diff --git a/blueprints/Color Adjustment.json b/blueprints/Color Adjustment.json index c599f7213..47f3df783 100644 --- a/blueprints/Color Adjustment.json +++ b/blueprints/Color Adjustment.json @@ -1 +1 @@ -{"revision": 0, "last_node_id": 14, "last_link_id": 0, "nodes": [{"id": 14, "type": "36677b92-5dd8-47a5-9380-4da982c1894f", "pos": [3610, -2630], "size": [270, 150], "flags": {}, "order": 3, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "IMAGE", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}], "properties": {"proxyWidgets": [["4", "value"], ["5", "value"], ["7", "value"], ["6", "value"]]}, "widgets_values": [], "title": "Color Adjustment"}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "36677b92-5dd8-47a5-9380-4da982c1894f", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 16, "lastLinkId": 36, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Color Adjustment", "inputNode": {"id": -10, "bounding": [3110, -3560, 120, 60]}, "outputNode": {"id": -20, "bounding": [4070, -3560, 120, 60]}, "inputs": [{"id": "0431d493-5f28-4430-bd00-84733997fc08", "name": "images.image0", "type": "IMAGE", "linkIds": [29], "localized_name": "images.image0", "label": "image", "pos": [3210, -3540]}], "outputs": [{"id": "bee8ea06-a114-4612-8937-939f2c927bdb", "name": "IMAGE0", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE0", "label": "IMAGE", "pos": [4090, -3540]}], "widgets": [], "nodes": [{"id": 15, "type": "GLSLShader", "pos": [3590, -3940], "size": [420, 252], "flags": {}, "order": 4, "mode": 0, "inputs": [{"label": "image0", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 29}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}, {"label": "u_float0", "localized_name": "floats.u_float0", "name": "floats.u_float0", "shape": 7, "type": "FLOAT", "link": 34}, {"label": "u_float1", "localized_name": "floats.u_float1", "name": "floats.u_float1", "shape": 7, "type": "FLOAT", "link": 30}, {"label": "u_float2", "localized_name": "floats.u_float2", "name": "floats.u_float2", "shape": 7, "type": "FLOAT", "link": 31}, {"label": "u_float3", "localized_name": "floats.u_float3", "name": "floats.u_float3", "shape": 7, "type": "FLOAT", "link": 33}, {"label": "u_float4", "localized_name": "floats.u_float4", "name": "floats.u_float4", "shape": 7, "type": "FLOAT", "link": null}, {"label": "u_int0", "localized_name": "ints.u_int0", "name": "ints.u_int0", "shape": 7, "type": "INT", "link": null}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}], "outputs": [{"localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [28]}, {"localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": null}], "properties": {"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}", "from_input"]}, {"id": 6, "type": "PrimitiveFloat", "pos": [3290, -3610], "size": [270, 58], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "vibrance", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [26, 31]}], "title": "Vibrance", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [128, 128, 128]}, {"offset": 1, "color": [255, 0, 0]}]}, "widgets_values": [0]}, {"id": 7, "type": "PrimitiveFloat", "pos": [3290, -3720], "size": [270, 58], "flags": {}, "order": 1, "mode": 0, "inputs": [{"label": "saturation", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [33]}], "title": "Saturation", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [128, 128, 128]}, {"offset": 1, "color": [255, 0, 0]}]}, "widgets_values": [0]}, {"id": 5, "type": "PrimitiveFloat", "pos": [3290, -3830], "size": [270, 58], "flags": {}, "order": 2, "mode": 0, "inputs": [{"label": "tint", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [30]}], "title": "Tint", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [0, 255, 0]}, {"offset": 0.5, "color": [255, 255, 255]}, {"offset": 1, "color": [255, 0, 255]}]}, "widgets_values": [0]}, {"id": 4, "type": "PrimitiveFloat", "pos": [3290, -3940], "size": [270, 58], "flags": {}, "order": 3, "mode": 0, "inputs": [{"label": "temperature", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [34]}], "title": "Temperature", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [68, 136, 255]}, {"offset": 0.5, "color": [255, 255, 255]}, {"offset": 1, "color": [255, 136, 0]}]}, "widgets_values": [100]}], "groups": [], "links": [{"id": 34, "origin_id": 4, "origin_slot": 0, "target_id": 15, "target_slot": 2, "type": "FLOAT"}, {"id": 30, "origin_id": 5, "origin_slot": 0, "target_id": 15, "target_slot": 3, "type": "FLOAT"}, {"id": 31, "origin_id": 6, "origin_slot": 0, "target_id": 15, "target_slot": 4, "type": "FLOAT"}, {"id": 33, "origin_id": 7, "origin_slot": 0, "target_id": 15, "target_slot": 5, "type": "FLOAT"}, {"id": 29, "origin_id": -10, "origin_slot": 0, "target_id": 15, "target_slot": 0, "type": "IMAGE"}, {"id": 28, "origin_id": 15, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Color adjust"}]}} +{"revision": 0, "last_node_id": 14, "last_link_id": 0, "nodes": [{"id": 14, "type": "36677b92-5dd8-47a5-9380-4da982c1894f", "pos": [3610, -2630], "size": [270, 150], "flags": {}, "order": 3, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "IMAGE", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}], "properties": {"proxyWidgets": [["4", "value"], ["5", "value"], ["7", "value"], ["6", "value"]]}, "widgets_values": [], "title": "Color Adjustment"}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "36677b92-5dd8-47a5-9380-4da982c1894f", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 16, "lastLinkId": 36, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Color Adjustment", "inputNode": {"id": -10, "bounding": [3110, -3560, 120, 60]}, "outputNode": {"id": -20, "bounding": [4070, -3560, 120, 60]}, "inputs": [{"id": "0431d493-5f28-4430-bd00-84733997fc08", "name": "images.image0", "type": "IMAGE", "linkIds": [29], "localized_name": "images.image0", "label": "image", "pos": [3210, -3540]}], "outputs": [{"id": "bee8ea06-a114-4612-8937-939f2c927bdb", "name": "IMAGE0", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE0", "label": "IMAGE", "pos": [4090, -3540]}], "widgets": [], "nodes": [{"id": 15, "type": "GLSLShader", "pos": [3590, -3940], "size": [420, 252], "flags": {}, "order": 4, "mode": 0, "inputs": [{"label": "image0", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 29}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}, {"label": "u_float0", "localized_name": "floats.u_float0", "name": "floats.u_float0", "shape": 7, "type": "FLOAT", "link": 34}, {"label": "u_float1", "localized_name": "floats.u_float1", "name": "floats.u_float1", "shape": 7, "type": "FLOAT", "link": 30}, {"label": "u_float2", "localized_name": "floats.u_float2", "name": "floats.u_float2", "shape": 7, "type": "FLOAT", "link": 31}, {"label": "u_float3", "localized_name": "floats.u_float3", "name": "floats.u_float3", "shape": 7, "type": "FLOAT", "link": 33}, {"label": "u_float4", "localized_name": "floats.u_float4", "name": "floats.u_float4", "shape": 7, "type": "FLOAT", "link": null}, {"label": "u_int0", "localized_name": "ints.u_int0", "name": "ints.u_int0", "shape": 7, "type": "INT", "link": null}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}], "outputs": [{"localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [28]}, {"localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": null}], "properties": {"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}", "from_input"]}, {"id": 6, "type": "PrimitiveFloat", "pos": [3290, -3610], "size": [270, 58], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "vibrance", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [26, 31]}], "title": "Vibrance", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [128, 128, 128]}, {"offset": 1, "color": [255, 0, 0]}]}, "widgets_values": [0]}, {"id": 7, "type": "PrimitiveFloat", "pos": [3290, -3720], "size": [270, 58], "flags": {}, "order": 1, "mode": 0, "inputs": [{"label": "saturation", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [33]}], "title": "Saturation", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [128, 128, 128]}, {"offset": 1, "color": [255, 0, 0]}]}, "widgets_values": [0]}, {"id": 5, "type": "PrimitiveFloat", "pos": [3290, -3830], "size": [270, 58], "flags": {}, "order": 2, "mode": 0, "inputs": [{"label": "tint", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [30]}], "title": "Tint", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [0, 255, 0]}, {"offset": 0.5, "color": [255, 255, 255]}, {"offset": 1, "color": [255, 0, 255]}]}, "widgets_values": [0]}, {"id": 4, "type": "PrimitiveFloat", "pos": [3290, -3940], "size": [270, 58], "flags": {}, "order": 3, "mode": 0, "inputs": [{"label": "temperature", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [34]}], "title": "Temperature", "properties": {"Node name for S&R": "PrimitiveFloat", "max": 100, "min": -100, "step": 1, "display": "gradientslider", "gradient_stops": [{"offset": 0, "color": [68, 136, 255]}, {"offset": 0.5, "color": [255, 255, 255]}, {"offset": 1, "color": [255, 136, 0]}]}, "widgets_values": [0]}], "groups": [], "links": [{"id": 34, "origin_id": 4, "origin_slot": 0, "target_id": 15, "target_slot": 2, "type": "FLOAT"}, {"id": 30, "origin_id": 5, "origin_slot": 0, "target_id": 15, "target_slot": 3, "type": "FLOAT"}, {"id": 31, "origin_id": 6, "origin_slot": 0, "target_id": 15, "target_slot": 4, "type": "FLOAT"}, {"id": 33, "origin_id": 7, "origin_slot": 0, "target_id": 15, "target_slot": 5, "type": "FLOAT"}, {"id": 29, "origin_id": -10, "origin_slot": 0, "target_id": 15, "target_slot": 0, "type": "IMAGE"}, {"id": 28, "origin_id": 15, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Color adjust"}]}} diff --git a/comfy/ldm/flux/model.py b/comfy/ldm/flux/model.py index 8e7912e6d..2020326c2 100644 --- a/comfy/ldm/flux/model.py +++ b/comfy/ldm/flux/model.py @@ -386,7 +386,7 @@ class Flux(nn.Module): h = max(h, ref.shape[-2] + h_offset) w = max(w, ref.shape[-1] + w_offset) - kontext, kontext_ids = self.process_img(ref, index=index, h_offset=h_offset, w_offset=w_offset) + kontext, kontext_ids = self.process_img(ref, index=index, h_offset=h_offset, w_offset=w_offset, transformer_options=transformer_options) img = torch.cat([img, kontext], dim=1) img_ids = torch.cat([img_ids, kontext_ids], dim=1) ref_num_tokens.append(kontext.shape[1]) diff --git a/comfy/ldm/lightricks/av_model.py b/comfy/ldm/lightricks/av_model.py index 08d686b7b..6f2ba41ef 100644 --- a/comfy/ldm/lightricks/av_model.py +++ b/comfy/ldm/lightricks/av_model.py @@ -681,6 +681,33 @@ class LTXAVModel(LTXVModel): additional_args["has_spatial_mask"] = has_spatial_mask ax, a_latent_coords = self.a_patchifier.patchify(ax) + + # Inject reference audio for ID-LoRA in-context conditioning + ref_audio = kwargs.get("ref_audio", None) + ref_audio_seq_len = 0 + if ref_audio is not None: + ref_tokens = ref_audio["tokens"].to(dtype=ax.dtype, device=ax.device) + if ref_tokens.shape[0] < ax.shape[0]: + ref_tokens = ref_tokens.expand(ax.shape[0], -1, -1) + ref_audio_seq_len = ref_tokens.shape[1] + B = ax.shape[0] + + # Compute negative temporal positions matching ID-LoRA convention: + # offset by -(end_of_last_token + time_per_latent) so reference ends just before t=0 + p = self.a_patchifier + tpl = p.hop_length * p.audio_latent_downsample_factor / p.sample_rate + ref_start = p._get_audio_latent_time_in_sec(0, ref_audio_seq_len, torch.float32, ax.device) + ref_end = p._get_audio_latent_time_in_sec(1, ref_audio_seq_len + 1, torch.float32, ax.device) + time_offset = ref_end[-1].item() + tpl + ref_start = (ref_start - time_offset).unsqueeze(0).expand(B, -1).unsqueeze(1) + ref_end = (ref_end - time_offset).unsqueeze(0).expand(B, -1).unsqueeze(1) + ref_pos = torch.stack([ref_start, ref_end], dim=-1) + + additional_args["ref_audio_seq_len"] = ref_audio_seq_len + additional_args["target_audio_seq_len"] = ax.shape[1] + ax = torch.cat([ref_tokens, ax], dim=1) + a_latent_coords = torch.cat([ref_pos.to(a_latent_coords), a_latent_coords], dim=2) + ax = self.audio_patchify_proj(ax) # additional_args.update({"av_orig_shape": list(x.shape)}) @@ -721,6 +748,14 @@ class LTXAVModel(LTXVModel): # Prepare audio timestep a_timestep = kwargs.get("a_timestep") + ref_audio_seq_len = kwargs.get("ref_audio_seq_len", 0) + if ref_audio_seq_len > 0 and a_timestep is not None: + # Reference tokens must have timestep=0, expand scalar/1D timestep to per-token so ref=0 and target=sigma. + target_len = kwargs.get("target_audio_seq_len") + if a_timestep.dim() <= 1: + a_timestep = a_timestep.view(-1, 1).expand(batch_size, target_len) + ref_ts = torch.zeros(batch_size, ref_audio_seq_len, *a_timestep.shape[2:], device=a_timestep.device, dtype=a_timestep.dtype) + a_timestep = torch.cat([ref_ts, a_timestep], dim=1) if a_timestep is not None: a_timestep_scaled = a_timestep * self.timestep_scale_multiplier a_timestep_flat = a_timestep_scaled.flatten() @@ -955,6 +990,13 @@ class LTXAVModel(LTXVModel): v_embedded_timestep = embedded_timestep[0] a_embedded_timestep = embedded_timestep[1] + # Trim reference audio tokens before unpatchification + ref_audio_seq_len = kwargs.get("ref_audio_seq_len", 0) + if ref_audio_seq_len > 0: + ax = ax[:, ref_audio_seq_len:] + if a_embedded_timestep.shape[1] > 1: + a_embedded_timestep = a_embedded_timestep[:, ref_audio_seq_len:] + # Expand compressed video timestep if needed if isinstance(v_embedded_timestep, CompressedTimestep): v_embedded_timestep = v_embedded_timestep.expand() diff --git a/comfy/ldm/wan/vae.py b/comfy/ldm/wan/vae.py index deeb8695b..57b0dabf7 100644 --- a/comfy/ldm/wan/vae.py +++ b/comfy/ldm/wan/vae.py @@ -376,11 +376,16 @@ class Decoder3d(nn.Module): return layer = self.upsamples[layer_idx] - if isinstance(layer, Resample) and layer.mode == 'upsample3d' and x.shape[2] > 1: - for frame_idx in range(x.shape[2]): + if feat_cache is not None: + x = layer(x, feat_cache, feat_idx) + else: + x = layer(x) + + if isinstance(layer, Resample) and layer.mode == 'upsample3d' and x.shape[2] > 2: + for frame_idx in range(0, x.shape[2], 2): self.run_up( - layer_idx, - [x[:, :, frame_idx:frame_idx + 1, :, :]], + layer_idx + 1, + [x[:, :, frame_idx:frame_idx + 2, :, :]], feat_cache, feat_idx.copy(), out_chunks, @@ -388,11 +393,6 @@ class Decoder3d(nn.Module): del x return - if feat_cache is not None: - x = layer(x, feat_cache, feat_idx) - else: - x = layer(x) - next_x_ref = [x] del x self.run_up(layer_idx + 1, next_x_ref, feat_cache, feat_idx, out_chunks) diff --git a/comfy/model_base.py b/comfy/model_base.py index 43ec93324..70aff886e 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -937,9 +937,10 @@ class LongCatImage(Flux): transformer_options = transformer_options.copy() rope_opts = transformer_options.get("rope_options", {}) rope_opts = dict(rope_opts) + pe_len = float(c_crossattn.shape[1]) if c_crossattn is not None else 512.0 rope_opts.setdefault("shift_t", 1.0) - rope_opts.setdefault("shift_y", 512.0) - rope_opts.setdefault("shift_x", 512.0) + rope_opts.setdefault("shift_y", pe_len) + rope_opts.setdefault("shift_x", pe_len) transformer_options["rope_options"] = rope_opts return super()._apply_model(x, t, c_concat, c_crossattn, control, transformer_options, **kwargs) @@ -1060,6 +1061,10 @@ class LTXAV(BaseModel): if guide_attention_entries is not None: out['guide_attention_entries'] = comfy.conds.CONDConstant(guide_attention_entries) + ref_audio = kwargs.get("ref_audio", None) + if ref_audio is not None: + out['ref_audio'] = comfy.conds.CONDConstant(ref_audio) + return out def process_timestep(self, timestep, x, denoise_mask=None, audio_denoise_mask=None, **kwargs): diff --git a/comfy/sample.py b/comfy/sample.py index e9c2259ab..653829582 100644 --- a/comfy/sample.py +++ b/comfy/sample.py @@ -8,12 +8,12 @@ import comfy.nested_tensor def prepare_noise_inner(latent_image, generator, noise_inds=None): if noise_inds is None: - return torch.randn(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu") + return torch.randn(latent_image.size(), dtype=torch.float32, layout=latent_image.layout, generator=generator, device="cpu").to(dtype=latent_image.dtype) unique_inds, inverse = np.unique(noise_inds, return_inverse=True) noises = [] for i in range(unique_inds[-1]+1): - noise = torch.randn([1] + list(latent_image.size())[1:], dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu") + noise = torch.randn([1] + list(latent_image.size())[1:], dtype=torch.float32, layout=latent_image.layout, generator=generator, device="cpu").to(dtype=latent_image.dtype) if i in unique_inds: noises.append(noise) noises = [noises[i] for i in inverse] diff --git a/comfy/samplers.py b/comfy/samplers.py index 8be449ef7..0a4d062db 100755 --- a/comfy/samplers.py +++ b/comfy/samplers.py @@ -985,8 +985,8 @@ class CFGGuider: self.inner_model, self.conds, self.loaded_models = comfy.sampler_helpers.prepare_sampling(self.model_patcher, noise.shape, self.conds, self.model_options) device = self.model_patcher.load_device - noise = noise.to(device) - latent_image = latent_image.to(device) + noise = noise.to(device=device, dtype=torch.float32) + latent_image = latent_image.to(device=device, dtype=torch.float32) sigmas = sigmas.to(device) cast_to_load_options(self.model_options, device=device, dtype=self.model_patcher.model_dtype()) @@ -1028,6 +1028,7 @@ class CFGGuider: denoise_mask, _ = comfy.utils.pack_latents(denoise_masks) else: denoise_mask = denoise_masks[0] + denoise_mask = denoise_mask.float() self.conds = {} for k in self.original_conds: diff --git a/comfy/text_encoders/llama.py b/comfy/text_encoders/llama.py index ccc200b7a..9fdea999c 100644 --- a/comfy/text_encoders/llama.py +++ b/comfy/text_encoders/llama.py @@ -1028,12 +1028,19 @@ class Qwen25_7BVLI(BaseLlama, BaseGenerate, torch.nn.Module): grid = e.get("extra", None) start = e.get("index") if position_ids is None: - position_ids = torch.zeros((3, embeds.shape[1]), device=embeds.device) + position_ids = torch.ones((3, embeds.shape[1]), device=embeds.device, dtype=torch.long) position_ids[:, :start] = torch.arange(0, start, device=embeds.device) end = e.get("size") + start len_max = int(grid.max()) // 2 start_next = len_max + start - position_ids[:, end:] = torch.arange(start_next + offset, start_next + (embeds.shape[1] - end) + offset, device=embeds.device) + if attention_mask is not None: + # Assign compact sequential positions to attended tokens only, + # skipping over padding so post-padding tokens aren't inflated. + after_mask = attention_mask[0, end:] + text_positions = after_mask.cumsum(0) - 1 + start_next + offset + position_ids[:, end:] = torch.where(after_mask.bool(), text_positions, position_ids[0, end:]) + else: + position_ids[:, end:] = torch.arange(start_next + offset, start_next + (embeds.shape[1] - end) + offset, device=embeds.device) position_ids[0, start:end] = start + offset max_d = int(grid[0][1]) // 2 position_ids[1, start:end] = torch.arange(start + offset, start + max_d + offset, device=embeds.device).unsqueeze(1).repeat(1, math.ceil((end - start) / max_d)).flatten(0)[:end - start] diff --git a/comfy/text_encoders/longcat_image.py b/comfy/text_encoders/longcat_image.py index 882d80901..0962779e3 100644 --- a/comfy/text_encoders/longcat_image.py +++ b/comfy/text_encoders/longcat_image.py @@ -64,7 +64,13 @@ class LongCatImageBaseTokenizer(Qwen25_7BVLITokenizer): return [output] +IMAGE_PAD_TOKEN_ID = 151655 + class LongCatImageTokenizer(sd1_clip.SD1Tokenizer): + T2I_PREFIX = "<|im_start|>system\nAs an image captioning expert, generate a descriptive text prompt based on an image content, suitable for input to a text-to-image model.<|im_end|>\n<|im_start|>user\n" + EDIT_PREFIX = "<|im_start|>system\nAs an image editing expert, first analyze the content and attributes of the input image(s). Then, based on the user's editing instructions, clearly and precisely determine how to modify the given image(s), ensuring that only the specified parts are altered and all other aspects remain consistent with the original(s).<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>" + SUFFIX = "<|im_end|>\n<|im_start|>assistant\n" + def __init__(self, embedding_directory=None, tokenizer_data={}): super().__init__( embedding_directory=embedding_directory, @@ -72,10 +78,8 @@ class LongCatImageTokenizer(sd1_clip.SD1Tokenizer): name="qwen25_7b", tokenizer=LongCatImageBaseTokenizer, ) - self.longcat_template_prefix = "<|im_start|>system\nAs an image captioning expert, generate a descriptive text prompt based on an image content, suitable for input to a text-to-image model.<|im_end|>\n<|im_start|>user\n" - self.longcat_template_suffix = "<|im_end|>\n<|im_start|>assistant\n" - def tokenize_with_weights(self, text, return_word_ids=False, **kwargs): + def tokenize_with_weights(self, text, return_word_ids=False, images=None, **kwargs): skip_template = False if text.startswith("<|im_start|>"): skip_template = True @@ -90,11 +94,14 @@ class LongCatImageTokenizer(sd1_clip.SD1Tokenizer): text, return_word_ids=return_word_ids, disable_weights=True, **kwargs ) else: + has_images = images is not None and len(images) > 0 + template_prefix = self.EDIT_PREFIX if has_images else self.T2I_PREFIX + prefix_ids = base_tok.tokenizer( - self.longcat_template_prefix, add_special_tokens=False + template_prefix, add_special_tokens=False )["input_ids"] suffix_ids = base_tok.tokenizer( - self.longcat_template_suffix, add_special_tokens=False + self.SUFFIX, add_special_tokens=False )["input_ids"] prompt_tokens = base_tok.tokenize_with_weights( @@ -106,6 +113,14 @@ class LongCatImageTokenizer(sd1_clip.SD1Tokenizer): suffix_pairs = [(t, 1.0) for t in suffix_ids] combined = prefix_pairs + prompt_pairs + suffix_pairs + + if has_images: + embed_count = 0 + for i in range(len(combined)): + if combined[i][0] == IMAGE_PAD_TOKEN_ID and embed_count < len(images): + combined[i] = ({"type": "image", "data": images[embed_count], "original_type": "image"}, combined[i][1]) + embed_count += 1 + tokens = {"qwen25_7b": [combined]} return tokens diff --git a/comfy/text_encoders/qwen_vl.py b/comfy/text_encoders/qwen_vl.py index 3b18ce730..98c350a12 100644 --- a/comfy/text_encoders/qwen_vl.py +++ b/comfy/text_encoders/qwen_vl.py @@ -425,4 +425,7 @@ class Qwen2VLVisionTransformer(nn.Module): hidden_states = block(hidden_states, position_embeddings, cu_seqlens_now, optimized_attention=optimized_attention) hidden_states = self.merger(hidden_states) + # Potentially important for spatially precise edits. This is present in the HF implementation. + reverse_indices = torch.argsort(window_index) + hidden_states = hidden_states[reverse_indices, :] return hidden_states diff --git a/comfy_api_nodes/apis/grok.py b/comfy_api_nodes/apis/grok.py index c56c8aecc..fbedb53e0 100644 --- a/comfy_api_nodes/apis/grok.py +++ b/comfy_api_nodes/apis/grok.py @@ -29,13 +29,21 @@ class ImageEditRequest(BaseModel): class VideoGenerationRequest(BaseModel): model: str = Field(...) prompt: str = Field(...) - image: InputUrlObject | None = Field(...) + image: InputUrlObject | None = Field(None) + reference_images: list[InputUrlObject] | None = Field(None) duration: int = Field(...) aspect_ratio: str | None = Field(...) resolution: str = Field(...) seed: int = Field(...) +class VideoExtensionRequest(BaseModel): + prompt: str = Field(...) + video: InputUrlObject = Field(...) + duration: int = Field(default=6) + model: str | None = Field(default=None) + + class VideoEditRequest(BaseModel): model: str = Field(...) prompt: str = Field(...) diff --git a/comfy_api_nodes/nodes_grok.py b/comfy_api_nodes/nodes_grok.py index 0716d6239..dabc899d6 100644 --- a/comfy_api_nodes/nodes_grok.py +++ b/comfy_api_nodes/nodes_grok.py @@ -8,6 +8,7 @@ from comfy_api_nodes.apis.grok import ( ImageGenerationResponse, InputUrlObject, VideoEditRequest, + VideoExtensionRequest, VideoGenerationRequest, VideoGenerationResponse, VideoStatusResponse, @@ -21,6 +22,7 @@ from comfy_api_nodes.util import ( poll_op, sync_op, tensor_to_base64_string, + upload_images_to_comfyapi, upload_video_to_comfyapi, validate_string, validate_video_duration, @@ -33,6 +35,13 @@ def _extract_grok_price(response) -> float | None: return None +def _extract_grok_video_price(response) -> float | None: + price = _extract_grok_price(response) + if price is not None: + return price * 1.43 + return None + + class GrokImageNode(IO.ComfyNode): @classmethod @@ -354,6 +363,8 @@ class GrokVideoNode(IO.ComfyNode): seed: int, image: Input.Image | None = None, ) -> IO.NodeOutput: + if model == "grok-imagine-video-beta": + model = "grok-imagine-video" image_url = None if image is not None: if get_number_of_images(image) != 1: @@ -462,6 +473,244 @@ class GrokVideoEditNode(IO.ComfyNode): return IO.NodeOutput(await download_url_to_video_output(response.video.url)) +class GrokVideoReferenceNode(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="GrokVideoReferenceNode", + display_name="Grok Reference-to-Video", + category="api node/video/Grok", + description="Generate video guided by reference images as style and content references.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + tooltip="Text description of the desired video.", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "grok-imagine-video", + [ + IO.Autogrow.Input( + "reference_images", + template=IO.Autogrow.TemplatePrefix( + IO.Image.Input("image"), + prefix="reference_", + min=1, + max=7, + ), + tooltip="Up to 7 reference images to guide the video generation.", + ), + IO.Combo.Input( + "resolution", + options=["480p", "720p"], + tooltip="The resolution of the output video.", + ), + IO.Combo.Input( + "aspect_ratio", + options=["16:9", "4:3", "3:2", "1:1", "2:3", "3:4", "9:16"], + tooltip="The aspect ratio of the output video.", + ), + IO.Int.Input( + "duration", + default=6, + min=2, + max=10, + step=1, + tooltip="The duration of the output video in seconds.", + display_mode=IO.NumberDisplay.slider, + ), + ], + ), + ], + tooltip="The model to use for video generation.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to determine if node should re-run; " + "actual results are nondeterministic regardless of seed.", + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends( + widgets=["model.duration", "model.resolution"], + input_groups=["model.reference_images"], + ), + expr=""" + ( + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $refs := inputGroups["model.reference_images"]; + $rate := $res = "720p" ? 0.07 : 0.05; + $price := ($rate * $dur + 0.002 * $refs) * 1.43; + {"type":"usd","usd": $price} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int, + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + ref_image_urls = await upload_images_to_comfyapi( + cls, + list(model["reference_images"].values()), + mime_type="image/png", + wait_label="Uploading base images", + max_images=7, + ) + initial_response = await sync_op( + cls, + ApiEndpoint(path="/proxy/xai/v1/videos/generations", method="POST"), + data=VideoGenerationRequest( + model=model["model"], + reference_images=[InputUrlObject(url=i) for i in ref_image_urls], + prompt=prompt, + resolution=model["resolution"], + duration=model["duration"], + aspect_ratio=model["aspect_ratio"], + seed=seed, + ), + response_model=VideoGenerationResponse, + ) + response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/xai/v1/videos/{initial_response.request_id}"), + status_extractor=lambda r: r.status if r.status is not None else "complete", + response_model=VideoStatusResponse, + price_extractor=_extract_grok_video_price, + ) + return IO.NodeOutput(await download_url_to_video_output(response.video.url)) + + +class GrokVideoExtendNode(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="GrokVideoExtendNode", + display_name="Grok Video Extend", + category="api node/video/Grok", + description="Extend an existing video with a seamless continuation based on a text prompt.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + tooltip="Text description of what should happen next in the video.", + ), + IO.Video.Input("video", tooltip="Source video to extend. MP4 format, 2-15 seconds."), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "grok-imagine-video", + [ + IO.Int.Input( + "duration", + default=8, + min=2, + max=10, + step=1, + tooltip="Length of the extension in seconds.", + display_mode=IO.NumberDisplay.slider, + ), + ], + ), + ], + tooltip="The model to use for video extension.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to determine if node should re-run; " + "actual results are nondeterministic regardless of seed.", + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model.duration"]), + expr=""" + ( + $dur := $lookup(widgets, "model.duration"); + { + "type": "range_usd", + "min_usd": (0.02 + 0.05 * $dur) * 1.43, + "max_usd": (0.15 + 0.05 * $dur) * 1.43 + } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + video: Input.Video, + model: dict, + seed: int, + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + validate_video_duration(video, min_duration=2, max_duration=15) + video_size = get_fs_object_size(video.get_stream_source()) + if video_size > 50 * 1024 * 1024: + raise ValueError(f"Video size ({video_size / 1024 / 1024:.1f}MB) exceeds 50MB limit.") + initial_response = await sync_op( + cls, + ApiEndpoint(path="/proxy/xai/v1/videos/extensions", method="POST"), + data=VideoExtensionRequest( + prompt=prompt, + video=InputUrlObject(url=await upload_video_to_comfyapi(cls, video)), + duration=model["duration"], + ), + response_model=VideoGenerationResponse, + ) + response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/xai/v1/videos/{initial_response.request_id}"), + status_extractor=lambda r: r.status if r.status is not None else "complete", + response_model=VideoStatusResponse, + price_extractor=_extract_grok_video_price, + ) + return IO.NodeOutput(await download_url_to_video_output(response.video.url)) + + class GrokExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: @@ -469,7 +718,9 @@ class GrokExtension(ComfyExtension): GrokImageNode, GrokImageEditNode, GrokVideoNode, + GrokVideoReferenceNode, GrokVideoEditNode, + GrokVideoExtendNode, ] diff --git a/comfy_extras/nodes_lt.py b/comfy_extras/nodes_lt.py index c05571143..d7c2e8744 100644 --- a/comfy_extras/nodes_lt.py +++ b/comfy_extras/nodes_lt.py @@ -3,6 +3,7 @@ import node_helpers import torch import comfy.model_management import comfy.model_sampling +import comfy.samplers import comfy.utils import math import numpy as np @@ -682,6 +683,84 @@ class LTXVSeparateAVLatent(io.ComfyNode): return io.NodeOutput(video_latent, audio_latent) +class LTXVReferenceAudio(io.ComfyNode): + @classmethod + def define_schema(cls) -> io.Schema: + return io.Schema( + node_id="LTXVReferenceAudio", + display_name="LTXV Reference Audio (ID-LoRA)", + category="conditioning/audio", + description="Set reference audio for ID-LoRA speaker identity transfer. Encodes a reference audio clip into the conditioning and optionally patches the model with identity guidance (extra forward pass without reference, amplifying the speaker identity effect).", + inputs=[ + io.Model.Input("model"), + io.Conditioning.Input("positive"), + io.Conditioning.Input("negative"), + io.Audio.Input("reference_audio", tooltip="Reference audio clip whose speaker identity to transfer. ~5 seconds recommended (training duration). Shorter or longer clips may degrade voice identity transfer."), + io.Vae.Input(id="audio_vae", display_name="Audio VAE", tooltip="LTXV Audio VAE for encoding."), + io.Float.Input("identity_guidance_scale", default=3.0, min=0.0, max=100.0, step=0.01, round=0.01, tooltip="Strength of identity guidance. Runs an extra forward pass without reference each step to amplify speaker identity. Set to 0 to disable (no extra pass)."), + io.Float.Input("start_percent", default=0.0, min=0.0, max=1.0, step=0.001, advanced=True, tooltip="Start of the sigma range where identity guidance is active."), + io.Float.Input("end_percent", default=1.0, min=0.0, max=1.0, step=0.001, advanced=True, tooltip="End of the sigma range where identity guidance is active."), + ], + outputs=[ + io.Model.Output(), + io.Conditioning.Output(display_name="positive"), + io.Conditioning.Output(display_name="negative"), + ], + ) + + @classmethod + def execute(cls, model, positive, negative, reference_audio, audio_vae, identity_guidance_scale, start_percent, end_percent) -> io.NodeOutput: + # Encode reference audio to latents and patchify + audio_latents = audio_vae.encode(reference_audio) + b, c, t, f = audio_latents.shape + ref_tokens = audio_latents.permute(0, 2, 1, 3).reshape(b, t, c * f) + ref_audio = {"tokens": ref_tokens} + + positive = node_helpers.conditioning_set_values(positive, {"ref_audio": ref_audio}) + negative = node_helpers.conditioning_set_values(negative, {"ref_audio": ref_audio}) + + # Patch model with identity guidance + m = model.clone() + scale = identity_guidance_scale + model_sampling = m.get_model_object("model_sampling") + sigma_start = model_sampling.percent_to_sigma(start_percent) + sigma_end = model_sampling.percent_to_sigma(end_percent) + + def post_cfg_function(args): + if scale == 0: + return args["denoised"] + + sigma = args["sigma"] + sigma_ = sigma[0].item() + if sigma_ > sigma_start or sigma_ < sigma_end: + return args["denoised"] + + cond_pred = args["cond_denoised"] + cond = args["cond"] + cfg_result = args["denoised"] + model_options = args["model_options"].copy() + x = args["input"] + + # Strip ref_audio from conditioning for the no-reference pass + noref_cond = [] + for entry in cond: + new_entry = entry.copy() + mc = new_entry.get("model_conds", {}).copy() + mc.pop("ref_audio", None) + new_entry["model_conds"] = mc + noref_cond.append(new_entry) + + (pred_noref,) = comfy.samplers.calc_cond_batch( + args["model"], [noref_cond], x, sigma, model_options + ) + + return cfg_result + (cond_pred - pred_noref) * scale + + m.set_model_sampler_post_cfg_function(post_cfg_function) + + return io.NodeOutput(m, positive, negative) + + class LtxvExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[io.ComfyNode]]: @@ -697,6 +776,7 @@ class LtxvExtension(ComfyExtension): LTXVCropGuides, LTXVConcatAVLatent, LTXVSeparateAVLatent, + LTXVReferenceAudio, ] diff --git a/comfyui_version.py b/comfyui_version.py index a3b7204dc..61d7672ca 100644 --- a/comfyui_version.py +++ b/comfyui_version.py @@ -1,3 +1,3 @@ # This file is automatically generated by the build process when version is # updated in pyproject.toml. -__version__ = "0.18.0" +__version__ = "0.18.1" diff --git a/main.py b/main.py index f99aee38e..cd4483c67 100644 --- a/main.py +++ b/main.py @@ -471,6 +471,9 @@ if __name__ == "__main__": if sys.version_info.major == 3 and sys.version_info.minor < 10: logging.warning("WARNING: You are using a python version older than 3.10, please upgrade to a newer one. 3.12 and above is recommended.") + if args.disable_dynamic_vram: + logging.warning("Dynamic vram disabled with argument. If you have any issues with dynamic vram enabled please give us a detailed reports as this argument will be removed soon.") + event_loop, _, start_all_func = start_comfyui() try: x = start_all_func() diff --git a/manager_requirements.txt b/manager_requirements.txt index 5b06b56f6..90a2be84e 100644 --- a/manager_requirements.txt +++ b/manager_requirements.txt @@ -1 +1 @@ -comfyui_manager==4.1b6 \ No newline at end of file +comfyui_manager==4.1b8 diff --git a/pyproject.toml b/pyproject.toml index 6db9b1267..1fc9402a1 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,6 +1,6 @@ [project] name = "ComfyUI" -version = "0.18.0" +version = "0.18.1" readme = "README.md" license = { file = "LICENSE" } requires-python = ">=3.10" diff --git a/requirements.txt b/requirements.txt index ad0344ed4..26cc94354 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,4 +1,4 @@ -comfyui-frontend-package==1.41.21 +comfyui-frontend-package==1.42.8 comfyui-workflow-templates==0.9.26 comfyui-embedded-docs==0.4.3 torch