFile save location\n\n```\nComfyUI/\n├───📂 models/\n│   ├───📂 diffusion_models/\n│   │   ├─── wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors\n│   │   └─── wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors\n│   ├───📂 loras/\n│   │   ├───wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors\n│   │   └───wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors\n│   ├───📂 text_encoders/\n│   │   └─── umt5_xxl_fp8_e4m3fn_scaled.safetensors \n│   └───📂 vae/\n│       └── wan_2.1_vae.safetensors\n```\n"
             ],
             "color": "#222",
             "bgcolor": "#000"
@@ -1263,7 +1263,7 @@
               "Node name for S&R": "CLIPTextEncode"
             },
             "widgets_values": [
-              "\u8272\u8c03\u8273\u4e3d\uff0c\u8fc7\u66dd\uff0c\u9759\u6001\uff0c\u7ec6\u8282\u6a21\u7cca\u4e0d\u6e05\uff0c\u5b57\u5e55\uff0c\u98ce\u683c\uff0c\u4f5c\u54c1\uff0c\u753b\u4f5c\uff0c\u753b\u9762\uff0c\u9759\u6b62\uff0c\u6574\u4f53\u53d1\u7070\uff0c\u6700\u5dee\u8d28\u91cf\uff0c\u4f4e\u8d28\u91cf\uff0cJPEG\u538b\u7f29\u6b8b\u7559\uff0c\u4e11\u964b\u7684\uff0c\u6b8b\u7f3a\u7684\uff0c\u591a\u4f59\u7684\u624b\u6307\uff0c\u753b\u5f97\u4e0d\u597d\u7684\u624b\u90e8\uff0c\u753b\u5f97\u4e0d\u597d\u7684\u8138\u90e8\uff0c\u7578\u5f62\u7684\uff0c\u6bc1\u5bb9\u7684\uff0c\u5f62\u6001\u7578\u5f62\u7684\u80a2\u4f53\uff0c\u624b\u6307\u878d\u5408\uff0c\u9759\u6b62\u4e0d\u52a8\u7684\u753b\u9762\uff0c\u6742\u4e71\u7684\u80cc\u666f\uff0c\u4e09\u6761\u817f\uff0c\u80cc\u666f\u4eba\u5f88\u591a\uff0c\u5012\u7740\u8d70\uff0c\u88f8\u9732\uff0cNSFW"
+              "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走，裸露，NSFW"
             ],
             "color": "#322",
             "bgcolor": "#533"

blueprints/Video Inpaint(Wan2.1 VACE).json

@@ -165,7 +165,7 @@
         },
         "revision": 0,
         "config": {},
-        "name": "local-Video Inpaint(Wan2.1 VACE)",
+        "name": "Video Inpaint(Wan2.1 VACE)",
         "inputNode": {
           "id": -10,
           "bounding": [
@@ -1116,7 +1116,7 @@
             "title": "About Video Size",
             "properties": {},
             "widgets_values": [
-              "| Model                                                         | 480P | 720P |\n| ------------------------------------------------------------ | ---- | ---- |\n| [VACE-1.3B](https://huggingface.co/Wan-AI/Wan2.1-VACE-1.3B) | \u2705   | \u274c   |\n| [VACE-14B](https://huggingface.co/Wan-AI/Wan2.1-VACE-14B)   | \u2705   | \u2705   |"
+              "| Model                                                         | 480P | 720P |\n| ------------------------------------------------------------ | ---- | ---- |\n| [VACE-1.3B](https://huggingface.co/Wan-AI/Wan2.1-VACE-1.3B) | ✅   | ❌   |\n| [VACE-14B](https://huggingface.co/Wan-AI/Wan2.1-VACE-14B)   | ✅   | ✅   |"
             ],
             "color": "#432",
             "bgcolor": "#000"
@@ -1516,7 +1516,7 @@
               "widget_ue_connectable": {}
             },
             "widgets_values": [
-              "Currently, it's difficult to perfectly draw dynamic masks for different frames using only core nodes. However, to avoid requiring users to install additional custom nodes, our templates only use core nodes. You can refer to this implementation idea to achieve video inpainting.\n\nYou can use KJNode\u2019s Points Editor and Sam2Segmentation to create some dynamic mask functions.\n\nCustom node links:\n- [ComfyUI-KJNodes](https://github.com/kijai/ComfyUI-KJNodes)\n- [ComfyUI-segment-anything-2](https://github.com/kijai/ComfyUI-segment-anything-2)"
+              "Currently, it's difficult to perfectly draw dynamic masks for different frames using only core nodes. However, to avoid requiring users to install additional custom nodes, our templates only use core nodes. You can refer to this implementation idea to achieve video inpainting.\n\nYou can use KJNode’s Points Editor and Sam2Segmentation to create some dynamic mask functions.\n\nCustom node links:\n- [ComfyUI-KJNodes](https://github.com/kijai/ComfyUI-KJNodes)\n- [ComfyUI-segment-anything-2](https://github.com/kijai/ComfyUI-segment-anything-2)"
             ],
             "color": "#432",
             "bgcolor": "#000"
@@ -1578,7 +1578,7 @@
               "widget_ue_connectable": {}
             },
             "widgets_values": [
-              "\u8fc7\u66dd\uff0c\u9759\u6001\uff0c\u7ec6\u8282\u6a21\u7cca\u4e0d\u6e05\uff0c\u5b57\u5e55\uff0c\u98ce\u683c\uff0c\u4f5c\u54c1\uff0c\u753b\u4f5c\uff0c\u753b\u9762\uff0c\u9759\u6b62\uff0c\u6574\u4f53\u53d1\u7070\uff0c\u6700\u5dee\u8d28\u91cf\uff0c\u4f4e\u8d28\u91cf\uff0cJPEG\u538b\u7f29\u6b8b\u7559\uff0c\u4e11\u964b\u7684\uff0c\u6b8b\u7f3a\u7684\uff0c\u591a\u4f59\u7684\u624b\u6307\uff0c\u753b\u5f97\u4e0d\u597d\u7684\u624b\u90e8\uff0c\u753b\u5f97\u4e0d\u597d\u7684\u8138\u90e8\uff0c\u7578\u5f62\u7684\uff0c\u6bc1\u5bb9\u7684\uff0c\u5f62\u6001\u7578\u5f62\u7684\u80a2\u4f53\uff0c\u624b\u6307\u878d\u5408\uff0c\u9759\u6b62\u4e0d\u52a8\u7684\u753b\u9762\uff0c\u6742\u4e71\u7684\u80cc\u666f\uff0c\u4e09\u6761\u817f\uff0c\u80cc\u666f\u4eba\u5f88\u591a\uff0c\u5012\u7740\u8d70,\u8fc7\u66dd\uff0c"
+              "过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走,过曝，"
             ],
             "color": "#223",
             "bgcolor": "#335"

blueprints/Video Upscale(GAN x4).json

@@ -413,7 +413,7 @@
           "workflowRendererVersion": "LG"
         },
         "category": "Video generation and editing/Enhance video",
-        "description": "Upscales video to 4\u00d7 resolution using a GAN-based upscaling model."
+        "description": "Upscales video to 4× resolution using a GAN-based upscaling model."
       }
     ]
   },

comfy/background_removal/birefnet.json

@@ -0,0 +1,7 @@
+{
+    "model_type": "birefnet",
+    "image_std": [1.0, 1.0, 1.0],
+    "image_mean": [0.0, 0.0, 0.0],
+    "image_size": 1024,
+    "resize_to_original": true
+}

comfy/background_removal/birefnet.py

@@ -0,0 +1,689 @@
+import torch
+import comfy.ops
+import numpy as np
+import torch.nn as nn
+from functools import partial
+import torch.nn.functional as F
+from torchvision.ops import deform_conv2d
+from comfy.ldm.modules.attention import optimized_attention_for_device
+
+CXT = [3072, 1536, 768, 384][1:][::-1][-3:]
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, device=None, dtype=None, operations=None):
+        super().__init__()
+
+        self.dim = dim
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim ** -0.5
+
+        self.q = operations.Linear(dim, dim, bias=qkv_bias, device=device, dtype=dtype)
+        self.kv = operations.Linear(dim, dim * 2, bias=qkv_bias, device=device, dtype=dtype)
+        self.proj = operations.Linear(dim, dim, device=device, dtype=dtype)
+
+    def forward(self, x):
+        B, N, C = x.shape
+        optimized_attention = optimized_attention_for_device(x.device, mask=False, small_input=True)
+        q = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
+        kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        k, v = kv[0], kv[1]
+
+        x = optimized_attention(
+            q, k, v, heads=self.num_heads, skip_output_reshape=True, skip_reshape=True
+        ).transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+
+        return x
+
+class Mlp(nn.Module):
+    def __init__(self, in_features, hidden_features=None, out_features=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = operations.Linear(in_features, hidden_features, device=device, dtype=dtype)
+        self.act = nn.GELU()
+        self.fc2 = operations.Linear(hidden_features, out_features, device=device, dtype=dtype)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.fc2(x)
+        return x
+
+
+def window_partition(x, window_size):
+    B, H, W, C = x.shape
+    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows
+
+
+def window_reverse(windows, window_size, H, W):
+    B = int(windows.shape[0] / (H * W / window_size / window_size))
+    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
+    return x
+
+
+class WindowAttention(nn.Module):
+    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, device=None, dtype=None, operations=None):
+
+        super().__init__()
+        self.dim = dim
+        self.window_size = window_size  # Wh, Ww
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim ** -0.5
+
+        self.relative_position_bias_table = nn.Parameter(
+            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads, device=device, dtype=dtype))
+
+        coords_h = torch.arange(self.window_size[0])
+        coords_w = torch.arange(self.window_size[1])
+        coords = torch.stack(torch.meshgrid([coords_h, coords_w], indexing='ij'))  # 2, Wh, Ww
+        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+        relative_coords[:, :, 0] += self.window_size[0] - 1
+        relative_coords[:, :, 1] += self.window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
+        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+        self.register_buffer("relative_position_index", relative_position_index)
+
+        self.qkv = operations.Linear(dim, dim * 3, bias=qkv_bias, device=device, dtype=dtype)
+        self.proj = operations.Linear(dim, dim, device=device, dtype=dtype)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x, mask=None):
+        B_, N, C = x.shape
+        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]
+
+        q = q * self.scale
+        attn = (q @ k.transpose(-2, -1))
+
+        relative_position_bias = self.relative_position_bias_table[self.relative_position_index.long().view(-1)].view(
+            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)  # Wh*Ww,Wh*Ww,nH
+        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
+        attn = attn + relative_position_bias.unsqueeze(0)
+
+        if mask is not None:
+            nW = mask.shape[0]
+            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
+            attn = attn.view(-1, self.num_heads, N, N)
+            attn = self.softmax(attn)
+        else:
+            attn = self.softmax(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
+        x = self.proj(x)
+        return x
+
+
+class SwinTransformerBlock(nn.Module):
+    def __init__(self, dim, num_heads, window_size=7, shift_size=0,
+                 mlp_ratio=4., qkv_bias=True, qk_scale=None,
+                 norm_layer=nn.LayerNorm, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.dim = dim
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.shift_size = shift_size
+        self.mlp_ratio = mlp_ratio
+
+        self.norm1 = norm_layer(dim, device=device, dtype=dtype)
+        self.attn = WindowAttention(
+            dim, window_size=(self.window_size, self.window_size), num_heads=num_heads,
+            qkv_bias=qkv_bias, qk_scale=qk_scale, device=device, dtype=dtype, operations=operations)
+
+        self.norm2 = norm_layer(dim, device=device, dtype=dtype)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, device=device, dtype=dtype, operations=operations)
+
+        self.H = None
+        self.W = None
+
+    def forward(self, x, mask_matrix):
+        B, L, C = x.shape
+        H, W = self.H, self.W
+
+        shortcut = x
+        x = self.norm1(x)
+        x = x.view(B, H, W, C)
+
+        pad_l = pad_t = 0
+        pad_r = (self.window_size - W % self.window_size) % self.window_size
+        pad_b = (self.window_size - H % self.window_size) % self.window_size
+        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
+        _, Hp, Wp, _ = x.shape
+
+        if self.shift_size > 0:
+            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
+            attn_mask = mask_matrix
+        else:
+            shifted_x = x
+            attn_mask = None
+
+        x_windows = window_partition(shifted_x, self.window_size)
+        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)
+
+        attn_windows = self.attn(x_windows, mask=attn_mask)
+
+        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
+        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # B H' W' C
+
+        if self.shift_size > 0:
+            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
+        else:
+            x = shifted_x
+
+        if pad_r > 0 or pad_b > 0:
+            x = x[:, :H, :W, :].contiguous()
+
+        x = x.view(B, H * W, C)
+
+        x = shortcut + x
+        x = x + self.mlp(self.norm2(x))
+
+        return x
+
+
+class PatchMerging(nn.Module):
+    def __init__(self, dim, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.dim = dim
+        self.reduction = operations.Linear(4 * dim, 2 * dim, bias=False, device=device, dtype=dtype)
+        self.norm = operations.LayerNorm(4 * dim, device=device, dtype=dtype)
+
+    def forward(self, x, H, W):
+        B, L, C = x.shape
+        x = x.view(B, H, W, C)
+
+        # padding
+        pad_input = (H % 2 == 1) or (W % 2 == 1)
+        if pad_input:
+            x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))
+
+        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
+        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
+        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
+        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
+        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
+        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
+
+        x = self.norm(x)
+        x = self.reduction(x)
+
+        return x
+
+
+class BasicLayer(nn.Module):
+    def __init__(self,
+                 dim,
+                 depth,
+                 num_heads,
+                 window_size=7,
+                 mlp_ratio=4.,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 norm_layer=nn.LayerNorm,
+                 downsample=None,
+                 device=None, dtype=None, operations=None):
+        super().__init__()
+        self.window_size = window_size
+        self.shift_size = window_size // 2
+        self.depth = depth
+
+        # build blocks
+        self.blocks = nn.ModuleList([
+            SwinTransformerBlock(
+                dim=dim,
+                num_heads=num_heads,
+                window_size=window_size,
+                shift_size=0 if (i % 2 == 0) else window_size // 2,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                norm_layer=norm_layer,
+                device=device, dtype=dtype, operations=operations)
+            for i in range(depth)])
+
+        # patch merging layer
+        if downsample is not None:
+            self.downsample = downsample(dim=dim, device=device, dtype=dtype, operations=operations)
+        else:
+            self.downsample = None
+
+    def forward(self, x, H, W):
+        Hp = int(np.ceil(H / self.window_size)) * self.window_size
+        Wp = int(np.ceil(W / self.window_size)) * self.window_size
+        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
+        h_slices = (slice(0, -self.window_size),
+                    slice(-self.window_size, -self.shift_size),
+                    slice(-self.shift_size, None))
+        w_slices = (slice(0, -self.window_size),
+                    slice(-self.window_size, -self.shift_size),
+                    slice(-self.shift_size, None))
+        cnt = 0
+        for h in h_slices:
+            for w in w_slices:
+                img_mask[:, h, w, :] = cnt
+                cnt += 1
+
+        mask_windows = window_partition(img_mask, self.window_size)
+        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
+        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
+
+        for blk in self.blocks:
+            blk.H, blk.W = H, W
+            x = blk(x, attn_mask)
+        if self.downsample is not None:
+            x_down = self.downsample(x, H, W)
+            Wh, Ww = (H + 1) // 2, (W + 1) // 2
+            return x, H, W, x_down, Wh, Ww
+        else:
+            return x, H, W, x, H, W
+
+
+class PatchEmbed(nn.Module):
+    def __init__(self, patch_size=4, in_channels=3, embed_dim=96, norm_layer=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        patch_size = (patch_size, patch_size)
+        self.patch_size = patch_size
+
+        self.in_channels = in_channels
+        self.embed_dim = embed_dim
+
+        self.proj = operations.Conv2d(in_channels, embed_dim, kernel_size=patch_size, stride=patch_size, device=device, dtype=dtype)
+        if norm_layer is not None:
+            self.norm = norm_layer(embed_dim, device=device, dtype=dtype)
+        else:
+            self.norm = None
+
+    def forward(self, x):
+        _, _, H, W = x.size()
+        if W % self.patch_size[1] != 0:
+            x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1]))
+        if H % self.patch_size[0] != 0:
+            x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0]))
+
+        x = self.proj(x)  # B C Wh Ww
+        if self.norm is not None:
+            Wh, Ww = x.size(2), x.size(3)
+            x = x.flatten(2).transpose(1, 2)
+            x = self.norm(x)
+            x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)
+
+        return x
+
+
+class SwinTransformer(nn.Module):
+    def __init__(self,
+                 pretrain_img_size=224,
+                 patch_size=4,
+                 in_channels=3,
+                 embed_dim=96,
+                 depths=[2, 2, 6, 2],
+                 num_heads=[3, 6, 12, 24],
+                 window_size=7,
+                 mlp_ratio=4.,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 patch_norm=True,
+                 out_indices=(0, 1, 2, 3),
+                 frozen_stages=-1,
+                 device=None, dtype=None, operations=None):
+        super().__init__()
+
+        norm_layer = partial(operations.LayerNorm, device=device, dtype=dtype)
+        self.pretrain_img_size = pretrain_img_size
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.patch_norm = patch_norm
+        self.out_indices = out_indices
+        self.frozen_stages = frozen_stages
+
+        self.patch_embed = PatchEmbed(
+            patch_size=patch_size, in_channels=in_channels, embed_dim=embed_dim,
+            device=device, dtype=dtype, operations=operations,
+            norm_layer=norm_layer if self.patch_norm else None)
+
+        self.layers = nn.ModuleList()
+        for i_layer in range(self.num_layers):
+            layer = BasicLayer(
+                dim=int(embed_dim * 2 ** i_layer),
+                depth=depths[i_layer],
+                num_heads=num_heads[i_layer],
+                window_size=window_size,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                norm_layer=norm_layer,
+                downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
+                device=device, dtype=dtype, operations=operations)
+            self.layers.append(layer)
+
+        num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)]
+        self.num_features = num_features
+
+        for i_layer in out_indices:
+            layer = norm_layer(num_features[i_layer])
+            layer_name = f'norm{i_layer}'
+            self.add_module(layer_name, layer)
+
+
+    def forward(self, x):
+        x = self.patch_embed(x)
+
+        Wh, Ww = x.size(2), x.size(3)
+
+        outs = []
+        x = x.flatten(2).transpose(1, 2)
+        for i in range(self.num_layers):
+            layer = self.layers[i]
+            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)
+
+            if i in self.out_indices:
+                norm_layer = getattr(self, f'norm{i}')
+                x_out = norm_layer(x_out)
+
+                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
+                outs.append(out)
+
+        return tuple(outs)
+
+class DeformableConv2d(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size=3,
+                 stride=1,
+                 padding=1,
+                 bias=False, device=None, dtype=None, operations=None):
+
+        super(DeformableConv2d, self).__init__()
+
+        kernel_size = kernel_size if type(kernel_size) is tuple else (kernel_size, kernel_size)
+        self.stride = stride if type(stride) is tuple else (stride, stride)
+        self.padding = padding
+
+        self.offset_conv = operations.Conv2d(in_channels,
+                                     2 * kernel_size[0] * kernel_size[1],
+                                     kernel_size=kernel_size,
+                                     stride=stride,
+                                     padding=self.padding,
+                                     bias=True, device=device, dtype=dtype)
+
+        self.modulator_conv = operations.Conv2d(in_channels,
+                                     1 * kernel_size[0] * kernel_size[1],
+                                     kernel_size=kernel_size,
+                                     stride=stride,
+                                     padding=self.padding,
+                                     bias=True, device=device, dtype=dtype)
+
+        self.regular_conv = operations.Conv2d(in_channels,
+                                      out_channels=out_channels,
+                                      kernel_size=kernel_size,
+                                      stride=stride,
+                                      padding=self.padding,
+                                      bias=bias, device=device, dtype=dtype)
+
+    def forward(self, x):
+        offset = self.offset_conv(x)
+        modulator = 2. * torch.sigmoid(self.modulator_conv(x))
+        weight, bias, offload_info = comfy.ops.cast_bias_weight(self.regular_conv, x, offloadable=True)
+
+        x = deform_conv2d(
+            input=x,
+            offset=offset,
+            weight=weight,
+            bias=None,
+            padding=self.padding,
+            mask=modulator,
+            stride=self.stride,
+        )
+        comfy.ops.uncast_bias_weight(self.regular_conv, weight, bias, offload_info)
+        return x
+
+class BasicDecBlk(nn.Module):
+    def __init__(self, in_channels=64, out_channels=64, inter_channels=64, device=None, dtype=None, operations=None):
+        super(BasicDecBlk, self).__init__()
+        inter_channels = 64
+        self.conv_in = operations.Conv2d(in_channels, inter_channels, 3, 1, padding=1, device=device, dtype=dtype)
+        self.relu_in = nn.ReLU(inplace=True)
+        self.dec_att = ASPPDeformable(in_channels=inter_channels, device=device, dtype=dtype, operations=operations)
+        self.conv_out = operations.Conv2d(inter_channels, out_channels, 3, 1, padding=1, device=device, dtype=dtype)
+        self.bn_in = operations.BatchNorm2d(inter_channels, device=device, dtype=dtype)
+        self.bn_out = operations.BatchNorm2d(out_channels, device=device, dtype=dtype)
+
+    def forward(self, x):
+        x = self.conv_in(x)
+        x = self.bn_in(x)
+        x = self.relu_in(x)
+        x = self.dec_att(x)
+        x = self.conv_out(x)
+        x = self.bn_out(x)
+        return x
+
+
+class BasicLatBlk(nn.Module):
+    def __init__(self, in_channels=64, out_channels=64, device=None, dtype=None, operations=None):
+        super(BasicLatBlk, self).__init__()
+        self.conv = operations.Conv2d(in_channels, out_channels, 1, 1, 0, device=device, dtype=dtype)
+
+    def forward(self, x):
+        x = self.conv(x)
+        return x
+
+
+class _ASPPModuleDeformable(nn.Module):
+    def __init__(self, in_channels, planes, kernel_size, padding, device, dtype, operations):
+        super(_ASPPModuleDeformable, self).__init__()
+        self.atrous_conv = DeformableConv2d(in_channels, planes, kernel_size=kernel_size,
+                                            stride=1, padding=padding, bias=False, device=device, dtype=dtype, operations=operations)
+        self.bn = operations.BatchNorm2d(planes, device=device, dtype=dtype)
+        self.relu = nn.ReLU(inplace=True)
+
+    def forward(self, x):
+        x = self.atrous_conv(x)
+        x = self.bn(x)
+
+        return self.relu(x)
+
+
+class ASPPDeformable(nn.Module):
+    def __init__(self, in_channels, out_channels=None, parallel_block_sizes=[1, 3, 7], device=None, dtype=None, operations=None):
+        super(ASPPDeformable, self).__init__()
+        self.down_scale = 1
+        if out_channels is None:
+            out_channels = in_channels
+        self.in_channelster = 256 // self.down_scale
+
+        self.aspp1 = _ASPPModuleDeformable(in_channels, self.in_channelster, 1, padding=0, device=device, dtype=dtype, operations=operations)
+        self.aspp_deforms = nn.ModuleList([
+            _ASPPModuleDeformable(in_channels, self.in_channelster, conv_size, padding=int(conv_size//2), device=device, dtype=dtype, operations=operations)
+              for conv_size in parallel_block_sizes
+        ])
+
+        self.global_avg_pool = nn.Sequential(nn.AdaptiveAvgPool2d((1, 1)),
+                                             operations.Conv2d(in_channels, self.in_channelster, 1, stride=1, bias=False, device=device, dtype=dtype),
+                                             operations.BatchNorm2d(self.in_channelster, device=device, dtype=dtype),
+                                             nn.ReLU(inplace=True))
+        self.conv1 = operations.Conv2d(self.in_channelster * (2 + len(self.aspp_deforms)), out_channels, 1, bias=False, device=device, dtype=dtype)
+        self.bn1 = operations.BatchNorm2d(out_channels, device=device, dtype=dtype)
+        self.relu = nn.ReLU(inplace=True)
+
+    def forward(self, x):
+        x1 = self.aspp1(x)
+        x_aspp_deforms = [aspp_deform(x) for aspp_deform in self.aspp_deforms]
+        x5 = self.global_avg_pool(x)
+        x5 = F.interpolate(x5, size=x1.size()[2:], mode='bilinear', align_corners=True)
+        x = torch.cat((x1, *x_aspp_deforms, x5), dim=1)
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+
+        return x
+
+class BiRefNet(nn.Module):
+    def __init__(self, config=None, dtype=None, device=None, operations=None):
+        super(BiRefNet, self).__init__()
+        self.bb = SwinTransformer(embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=12, device=device, dtype=dtype, operations=operations)
+
+        channels = [1536, 768, 384, 192]
+        channels = [c * 2 for c in channels]
+        self.cxt = channels[1:][::-1][-3:]
+        self.squeeze_module = nn.Sequential(*[
+            BasicDecBlk(channels[0]+sum(self.cxt), channels[0], device=device, dtype=dtype, operations=operations)
+            for _ in range(1)
+        ])
+
+        self.decoder = Decoder(channels, device=device, dtype=dtype, operations=operations)
+
+    def forward_enc(self, x):
+        x1, x2, x3, x4 = self.bb(x)
+        B, C, H, W = x.shape
+        x1_, x2_, x3_, x4_ = self.bb(F.interpolate(x, size=(H//2, W//2), mode='bilinear', align_corners=True))
+        x1 = torch.cat([x1, F.interpolate(x1_, size=x1.shape[2:], mode='bilinear', align_corners=True)], dim=1)
+        x2 = torch.cat([x2, F.interpolate(x2_, size=x2.shape[2:], mode='bilinear', align_corners=True)], dim=1)
+        x3 = torch.cat([x3, F.interpolate(x3_, size=x3.shape[2:], mode='bilinear', align_corners=True)], dim=1)
+        x4 = torch.cat([x4, F.interpolate(x4_, size=x4.shape[2:], mode='bilinear', align_corners=True)], dim=1)
+        x4 = torch.cat(
+            (
+                *[
+                    F.interpolate(x1, size=x4.shape[2:], mode='bilinear', align_corners=True),
+                    F.interpolate(x2, size=x4.shape[2:], mode='bilinear', align_corners=True),
+                    F.interpolate(x3, size=x4.shape[2:], mode='bilinear', align_corners=True),
+                ][-len(CXT):],
+                x4
+            ),
+            dim=1
+        )
+        return (x1, x2, x3, x4)
+
+    def forward_ori(self, x):
+        (x1, x2, x3, x4) = self.forward_enc(x)
+        x4 = self.squeeze_module(x4)
+        features = [x, x1, x2, x3, x4]
+        scaled_preds = self.decoder(features)
+        return scaled_preds
+
+    def forward(self, pixel_values, intermediate_output=None):
+        scaled_preds = self.forward_ori(pixel_values)
+        return scaled_preds
+
+
+class Decoder(nn.Module):
+    def __init__(self, channels, device, dtype, operations):
+        super(Decoder, self).__init__()
+        # factory kwargs
+        fk = {"device":device, "dtype":dtype, "operations":operations}
+        DecoderBlock = partial(BasicDecBlk, **fk)
+        LateralBlock = partial(BasicLatBlk, **fk)
+        DBlock = partial(SimpleConvs, **fk)
+
+        self.split = True
+        N_dec_ipt = 64
+        ic = 64
+        ipt_cha_opt = 1
+        self.ipt_blk5 = DBlock(2**10*3 if self.split else 3, [N_dec_ipt, channels[0]//8][ipt_cha_opt], inter_channels=ic)
+        self.ipt_blk4 = DBlock(2**8*3 if self.split else 3, [N_dec_ipt, channels[0]//8][ipt_cha_opt], inter_channels=ic)
+        self.ipt_blk3 = DBlock(2**6*3 if self.split else 3, [N_dec_ipt, channels[1]//8][ipt_cha_opt], inter_channels=ic)
+        self.ipt_blk2 = DBlock(2**4*3 if self.split else 3, [N_dec_ipt, channels[2]//8][ipt_cha_opt], inter_channels=ic)
+        self.ipt_blk1 = DBlock(2**0*3 if self.split else 3, [N_dec_ipt, channels[3]//8][ipt_cha_opt], inter_channels=ic)
+
+        self.decoder_block4 = DecoderBlock(channels[0]+([N_dec_ipt, channels[0]//8][ipt_cha_opt]), channels[1])
+        self.decoder_block3 = DecoderBlock(channels[1]+([N_dec_ipt, channels[0]//8][ipt_cha_opt]), channels[2])
+        self.decoder_block2 = DecoderBlock(channels[2]+([N_dec_ipt, channels[1]//8][ipt_cha_opt]), channels[3])
+        self.decoder_block1 = DecoderBlock(channels[3]+([N_dec_ipt, channels[2]//8][ipt_cha_opt]), channels[3]//2)
+
+        fk = {"device":device, "dtype":dtype}
+
+        self.conv_out1 = nn.Sequential(operations.Conv2d(channels[3]//2+([N_dec_ipt, channels[3]//8][ipt_cha_opt]), 1, 1, 1, 0, **fk))
+
+        self.lateral_block4 = LateralBlock(channels[1], channels[1])
+        self.lateral_block3 = LateralBlock(channels[2], channels[2])
+        self.lateral_block2 = LateralBlock(channels[3], channels[3])
+
+        self.conv_ms_spvn_4 = operations.Conv2d(channels[1], 1, 1, 1, 0, **fk)
+        self.conv_ms_spvn_3 = operations.Conv2d(channels[2], 1, 1, 1, 0, **fk)
+        self.conv_ms_spvn_2 = operations.Conv2d(channels[3], 1, 1, 1, 0, **fk)
+
+        _N = 16
+
+        self.gdt_convs_4 = nn.Sequential(operations.Conv2d(channels[0] // 2, _N, 3, 1, 1, **fk), operations.BatchNorm2d(_N, **fk), nn.ReLU(inplace=True))
+        self.gdt_convs_3 = nn.Sequential(operations.Conv2d(channels[1] // 2, _N, 3, 1, 1, **fk), operations.BatchNorm2d(_N, **fk), nn.ReLU(inplace=True))
+        self.gdt_convs_2 = nn.Sequential(operations.Conv2d(channels[2] // 2, _N, 3, 1, 1, **fk), operations.BatchNorm2d(_N, **fk), nn.ReLU(inplace=True))
+
+        [setattr(self, f"gdt_convs_pred_{i}", nn.Sequential(operations.Conv2d(_N, 1, 1, 1, 0, **fk))) for i in range(2, 5)]
+        [setattr(self, f"gdt_convs_attn_{i}", nn.Sequential(operations.Conv2d(_N, 1, 1, 1, 0, **fk))) for i in range(2, 5)]
+
+    def get_patches_batch(self, x, p):
+        _size_h, _size_w = p.shape[2:]
+        patches_batch = []
+        for idx in range(x.shape[0]):
+            columns_x = torch.split(x[idx], split_size_or_sections=_size_w, dim=-1)
+            patches_x = []
+            for column_x in columns_x:
+                patches_x += [p.unsqueeze(0) for p in torch.split(column_x, split_size_or_sections=_size_h, dim=-2)]
+            patch_sample = torch.cat(patches_x, dim=1)
+            patches_batch.append(patch_sample)
+        return torch.cat(patches_batch, dim=0)
+
+    def forward(self, features):
+        x, x1, x2, x3, x4 = features
+
+        patches_batch = self.get_patches_batch(x, x4) if self.split else x
+        x4 = torch.cat((x4, self.ipt_blk5(F.interpolate(patches_batch, size=x4.shape[2:], mode='bilinear', align_corners=True))), 1)
+        p4 = self.decoder_block4(x4)
+        p4_gdt = self.gdt_convs_4(p4)
+        gdt_attn_4 = self.gdt_convs_attn_4(p4_gdt).sigmoid()
+        p4 = p4 * gdt_attn_4
+        _p4 = F.interpolate(p4, size=x3.shape[2:], mode='bilinear', align_corners=True)
+        _p3 = _p4 + self.lateral_block4(x3)
+
+        patches_batch = self.get_patches_batch(x, _p3) if self.split else x
+        _p3 = torch.cat((_p3, self.ipt_blk4(F.interpolate(patches_batch, size=x3.shape[2:], mode='bilinear', align_corners=True))), 1)
+        p3 = self.decoder_block3(_p3)
+
+        p3_gdt = self.gdt_convs_3(p3)
+        gdt_attn_3 = self.gdt_convs_attn_3(p3_gdt).sigmoid()
+        p3 = p3 * gdt_attn_3
+        _p3 = F.interpolate(p3, size=x2.shape[2:], mode='bilinear', align_corners=True)
+        _p2 = _p3 + self.lateral_block3(x2)
+
+        patches_batch = self.get_patches_batch(x, _p2) if self.split else x
+        _p2 = torch.cat((_p2, self.ipt_blk3(F.interpolate(patches_batch, size=x2.shape[2:], mode='bilinear', align_corners=True))), 1)
+        p2 = self.decoder_block2(_p2)
+
+        p2_gdt = self.gdt_convs_2(p2)
+        gdt_attn_2 = self.gdt_convs_attn_2(p2_gdt).sigmoid()
+        p2 = p2 * gdt_attn_2
+
+        _p2 = F.interpolate(p2, size=x1.shape[2:], mode='bilinear', align_corners=True)
+        _p1 = _p2 + self.lateral_block2(x1)
+
+        patches_batch = self.get_patches_batch(x, _p1) if self.split else x
+        _p1 = torch.cat((_p1, self.ipt_blk2(F.interpolate(patches_batch, size=x1.shape[2:], mode='bilinear', align_corners=True))), 1)
+        _p1 = self.decoder_block1(_p1)
+        _p1 = F.interpolate(_p1, size=x.shape[2:], mode='bilinear', align_corners=True)
+
+        patches_batch = self.get_patches_batch(x, _p1) if self.split else x
+        _p1 = torch.cat((_p1, self.ipt_blk1(F.interpolate(patches_batch, size=x.shape[2:], mode='bilinear', align_corners=True))), 1)
+        p1_out = self.conv_out1(_p1)
+        return p1_out
+
+
+class SimpleConvs(nn.Module):
+    def __init__(
+        self, in_channels: int, out_channels: int, inter_channels=64, device=None, dtype=None, operations=None
+    ) -> None:
+        super().__init__()
+        self.conv1 = operations.Conv2d(in_channels, inter_channels, 3, 1, 1, device=device, dtype=dtype)
+        self.conv_out = operations.Conv2d(inter_channels, out_channels, 3, 1, 1, device=device, dtype=dtype)
+
+    def forward(self, x):
+        return self.conv_out(self.conv1(x))

comfy/bg_removal_model.py

@@ -0,0 +1,78 @@
+from .utils import load_torch_file
+import os
+import json
+import torch
+import logging
+
+import comfy.ops
+import comfy.model_patcher
+import comfy.model_management
+import comfy.clip_model
+import comfy.background_removal.birefnet
+
+BG_REMOVAL_MODELS = {
+    "birefnet": comfy.background_removal.birefnet.BiRefNet
+}
+
+class BackgroundRemovalModel():
+    def __init__(self, json_config):
+        with open(json_config) as f:
+            config = json.load(f)
+
+        self.image_size = config.get("image_size", 1024)
+        self.image_mean = config.get("image_mean", [0.0, 0.0, 0.0])
+        self.image_std = config.get("image_std", [1.0, 1.0, 1.0])
+        self.model_type = config.get("model_type", "birefnet")
+        self.config = config.copy()
+        model_class = BG_REMOVAL_MODELS.get(self.model_type)
+
+        self.load_device = comfy.model_management.text_encoder_device()
+        offload_device = comfy.model_management.text_encoder_offload_device()
+        self.dtype = comfy.model_management.text_encoder_dtype(self.load_device)
+        self.model = model_class(config, self.dtype, offload_device, comfy.ops.manual_cast)
+        self.model.eval()
+
+        self.patcher = comfy.model_patcher.CoreModelPatcher(self.model, load_device=self.load_device, offload_device=offload_device)
+
+    def load_sd(self, sd):
+        return self.model.load_state_dict(sd, strict=False, assign=self.patcher.is_dynamic())
+
+    def get_sd(self):
+        return self.model.state_dict()
+
+    def encode_image(self, image):
+        comfy.model_management.load_model_gpu(self.patcher)
+        H, W = image.shape[1], image.shape[2]
+        pixel_values = comfy.clip_model.clip_preprocess(image.to(self.load_device), size=self.image_size, mean=self.image_mean, std=self.image_std, crop=False)
+        out = self.model(pixel_values=pixel_values)
+        out = torch.nn.functional.interpolate(out, size=(H, W), mode="bicubic", antialias=False)
+
+        mask = out.sigmoid().to(device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype())
+        if mask.ndim == 3:
+            mask = mask.unsqueeze(0)
+        if mask.shape[1] != 1:
+            mask = mask.movedim(-1, 1)
+
+        return mask
+
+
+def load_background_removal_model(sd):
+    if "bb.layers.1.blocks.0.attn.relative_position_index" in sd:
+        json_config = os.path.join(os.path.join(os.path.dirname(os.path.realpath(__file__)), "background_removal"), "birefnet.json")
+    else:
+        return None
+
+    bg_model = BackgroundRemovalModel(json_config)
+    m, u = bg_model.load_sd(sd)
+    if len(m) > 0:
+        logging.warning("missing background removal: {}".format(m))
+    u = set(u)
+    keys = list(sd.keys())
+    for k in keys:
+        if k not in u:
+            sd.pop(k)
+    return bg_model
+
+def load(ckpt_path):
+    sd = load_torch_file(ckpt_path)
+    return load_background_removal_model(sd)

comfy/hooks.py

@@ -93,7 +93,7 @@ class Hook:
         self.hook_scope = hook_scope
         '''Scope of where this hook should apply in terms of the conds used in sampling run.'''
         self.custom_should_register = default_should_register
-        '''Can be overriden with a compatible function to decide if this hook should be registered without the need to override .should_register'''
+        '''Can be overridden with a compatible function to decide if this hook should be registered without the need to override .should_register'''
 
     @property
     def strength(self):

comfy/ldm/modules/diffusionmodules/util.py

@@ -140,7 +140,7 @@ def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
     alphas = alphacums[ddim_timesteps]
     alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())
 
-    # according the the formula provided in https://arxiv.org/abs/2010.02502
+    # according to the formula provided in https://arxiv.org/abs/2010.02502
     sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev))
     if verbose:
         logging.info(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}')

comfy/ops.py

@@ -562,6 +562,25 @@ class disable_weight_init:
             else:
                 return super().forward(*args, **kwargs)
 
+    class BatchNorm2d(torch.nn.BatchNorm2d, CastWeightBiasOp):
+        def reset_parameters(self):
+            return None
+
+        def forward_comfy_cast_weights(self, input):
+            weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True)
+            running_mean = self.running_mean.to(device=input.device, dtype=weight.dtype) if self.running_mean is not None else None
+            running_var = self.running_var.to(device=input.device, dtype=weight.dtype) if self.running_var is not None else None
+            x = torch.nn.functional.batch_norm(input, running_mean, running_var, weight, bias, self.training, self.momentum, self.eps)
+            uncast_bias_weight(self, weight, bias, offload_stream)
+            return x
+
+        def forward(self, *args, **kwargs):
+            run_every_op()
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
+                return self.forward_comfy_cast_weights(*args, **kwargs)
+            else:
+                return super().forward(*args, **kwargs)
+
     class LayerNorm(torch.nn.LayerNorm, CastWeightBiasOp):
         def reset_parameters(self):
             return None
@@ -749,6 +768,9 @@ class manual_cast(disable_weight_init):
     class Conv3d(disable_weight_init.Conv3d):
         comfy_cast_weights = True
 
+    class BatchNorm2d(disable_weight_init.BatchNorm2d):
+        comfy_cast_weights = True
+
     class GroupNorm(disable_weight_init.GroupNorm):
         comfy_cast_weights = True
 

comfy_api/latest/_io.py

@@ -17,6 +17,7 @@ if TYPE_CHECKING:
     from spandrel import ImageModelDescriptor
     from comfy.clip_vision import ClipVisionModel
     from comfy.clip_vision import Output as ClipVisionOutput_
+    from comfy.bg_removal_model import BackgroundRemovalModel
     from comfy.controlnet import ControlNet
     from comfy.hooks import HookGroup, HookKeyframeGroup
     from comfy.model_patcher import ModelPatcher
@@ -614,6 +615,11 @@ class Model(ComfyTypeIO):
     if TYPE_CHECKING:
         Type = ModelPatcher
 
+@comfytype(io_type="BACKGROUND_REMOVAL")
+class BackgroundRemoval(ComfyTypeIO):
+    if TYPE_CHECKING:
+        Type = BackgroundRemovalModel
+
 @comfytype(io_type="CLIP_VISION")
 class ClipVision(ComfyTypeIO):
     if TYPE_CHECKING:
@@ -2257,6 +2263,7 @@ __all__ = [
     "ModelPatch",
     "ClipVision",
     "ClipVisionOutput",
+    "BackgroundRemoval",
     "AudioEncoder",
     "AudioEncoderOutput",
     "StyleModel",

comfy_api_nodes/nodes_bytedance.py

@@ -1271,7 +1271,7 @@ PRICE_BADGE_VIDEO = IO.PriceBadge(
 )
 
 
-def _seedance2_text_inputs(resolutions: list[str]):
+def _seedance2_text_inputs(resolutions: list[str], default_ratio: str = "16:9"):
     return [
         IO.String.Input(
             "prompt",
@@ -1287,6 +1287,7 @@ def _seedance2_text_inputs(resolutions: list[str]):
         IO.Combo.Input(
             "ratio",
             options=["16:9", "4:3", "1:1", "3:4", "9:16", "21:9", "adaptive"],
+            default=default_ratio,
             tooltip="Aspect ratio of the output video.",
         ),
         IO.Int.Input(
@@ -1420,8 +1421,14 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "model",
                     options=[
-                        IO.DynamicCombo.Option("Seedance 2.0", _seedance2_text_inputs(["480p", "720p", "1080p"])),
-                        IO.DynamicCombo.Option("Seedance 2.0 Fast", _seedance2_text_inputs(["480p", "720p"])),
+                        IO.DynamicCombo.Option(
+                            "Seedance 2.0",
+                            _seedance2_text_inputs(["480p", "720p", "1080p"], default_ratio="adaptive"),
+                        ),
+                        IO.DynamicCombo.Option(
+                            "Seedance 2.0 Fast",
+                            _seedance2_text_inputs(["480p", "720p"], default_ratio="adaptive"),
+                        ),
                     ],
                     tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.",
                 ),
@@ -1588,9 +1595,9 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
         return IO.NodeOutput(await download_url_to_video_output(response.content.video_url))
 
 
-def _seedance2_reference_inputs(resolutions: list[str]):
+def _seedance2_reference_inputs(resolutions: list[str], default_ratio: str = "16:9"):
     return [
-        *_seedance2_text_inputs(resolutions),
+        *_seedance2_text_inputs(resolutions, default_ratio=default_ratio),
         IO.Autogrow.Input(
             "reference_images",
             template=IO.Autogrow.TemplateNames(
@@ -1668,8 +1675,14 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "model",
                     options=[
-                        IO.DynamicCombo.Option("Seedance 2.0", _seedance2_reference_inputs(["480p", "720p", "1080p"])),
-                        IO.DynamicCombo.Option("Seedance 2.0 Fast", _seedance2_reference_inputs(["480p", "720p"])),
+                        IO.DynamicCombo.Option(
+                            "Seedance 2.0",
+                            _seedance2_reference_inputs(["480p", "720p", "1080p"], default_ratio="adaptive"),
+                        ),
+                        IO.DynamicCombo.Option(
+                            "Seedance 2.0 Fast",
+                            _seedance2_reference_inputs(["480p", "720p"], default_ratio="adaptive"),
+                        ),
                     ],
                     tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.",
                 ),

comfy_extras/nodes_advanced_samplers.py

@@ -92,7 +92,7 @@ class SamplerEulerCFGpp(io.ComfyNode):
         return io.Schema(
             node_id="SamplerEulerCFGpp",
             display_name="SamplerEulerCFG++",
-            category="_for_testing",  # "sampling/custom_sampling/samplers"
+            category="experimental",  # "sampling/custom_sampling/samplers"
             inputs=[
                 io.Combo.Input("version", options=["regular", "alternative"], advanced=True),
             ],

comfy_extras/nodes_attention_multiply.py

@@ -25,7 +25,7 @@ class UNetSelfAttentionMultiply(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="UNetSelfAttentionMultiply",
-            category="_for_testing/attention_experiments",
+            category="experimental/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
                 io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
@@ -48,7 +48,7 @@ class UNetCrossAttentionMultiply(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="UNetCrossAttentionMultiply",
-            category="_for_testing/attention_experiments",
+            category="experimental/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
                 io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
@@ -72,7 +72,7 @@ class CLIPAttentionMultiply(io.ComfyNode):
         return io.Schema(
             node_id="CLIPAttentionMultiply",
             search_aliases=["clip attention scale", "text encoder attention"],
-            category="_for_testing/attention_experiments",
+            category="experimental/attention_experiments",
             inputs=[
                 io.Clip.Input("clip"),
                 io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
@@ -106,7 +106,7 @@ class UNetTemporalAttentionMultiply(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="UNetTemporalAttentionMultiply",
-            category="_for_testing/attention_experiments",
+            category="experimental/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
                 io.Float.Input("self_structural", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),

comfy_extras/nodes_audio_encoder.py

@@ -10,6 +10,7 @@ class AudioEncoderLoader(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="AudioEncoderLoader",
+            display_name="Load Audio Encoder",
             category="loaders",
             inputs=[
                 io.Combo.Input(

comfy_extras/nodes_bg_removal.py

@@ -0,0 +1,60 @@
+import folder_paths
+from typing_extensions import override
+from comfy_api.latest import ComfyExtension, IO
+from comfy.bg_removal_model import load
+
+
+class LoadBackgroundRemovalModel(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        files = folder_paths.get_filename_list("background_removal")
+        return IO.Schema(
+            node_id="LoadBackgroundRemovalModel",
+            display_name="Load Background Removal Model",
+            category="loaders",
+            inputs=[
+                IO.Combo.Input("bg_removal_name", options=sorted(files), tooltip="The model used to remove backgrounds from images"),
+            ],
+            outputs=[
+                IO.BackgroundRemoval.Output("bg_model")
+            ]
+        )
+    @classmethod
+    def execute(cls, bg_removal_name):
+        path = folder_paths.get_full_path_or_raise("background_removal", bg_removal_name)
+        bg = load(path)
+        if bg is None:
+            raise RuntimeError("ERROR: background model file is invalid and does not contain a valid background removal model.")
+        return IO.NodeOutput(bg)
+
+class RemoveBackground(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RemoveBackground",
+            display_name="Remove Background",
+            category="image/background removal",
+            inputs=[
+                IO.Image.Input("image", tooltip="Input image to remove the background from"),
+                IO.BackgroundRemoval.Input("bg_removal_model", tooltip="Background removal model used to generate the mask")
+            ],
+            outputs=[
+                IO.Mask.Output("mask", tooltip="Generated foreground mask")
+            ]
+        )
+    @classmethod
+    def execute(cls, image, bg_removal_model):
+        mask = bg_removal_model.encode_image(image)
+        return IO.NodeOutput(mask)
+
+class BackgroundRemovalExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
+        return [
+            LoadBackgroundRemovalModel,
+            RemoveBackground
+        ]
+
+
+async def comfy_entrypoint() -> BackgroundRemovalExtension:
+    return BackgroundRemovalExtension()

comfy_extras/nodes_camera_trajectory.py

@@ -153,7 +153,7 @@ class WanCameraEmbedding(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="WanCameraEmbedding",
-            category="camera",
+            category="conditioning/video_models",
             inputs=[
                 io.Combo.Input(
                     "camera_pose",

comfy_extras/nodes_compositing.py

@@ -203,7 +203,7 @@ class JoinImageWithAlpha(io.ComfyNode):
     @classmethod
     def execute(cls, image: torch.Tensor, alpha: torch.Tensor) -> io.NodeOutput:
         batch_size = max(len(image), len(alpha))
-        alpha = 1.0 - resize_mask(alpha, image.shape[1:])
+        alpha = 1.0 - resize_mask(alpha.to(image), image.shape[1:])
         alpha = comfy.utils.repeat_to_batch_size(alpha, batch_size)
         image = comfy.utils.repeat_to_batch_size(image, batch_size)
         return io.NodeOutput(torch.cat((image[..., :3], alpha.unsqueeze(-1)), dim=-1))

comfy_extras/nodes_cond.py

@@ -8,7 +8,7 @@ class CLIPTextEncodeControlnet(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="CLIPTextEncodeControlnet",
-            category="_for_testing/conditioning",
+            category="experimental/conditioning",
             inputs=[
                 io.Clip.Input("clip"),
                 io.Conditioning.Input("conditioning"),
@@ -35,7 +35,7 @@ class T5TokenizerOptions(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="T5TokenizerOptions",
-            category="_for_testing/conditioning",
+            category="experimental/conditioning",
             inputs=[
                 io.Clip.Input("clip"),
                 io.Int.Input("min_padding", default=0, min=0, max=10000, step=1, advanced=True),

comfy_extras/nodes_context_windows.py

@@ -10,7 +10,7 @@ class ContextWindowsManualNode(io.ComfyNode):
         return io.Schema(
             node_id="ContextWindowsManual",
             display_name="Context Windows (Manual)",
-            category="context",
+            category="model_patches",
             description="Manually set context windows.",
             inputs=[
                 io.Model.Input("model", tooltip="The model to apply context windows to during sampling."),

comfy_extras/nodes_custom_sampler.py

@@ -984,7 +984,7 @@ class AddNoise(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="AddNoise",
-            category="_for_testing/custom_sampling/noise",
+            category="experimental/custom_sampling/noise",
             is_experimental=True,
             inputs=[
                 io.Model.Input("model"),
@@ -1034,7 +1034,7 @@ class ManualSigmas(io.ComfyNode):
         return io.Schema(
             node_id="ManualSigmas",
             search_aliases=["custom noise schedule", "define sigmas"],
-            category="_for_testing/custom_sampling",
+            category="experimental/custom_sampling",
             is_experimental=True,
             inputs=[
                 io.String.Input("sigmas", default="1, 0.5", multiline=False)

comfy_extras/nodes_differential_diffusion.py

@@ -13,7 +13,7 @@ class DifferentialDiffusion(io.ComfyNode):
             node_id="DifferentialDiffusion",
             search_aliases=["inpaint gradient", "variable denoise strength"],
             display_name="Differential Diffusion",
-            category="_for_testing",
+            category="experimental",
             inputs=[
                 io.Model.Input("model"),
                 io.Float.Input(

comfy_extras/nodes_flux.py

@@ -102,7 +102,7 @@ class FluxDisableGuidance(io.ComfyNode):
     append = execute  # TODO: remove
 
 
-PREFERED_KONTEXT_RESOLUTIONS = [
+PREFERRED_KONTEXT_RESOLUTIONS = [
     (672, 1568),
     (688, 1504),
     (720, 1456),
@@ -143,7 +143,7 @@ class FluxKontextImageScale(io.ComfyNode):
         width = image.shape[2]
         height = image.shape[1]
         aspect_ratio = width / height
-        _, width, height = min((abs(aspect_ratio - w / h), w, h) for w, h in PREFERED_KONTEXT_RESOLUTIONS)
+        _, width, height = min((abs(aspect_ratio - w / h), w, h) for w, h in PREFERRED_KONTEXT_RESOLUTIONS)
         image = comfy.utils.common_upscale(image.movedim(-1, 1), width, height, "lanczos", "center").movedim(1, -1)
         return io.NodeOutput(image)
 

comfy_extras/nodes_fresca.py

@@ -60,7 +60,7 @@ class FreSca(io.ComfyNode):
             node_id="FreSca",
             search_aliases=["frequency guidance"],
             display_name="FreSca",
-            category="_for_testing",
+            category="experimental",
             description="Applies frequency-dependent scaling to the guidance",
             inputs=[
                 io.Model.Input("model"),

comfy_extras/nodes_hunyuan.py

@@ -131,6 +131,8 @@ class HunyuanVideo15SuperResolution(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="HunyuanVideo15SuperResolution",
+            display_name="Hunyuan Video 1.5 Super Resolution",
+            category="conditioning/video_models",
             inputs=[
                 io.Conditioning.Input("positive"),
                 io.Conditioning.Input("negative"),
@@ -381,6 +383,8 @@ class HunyuanRefinerLatent(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="HunyuanRefinerLatent",
+            display_name="Hunyuan Latent Refiner",
+            category="conditioning/video_models",
             inputs=[
                 io.Conditioning.Input("positive"),
                 io.Conditioning.Input("negative"),

comfy_extras/nodes_hunyuan3d.py

@@ -40,7 +40,7 @@ class Hunyuan3Dv2Conditioning(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="Hunyuan3Dv2Conditioning",
-            category="conditioning/video_models",
+            category="conditioning/3d_models",
             inputs=[
                 IO.ClipVisionOutput.Input("clip_vision_output"),
             ],
@@ -65,7 +65,7 @@ class Hunyuan3Dv2ConditioningMultiView(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="Hunyuan3Dv2ConditioningMultiView",
-            category="conditioning/video_models",
+            category="conditioning/3d_models",
             inputs=[
                 IO.ClipVisionOutput.Input("front", optional=True),
                 IO.ClipVisionOutput.Input("left", optional=True),
@@ -424,6 +424,7 @@ class VoxelToMeshBasic(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="VoxelToMeshBasic",
+            display_name="Voxel to Mesh (Basic)",
             category="3d",
             inputs=[
                 IO.Voxel.Input("voxel"),
@@ -453,6 +454,7 @@ class VoxelToMesh(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="VoxelToMesh",
+            display_name="Voxel to Mesh",
             category="3d",
             inputs=[
                 IO.Voxel.Input("voxel"),

comfy_extras/nodes_hypernetwork.py

@@ -102,6 +102,7 @@ class HypernetworkLoader(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="HypernetworkLoader",
+            display_name="Load Hypernetwork",
             category="loaders",
             inputs=[
                 IO.Model.Input("model"),

comfy_extras/nodes_lora_extract.py

@@ -91,7 +91,7 @@ class LoraSave(io.ComfyNode):
             node_id="LoraSave",
             search_aliases=["export lora"],
             display_name="Extract and Save Lora",
-            category="_for_testing",
+            category="experimental",
             inputs=[
                 io.String.Input("filename_prefix", default="loras/ComfyUI_extracted_lora"),
                 io.Int.Input("rank", default=8, min=1, max=4096, step=1, advanced=True),

comfy_extras/nodes_lt.py

@@ -106,12 +106,12 @@ class LTXVImgToVideoInplace(io.ComfyNode):
         if bypass:
             return (latent,)
 
-        samples = latent["samples"]
+        samples = latent["samples"].clone()
         _, height_scale_factor, width_scale_factor = (
             vae.downscale_index_formula
         )
 
-        batch, _, latent_frames, latent_height, latent_width = samples.shape
+        _, _, _, latent_height, latent_width = samples.shape
         width = latent_width * width_scale_factor
         height = latent_height * height_scale_factor
 
@@ -124,11 +124,7 @@ class LTXVImgToVideoInplace(io.ComfyNode):
 
         samples[:, :, :t.shape[2]] = t
 
-        conditioning_latent_frames_mask = torch.ones(
-            (batch, 1, latent_frames, 1, 1),
-            dtype=torch.float32,
-            device=samples.device,
-        )
+        conditioning_latent_frames_mask = get_noise_mask(latent)
         conditioning_latent_frames_mask[:, :, :t.shape[2]] = 1.0 - strength
 
         return io.NodeOutput({"samples": samples, "noise_mask": conditioning_latent_frames_mask})
@@ -236,7 +232,7 @@ class LTXVAddGuide(io.ComfyNode):
     def encode(cls, vae, latent_width, latent_height, images, scale_factors):
         time_scale_factor, width_scale_factor, height_scale_factor = scale_factors
         images = images[:(images.shape[0] - 1) // time_scale_factor * time_scale_factor + 1]
-        pixels = comfy.utils.common_upscale(images.movedim(-1, 1), latent_width * width_scale_factor, latent_height * height_scale_factor, "bilinear", crop="disabled").movedim(1, -1)
+        pixels = comfy.utils.common_upscale(images.movedim(-1, 1), latent_width * width_scale_factor, latent_height * height_scale_factor, "bilinear", crop="center").movedim(1, -1)
         encode_pixels = pixels[:, :, :, :3]
         t = vae.encode(encode_pixels)
         return encode_pixels, t
@@ -594,7 +590,8 @@ class LTXVPreprocess(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="LTXVPreprocess",
-            category="image",
+            display_name="LTXV Preprocess",
+            category="video/preprocessors",
             inputs=[
                 io.Image.Input("image"),
                 io.Int.Input(

comfy_extras/nodes_mahiro.py

@@ -11,7 +11,7 @@ class Mahiro(io.ComfyNode):
         return io.Schema(
             node_id="Mahiro",
             display_name="Positive-Biased Guidance",
-            category="_for_testing",
+            category="experimental",
             description="Modify the guidance to scale more on the 'direction' of the positive prompt rather than the difference between the negative prompt.",
             inputs=[
                 io.Model.Input("model"),

comfy_extras/nodes_mask.py

@@ -40,10 +40,21 @@ def composite(destination, source, x, y, mask = None, multiplier = 8, resize_sou
 
     inverse_mask = torch.ones_like(mask) - mask
 
-    source_portion = mask * source[..., :visible_height, :visible_width]
-    destination_portion = inverse_mask  * destination[..., top:bottom, left:right]
+    source_rgb = source[:, :3, :visible_height, :visible_width]
+    dest_slice = destination[..., top:bottom, left:right]
+
+    if destination.shape[1] == 4:
+        if torch.max(dest_slice) == 0:
+            destination[:, :3, top:bottom, left:right] = source_rgb
+            destination[:, 3:4, top:bottom, left:right] = mask
+        else:
+            destination[:, :3, top:bottom, left:right] = (mask * source_rgb) + (inverse_mask * dest_slice[:, :3])
+            destination[:, 3:4, top:bottom, left:right] = torch.max(mask, dest_slice[:, 3:4])
+    else:
+        source_portion = mask * source_rgb
+        destination_portion = inverse_mask * dest_slice
+        destination[..., top:bottom, left:right] = source_portion + destination_portion
 
-    destination[..., top:bottom, left:right] = source_portion + destination_portion
     return destination
 
 class LatentCompositeMasked(IO.ComfyNode):
@@ -84,18 +95,23 @@ class ImageCompositeMasked(IO.ComfyNode):
             display_name="Image Composite Masked",
             category="image",
             inputs=[
-                IO.Image.Input("destination"),
                 IO.Image.Input("source"),
                 IO.Int.Input("x", default=0, min=0, max=nodes.MAX_RESOLUTION, step=1),
                 IO.Int.Input("y", default=0, min=0, max=nodes.MAX_RESOLUTION, step=1),
                 IO.Boolean.Input("resize_source", default=False),
+                IO.Image.Input("destination", optional=True),
                 IO.Mask.Input("mask", optional=True),
             ],
             outputs=[IO.Image.Output()],
         )
 
     @classmethod
-    def execute(cls, destination, source, x, y, resize_source, mask = None) -> IO.NodeOutput:
+    def execute(cls, source, x, y, resize_source, destination = None, mask = None) -> IO.NodeOutput:
+        if destination is None: # transparent rgba
+            B, H, W, C = source.shape
+            destination = torch.zeros((B, H, W, 4), dtype=source.dtype, device=source.device)
+            if C == 3:
+                source = torch.nn.functional.pad(source, (0, 1), value=1.0)
         destination, source = node_helpers.image_alpha_fix(destination, source)
         destination = destination.clone().movedim(-1, 1)
         output = composite(destination, source.movedim(-1, 1), x, y, mask, 1, resize_source).movedim(1, -1)
@@ -381,7 +397,6 @@ class GrowMask(IO.ComfyNode):
 
     expand_mask = execute  # TODO: remove
 
-
 class ThresholdMask(IO.ComfyNode):
     @classmethod
     def define_schema(cls):

comfy_extras/nodes_math.py

@@ -70,7 +70,7 @@ class MathExpressionNode(io.ComfyNode):
         return io.Schema(
             node_id="ComfyMathExpression",
             display_name="Math Expression",
-            category="math",
+            category="logic",
             search_aliases=[
                 "expression", "formula", "calculate", "calculator",
                 "eval", "math",

comfy_extras/nodes_number_convert.py

@@ -21,7 +21,7 @@ class NumberConvertNode(io.ComfyNode):
         return io.Schema(
             node_id="ComfyNumberConvert",
             display_name="Number Convert",
-            category="math",
+            category="utils",
             search_aliases=[
                 "int to float", "float to int", "number convert",
                 "int2float", "float2int", "cast", "parse number",

comfy_extras/nodes_perpneg.py

@@ -24,8 +24,8 @@ class PerpNeg(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="PerpNeg",
-            display_name="Perp-Neg (DEPRECATED by PerpNegGuider)",
-            category="_for_testing",
+            display_name="Perp-Neg (DEPRECATED by Perp-Neg Guider)",
+            category="experimental",
             inputs=[
                 io.Model.Input("model"),
                 io.Conditioning.Input("empty_conditioning"),
@@ -127,7 +127,8 @@ class PerpNegGuider(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="PerpNegGuider",
-            category="_for_testing",
+            display_name="Perp-Neg Guider",
+            category="experimental",
             inputs=[
                 io.Model.Input("model"),
                 io.Conditioning.Input("positive"),

comfy_extras/nodes_photomaker.py

@@ -123,7 +123,7 @@ class PhotoMakerLoader(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="PhotoMakerLoader",
-            category="_for_testing/photomaker",
+            category="experimental/photomaker",
             inputs=[
                 io.Combo.Input("photomaker_model_name", options=folder_paths.get_filename_list("photomaker")),
             ],
@@ -149,7 +149,7 @@ class PhotoMakerEncode(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="PhotoMakerEncode",
-            category="_for_testing/photomaker",
+            category="experimental/photomaker",
             inputs=[
                 io.Photomaker.Input("photomaker"),
                 io.Image.Input("image"),

comfy_extras/nodes_post_processing.py

@@ -116,6 +116,7 @@ class Quantize(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ImageQuantize",
+            display_name="Quantize Image",
             category="image/postprocessing",
             inputs=[
                 io.Image.Input("image"),
@@ -181,6 +182,7 @@ class Sharpen(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ImageSharpen",
+            display_name="Sharpen Image",
             category="image/postprocessing",
             inputs=[
                 io.Image.Input("image"),
@@ -436,7 +438,7 @@ class ResizeImageMaskNode(io.ComfyNode):
             node_id="ResizeImageMaskNode",
             display_name="Resize Image/Mask",
             description="Resize an image or mask using various scaling methods.",
-            category="transform",
+            category="image/transform",
             search_aliases=["resize", "resize image", "resize mask", "scale", "scale image", "scale mask", "image resize", "change size", "dimensions", "shrink", "enlarge"],
             inputs=[
                 io.MatchType.Input("input", template=template),

comfy_extras/nodes_rtdetr.py

@@ -15,7 +15,7 @@ class RTDETR_detect(io.ComfyNode):
         return io.Schema(
             node_id="RTDETR_detect",
             display_name="RT-DETR Detect",
-            category="detection/",
+            category="detection",
             search_aliases=["bbox", "bounding box", "object detection", "coco"],
             inputs=[
                 io.Model.Input("model", display_name="model"),
@@ -71,7 +71,7 @@ class DrawBBoxes(io.ComfyNode):
         return io.Schema(
             node_id="DrawBBoxes",
             display_name="Draw BBoxes",
-            category="detection/",
+            category="detection",
             search_aliases=["bbox", "bounding box", "object detection", "rt_detr", "visualize detections", "coco"],
             inputs=[
                 io.Image.Input("image", optional=True),

comfy_extras/nodes_sag.py

@@ -113,7 +113,7 @@ class SelfAttentionGuidance(io.ComfyNode):
         return io.Schema(
             node_id="SelfAttentionGuidance",
             display_name="Self-Attention Guidance",
-            category="_for_testing",
+            category="experimental",
             inputs=[
                 io.Model.Input("model"),
                 io.Float.Input("scale", default=0.5, min=-2.0, max=5.0, step=0.01),

comfy_extras/nodes_sam3.py

@@ -93,7 +93,7 @@ class SAM3_Detect(io.ComfyNode):
         return io.Schema(
             node_id="SAM3_Detect",
             display_name="SAM3 Detect",
-            category="detection/",
+            category="detection",
             search_aliases=["sam3", "segment anything", "open vocabulary", "text detection", "segment"],
             inputs=[
                 io.Model.Input("model", display_name="model"),
@@ -265,7 +265,7 @@ class SAM3_VideoTrack(io.ComfyNode):
         return io.Schema(
             node_id="SAM3_VideoTrack",
             display_name="SAM3 Video Track",
-            category="detection/",
+            category="detection",
             search_aliases=["sam3", "video", "track", "propagate"],
             inputs=[
                 io.Image.Input("images", display_name="images", tooltip="Video frames as batched images"),
@@ -320,7 +320,7 @@ class SAM3_TrackPreview(io.ComfyNode):
         return io.Schema(
             node_id="SAM3_TrackPreview",
             display_name="SAM3 Track Preview",
-            category="detection/",
+            category="detection",
             inputs=[
                 SAM3TrackData.Input("track_data", display_name="track_data"),
                 io.Image.Input("images", display_name="images", optional=True),
@@ -478,7 +478,7 @@ class SAM3_TrackToMask(io.ComfyNode):
         return io.Schema(
             node_id="SAM3_TrackToMask",
             display_name="SAM3 Track to Mask",
-            category="detection/",
+            category="detection",
             inputs=[
                 SAM3TrackData.Input("track_data", display_name="track_data"),
                 io.String.Input("object_indices", display_name="object_indices", default="",

comfy_extras/nodes_stable_cascade.py

@@ -119,7 +119,7 @@ class StableCascade_SuperResolutionControlnet(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StableCascade_SuperResolutionControlnet",
-            category="_for_testing/stable_cascade",
+            category="experimental/stable_cascade",
             is_experimental=True,
             inputs=[
                 io.Image.Input("image"),

comfy_extras/nodes_textgen.py

@@ -26,7 +26,8 @@ class TextGenerate(io.ComfyNode):
 
         return io.Schema(
             node_id="TextGenerate",
-            category="textgen",
+            display_name="Generate Text",
+            category="text",
             search_aliases=["LLM", "gemma"],
             inputs=[
                 io.Clip.Input("clip"),
@@ -157,6 +158,7 @@ class TextGenerateLTX2Prompt(TextGenerate):
         parent_schema = super().define_schema()
         return io.Schema(
             node_id="TextGenerateLTX2Prompt",
+            display_name="Generate LTX2 Prompt",
             category=parent_schema.category,
             inputs=parent_schema.inputs,
             outputs=parent_schema.outputs,

comfy_extras/nodes_torch_compile.py

@@ -10,7 +10,7 @@ class TorchCompileModel(io.ComfyNode):
     def define_schema(cls) -> io.Schema:
         return io.Schema(
             node_id="TorchCompileModel",
-            category="_for_testing",
+            category="experimental",
             inputs=[
                 io.Model.Input("model"),
                 io.Combo.Input(

comfy_extras/nodes_train.py

@@ -1361,7 +1361,7 @@ class SaveLoRA(io.ComfyNode):
             node_id="SaveLoRA",
             search_aliases=["export lora"],
             display_name="Save LoRA Weights",
-            category="loaders",
+            category="advanced/model_merging",
             is_experimental=True,
             is_output_node=True,
             inputs=[

comfy_extras/nodes_video_model.py

@@ -15,7 +15,7 @@ class ImageOnlyCheckpointLoader:
     RETURN_TYPES = ("MODEL", "CLIP_VISION", "VAE")
     FUNCTION = "load_checkpoint"
 
-    CATEGORY = "loaders/video_models"
+    CATEGORY = "loaders"
 
     def load_checkpoint(self, ckpt_name, output_vae=True, output_clip=True):
         ckpt_path = folder_paths.get_full_path_or_raise("checkpoints", ckpt_name)

custom_nodes/websocket_image_save.py

@@ -22,7 +22,7 @@ class SaveImageWebsocket:
 
     OUTPUT_NODE = True
 
-    CATEGORY = "api/image"
+    CATEGORY = "image"
 
     def save_images(self, images):
         pbar = comfy.utils.ProgressBar(images.shape[0])
@@ -42,3 +42,7 @@ class SaveImageWebsocket:
 NODE_CLASS_MAPPINGS = {
     "SaveImageWebsocket": SaveImageWebsocket,
 }
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "SaveImageWebsocket": "Save Image (Websocket)",
+}

folder_paths.py

@@ -52,6 +52,8 @@ folder_names_and_paths["model_patches"] = ([os.path.join(models_dir, "model_patc
 
 folder_names_and_paths["audio_encoders"] = ([os.path.join(models_dir, "audio_encoders")], supported_pt_extensions)
 
+folder_names_and_paths["background_removal"] = ([os.path.join(models_dir, "background_removal")], supported_pt_extensions)
+
 folder_names_and_paths["frame_interpolation"] = ([os.path.join(models_dir, "frame_interpolation")], supported_pt_extensions)
 
 folder_names_and_paths["optical_flow"] = ([os.path.join(models_dir, "optical_flow")], supported_pt_extensions)

nodes.py

@@ -330,7 +330,7 @@ class VAEDecodeTiled:
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "decode"
 
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
 
     def decode(self, vae, samples, tile_size, overlap=64, temporal_size=64, temporal_overlap=8):
         if tile_size < overlap * 4:
@@ -377,7 +377,7 @@ class VAEEncodeTiled:
     RETURN_TYPES = ("LATENT",)
     FUNCTION = "encode"
 
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
 
     def encode(self, vae, pixels, tile_size, overlap, temporal_size=64, temporal_overlap=8):
         t = vae.encode_tiled(pixels, tile_x=tile_size, tile_y=tile_size, overlap=overlap, tile_t=temporal_size, overlap_t=temporal_overlap)
@@ -493,7 +493,7 @@ class SaveLatent:
 
     OUTPUT_NODE = True
 
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
 
     def save(self, samples, filename_prefix="ComfyUI", prompt=None, extra_pnginfo=None):
         full_output_folder, filename, counter, subfolder, filename_prefix = folder_paths.get_save_image_path(filename_prefix, self.output_dir)
@@ -538,7 +538,7 @@ class LoadLatent:
         files = [f for f in os.listdir(input_dir) if os.path.isfile(os.path.join(input_dir, f)) and f.endswith(".latent")]
         return {"required": {"latent": [sorted(files), ]}, }
 
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
 
     RETURN_TYPES = ("LATENT", )
     FUNCTION = "load"
@@ -1443,7 +1443,7 @@ class LatentBlend:
     RETURN_TYPES = ("LATENT",)
     FUNCTION = "blend"
 
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
 
     def blend(self, samples1, samples2, blend_factor:float, blend_mode: str="normal"):
 
@@ -2092,6 +2092,8 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "StyleModelLoader": "Load Style Model",
     "CLIPVisionLoader": "Load CLIP Vision",
     "UNETLoader": "Load Diffusion Model",
+    "unCLIPCheckpointLoader": "Load unCLIP Checkpoint",
+    "GLIGENLoader": "Load GLIGEN Model",
     # Conditioning
     "CLIPVisionEncode": "CLIP Vision Encode",
     "StyleModelApply": "Apply Style Model",
@@ -2140,7 +2142,7 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "ImageSharpen": "Sharpen Image",
     "ImageScaleToTotalPixels": "Scale Image to Total Pixels",
     "GetImageSize": "Get Image Size",
-    # _for_testing
+    # experimental
     "VAEDecodeTiled": "VAE Decode (Tiled)",
     "VAEEncodeTiled": "VAE Encode (Tiled)",
 }
@@ -2427,6 +2429,7 @@ async def init_builtin_extra_nodes():
         "nodes_number_convert.py",
         "nodes_painter.py",
         "nodes_curve.py",
+        "nodes_bg_removal.py",
         "nodes_rtdetr.py",
         "nodes_frame_interpolation.py",
         "nodes_sam3.py",

tests-unit/app_test/node_replace_manager_test.py

@@ -0,0 +1,90 @@
+"""Tests for NodeReplaceManager registration behavior."""
+import importlib
+import sys
+import types
+
+import pytest
+
+
+@pytest.fixture
+def NodeReplaceManager(monkeypatch):
+    """Provide NodeReplaceManager with `nodes` stubbed.
+
+    `app.node_replace_manager` does `import nodes` at module level, which pulls in
+    torch + the full ComfyUI graph. register() doesn't actually need it, so we
+    stub `nodes` per-test (via monkeypatch so it's torn down) and reload the
+    module so it picks up the stub instead of any cached real import.
+    """
+    fake_nodes = types.ModuleType("nodes")
+    fake_nodes.NODE_CLASS_MAPPINGS = {}
+    monkeypatch.setitem(sys.modules, "nodes", fake_nodes)
+    monkeypatch.delitem(sys.modules, "app.node_replace_manager", raising=False)
+    module = importlib.import_module("app.node_replace_manager")
+    yield module.NodeReplaceManager
+    # Drop the freshly-imported module so the next test (or a later real import
+    # of `nodes`) starts from a clean slate.
+    sys.modules.pop("app.node_replace_manager", None)
+
+
+class FakeNodeReplace:
+    """Lightweight stand-in for comfy_api.latest._io.NodeReplace."""
+    def __init__(self, new_node_id, old_node_id, old_widget_ids=None,
+                 input_mapping=None, output_mapping=None):
+        self.new_node_id = new_node_id
+        self.old_node_id = old_node_id
+        self.old_widget_ids = old_widget_ids
+        self.input_mapping = input_mapping
+        self.output_mapping = output_mapping
+
+
+def test_register_adds_replacement(NodeReplaceManager):
+    manager = NodeReplaceManager()
+    manager.register(FakeNodeReplace(new_node_id="NewNode", old_node_id="OldNode"))
+    assert manager.has_replacement("OldNode")
+    assert len(manager.get_replacement("OldNode")) == 1
+
+
+def test_register_allows_multiple_alternatives_for_same_old_node(NodeReplaceManager):
+    """Different new_node_ids for the same old_node_id should all be kept."""
+    manager = NodeReplaceManager()
+    manager.register(FakeNodeReplace(new_node_id="AltA", old_node_id="OldNode"))
+    manager.register(FakeNodeReplace(new_node_id="AltB", old_node_id="OldNode"))
+    replacements = manager.get_replacement("OldNode")
+    assert len(replacements) == 2
+    assert {r.new_node_id for r in replacements} == {"AltA", "AltB"}
+
+
+def test_register_is_idempotent_for_duplicate_pair(NodeReplaceManager):
+    """Re-registering the same (old_node_id, new_node_id) should be a no-op."""
+    manager = NodeReplaceManager()
+    manager.register(FakeNodeReplace(new_node_id="NewNode", old_node_id="OldNode"))
+    manager.register(FakeNodeReplace(new_node_id="NewNode", old_node_id="OldNode"))
+    manager.register(FakeNodeReplace(new_node_id="NewNode", old_node_id="OldNode"))
+    assert len(manager.get_replacement("OldNode")) == 1
+
+
+def test_register_idempotent_preserves_first_registration(NodeReplaceManager):
+    """First registration wins; later duplicates with different mappings are ignored."""
+    manager = NodeReplaceManager()
+    first = FakeNodeReplace(
+        new_node_id="NewNode", old_node_id="OldNode",
+        input_mapping=[{"new_id": "a", "old_id": "x"}],
+    )
+    second = FakeNodeReplace(
+        new_node_id="NewNode", old_node_id="OldNode",
+        input_mapping=[{"new_id": "b", "old_id": "y"}],
+    )
+    manager.register(first)
+    manager.register(second)
+    replacements = manager.get_replacement("OldNode")
+    assert len(replacements) == 1
+    assert replacements[0] is first
+
+
+def test_register_dedupe_does_not_affect_other_old_nodes(NodeReplaceManager):
+    manager = NodeReplaceManager()
+    manager.register(FakeNodeReplace(new_node_id="NewA", old_node_id="OldA"))
+    manager.register(FakeNodeReplace(new_node_id="NewA", old_node_id="OldA"))
+    manager.register(FakeNodeReplace(new_node_id="NewB", old_node_id="OldB"))
+    assert len(manager.get_replacement("OldA")) == 1
+    assert len(manager.get_replacement("OldB")) == 1

tests/execution/testing_nodes/testing-pack/api_test_nodes.py

@@ -21,7 +21,7 @@ class TestAsyncProgressUpdate(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "execute"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def execute(self, value, sleep_seconds):
         start = time.time()
@@ -51,7 +51,7 @@ class TestSyncProgressUpdate(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "execute"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     def execute(self, value, sleep_seconds):
         start = time.time()

tests/execution/testing_nodes/testing-pack/async_test_nodes.py

@@ -21,7 +21,7 @@ class TestAsyncValidation(ComfyNodeABC):
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "process"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     @classmethod
     async def VALIDATE_INPUTS(cls, value, threshold):
@@ -53,7 +53,7 @@ class TestAsyncError(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "error_execution"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def error_execution(self, value, error_after):
         await asyncio.sleep(error_after)
@@ -74,7 +74,7 @@ class TestAsyncValidationError(ComfyNodeABC):
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "process"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     @classmethod
     async def VALIDATE_INPUTS(cls, value, max_value):
@@ -105,7 +105,7 @@ class TestAsyncTimeout(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "timeout_execution"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def timeout_execution(self, value, timeout, operation_time):
         try:
@@ -129,7 +129,7 @@ class TestSyncError(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "sync_error"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     def sync_error(self, value):
         raise RuntimeError("Intentional sync execution error for testing")
@@ -150,7 +150,7 @@ class TestAsyncLazyCheck(ComfyNodeABC):
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "process"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def check_lazy_status(self, condition, input1, input2):
         # Simulate async checking (e.g., querying remote service)
@@ -184,7 +184,7 @@ class TestDynamicAsyncGeneration(ComfyNodeABC):
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "generate_async_workflow"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     def generate_async_workflow(self, image1, image2, num_async_nodes, sleep_duration):
         g = GraphBuilder()
@@ -229,7 +229,7 @@ class TestAsyncResourceUser(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "use_resource"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def use_resource(self, value, resource_id, duration):
         # Check if resource is already in use
@@ -265,7 +265,7 @@ class TestAsyncBatchProcessing(ComfyNodeABC):
 
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "process_batch"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def process_batch(self, images, process_time_per_item, unique_id):
         batch_size = images.shape[0]
@@ -305,7 +305,7 @@ class TestAsyncConcurrentLimit(ComfyNodeABC):
 
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "limited_execution"
-    CATEGORY = "_for_testing/async"
+    CATEGORY = "experimental/async"
 
     async def limited_execution(self, value, duration, node_id):
         async with self._semaphore:

tests/execution/testing_nodes/testing-pack/specific_tests.py

@@ -409,7 +409,7 @@ class TestSleep(ComfyNodeABC):
     RETURN_TYPES = (IO.ANY,)
     FUNCTION = "sleep"
 
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
 
     async def sleep(self, value, seconds, unique_id):
         pbar = ProgressBar(seconds, node_id=unique_id)
@@ -440,7 +440,7 @@ class TestParallelSleep(ComfyNodeABC):
         }
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "parallel_sleep"
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
     OUTPUT_NODE = True
 
     def parallel_sleep(self, image1, image2, image3, sleep1, sleep2, sleep3, unique_id):
@@ -474,7 +474,7 @@ class TestOutputNodeWithSocketOutput:
         }
     RETURN_TYPES = ("IMAGE",)
     FUNCTION = "process"
-    CATEGORY = "_for_testing"
+    CATEGORY = "experimental"
     OUTPUT_NODE = True
 
     def process(self, image, value):