Merge 9a5e6233f4 into b08debceca

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

.coderabbit.yaml

@@ -4,12 +4,12 @@ early_access: false
 tone_instructions: "Only comment on issues introduced by this PR's changes. Do not flag pre-existing problems in moved, re-indented, or reformatted code."
 
 reviews:
-  profile: "chill"
-  request_changes_workflow: false
+  profile: "assertive"
+  request_changes_workflow: true
   high_level_summary: false
   poem: false
   review_status: false
-  review_details: false
+  review_details: true
   commit_status: true
   collapse_walkthrough: true
   changed_files_summary: false
@@ -39,6 +39,14 @@ reviews:
     - path: "**"
       instructions: |
         IMPORTANT: Only comment on issues directly introduced by this PR's code changes.
+        Treat AGENTS.md as mandatory repository policy, not optional style guidance.
+        Flag PR changes that violate AGENTS.md even when the code is otherwise functional.
+        In particular, enforce architecture boundaries, dtype/device/memory rules,
+        interface contracts, import style, no unnecessary try/except blocks, no inline
+        imports, no outbound internet paths in core ComfyUI, and narrow scoped fixes.
+        Prefer direct findings over suggestions when a rule is violated. Only ignore
+        AGENTS.md when it clearly conflicts with a newer explicit maintainer instruction
+        in the PR.
         Do NOT flag pre-existing issues in code that was merely moved, re-indented,
         de-indented, or reformatted without logic changes. If code appears in the diff
         only due to whitespace or structural reformatting (e.g., removing a `with:` block),
@@ -123,5 +131,10 @@ chat:
 
 knowledge_base:
   opt_out: false
+  code_guidelines:
+    enabled: true
+    filePatterns:
+      - files: "AGENTS.md"
+        applyTo: "**"
   learnings:
     scope: "auto"

comfy/samplers.py

@@ -782,15 +782,23 @@ def resolve_areas_and_cond_masks_multidim(conditions, dims, device):
             mask = c['mask']
             mask = mask.to(device=device)
             modified = c.copy()
-            if len(mask.shape) == len(dims):
+            # Normalize mask to [batch, *spatial_dims]
+            target_ndim = len(dims) + 1
+            while mask.ndim > target_ndim and mask.shape[0] == 1:
+                mask = mask.squeeze(0)
+            while mask.ndim < target_ndim:
                 mask = mask.unsqueeze(0)
             if mask.shape[1:] != dims:
-                if mask.ndim < 4:
+                if len(dims) == 1:
+                    mask = torch.nn.functional.interpolate(
+                        mask.unsqueeze(1), size=dims[0],
+                        mode='linear', align_corners=False).squeeze(1)
+                elif mask.ndim < 4:
                     mask = comfy.utils.common_upscale(mask.unsqueeze(1), dims[-1], dims[-2], 'bilinear', 'none').squeeze(1)
                 else:
                     mask = comfy.utils.common_upscale(mask, dims[-1], dims[-2], 'bilinear', 'none')
 
-            if modified.get("set_area_to_bounds", False): #TODO: handle dim != 2
+            if modified.get("set_area_to_bounds", False) and len(dims) == 2:
                 bounds = torch.max(torch.abs(mask),dim=0).values.unsqueeze(0)
                 boxes, is_empty = get_mask_aabb(bounds)
                 if is_empty[0]:

comfy/sd1_clip.py

@@ -543,18 +543,24 @@ class SDTokenizer:
     def _try_get_embedding(self, embedding_name:str):
         '''
         Takes a potential embedding name and tries to retrieve it.
-        Returns a Tuple consisting of the embedding and any leftover string, embedding can be None.
+        Returns a Tuple consisting of the embedding, the cleaned embedding name, and any leftover string, embedding can be None.
         '''
         split_embed = embedding_name.split()
         embedding_name = split_embed[0]
         leftover = ' '.join(split_embed[1:])
+
+        match = re.search(r'[<\[]', embedding_name)
+        if match is not None:
+            leftover = embedding_name[match.start():] + (" " + leftover if leftover else "")
+            embedding_name = embedding_name[:match.start()]
+
         embed = load_embed(embedding_name, self.embedding_directory, self.embedding_size, self.embedding_key)
         if embed is None:
             stripped = embedding_name.strip(',')
             if len(stripped) < len(embedding_name):
                 embed = load_embed(stripped, self.embedding_directory, self.embedding_size, self.embedding_key)
-                return (embed, "{} {}".format(embedding_name[len(stripped):], leftover))
-        return (embed, leftover)
+                return (embed, embedding_name, "{} {}".format(embedding_name[len(stripped):], leftover))
+        return (embed, embedding_name, leftover)
 
     def pad_tokens(self, tokens, amount):
         if self.pad_left:
@@ -585,7 +591,7 @@ class SDTokenizer:
         tokens = []
         for weighted_segment, weight in parsed_weights:
             to_tokenize = unescape_important(weighted_segment)
-            split = re.split(' {0}|\n{0}'.format(self.embedding_identifier), to_tokenize)
+            split = re.split(r'(?<=\s){}'.format(re.escape(self.embedding_identifier)), to_tokenize)
             to_tokenize = [split[0]]
             for i in range(1, len(split)):
                 to_tokenize.append("{}{}".format(self.embedding_identifier, split[i]))
@@ -595,7 +601,7 @@ class SDTokenizer:
                 # if we find an embedding, deal with the embedding
                 if word.startswith(self.embedding_identifier) and self.embedding_directory is not None:
                     embedding_name = word[len(self.embedding_identifier):].strip('\n')
-                    embed, leftover = self._try_get_embedding(embedding_name)
+                    embed, embedding_name, leftover = self._try_get_embedding(embedding_name)
                     if embed is None:
                         logging.warning(f"warning, embedding:{embedding_name} does not exist, ignoring")
                     else:

comfy/text_encoders/llama.py

@@ -937,22 +937,41 @@ class BaseGenerate:
             return torch.argmax(logits, dim=-1, keepdim=True)
 
         # Sampling mode
-        if repetition_penalty != 1.0:
-            for i in range(logits.shape[0]):
-                for token_id in set(token_history):
-                    logits[i, token_id] *= repetition_penalty if logits[i, token_id] < 0 else 1/repetition_penalty
-
-        if presence_penalty is not None and presence_penalty != 0.0:
-            for i in range(logits.shape[0]):
-                for token_id in set(token_history):
-                    logits[i, token_id] -= presence_penalty
+        if len(token_history) > 0 and (repetition_penalty != 1.0 or (presence_penalty is not None and presence_penalty != 0.0)):
+            token_ids = torch.tensor(list(set(token_history)), device=logits.device)
+            token_logits = logits[:, token_ids]
+            if repetition_penalty != 1.0:
+                token_logits = torch.where(token_logits < 0, token_logits * repetition_penalty, token_logits / repetition_penalty)
+            if presence_penalty is not None and presence_penalty != 0.0:
+                token_logits = token_logits - presence_penalty
+            logits[:, token_ids] = token_logits
 
         if temperature != 1.0:
             logits = logits / temperature
 
         if top_k > 0:
-            indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
-            logits[indices_to_remove] = torch.finfo(logits.dtype).min
+            top_k = min(top_k, logits.shape[-1])
+            logits, top_indices = torch.topk(logits, top_k)
+
+            if min_p > 0.0:
+                probs_before_filter = torch.nn.functional.softmax(logits, dim=-1)
+                top_probs, _ = probs_before_filter.max(dim=-1, keepdim=True)
+                min_threshold = min_p * top_probs
+                indices_to_remove = probs_before_filter < min_threshold
+                logits[indices_to_remove] = torch.finfo(logits.dtype).min
+
+            if top_p < 1.0:
+                sorted_logits, sorted_indices = torch.sort(logits, descending=True)
+                cumulative_probs = torch.cumsum(torch.nn.functional.softmax(sorted_logits, dim=-1), dim=-1)
+                sorted_indices_to_remove = cumulative_probs > top_p
+                sorted_indices_to_remove[..., 0] = False
+                indices_to_remove = torch.zeros_like(logits, dtype=torch.bool)
+                indices_to_remove.scatter_(1, sorted_indices, sorted_indices_to_remove)
+                logits[indices_to_remove] = torch.finfo(logits.dtype).min
+
+            probs = torch.nn.functional.softmax(logits, dim=-1)
+            next_token = torch.multinomial(probs, num_samples=1, generator=generator)
+            return top_indices.gather(1, next_token)
 
         if min_p > 0.0:
             probs_before_filter = torch.nn.functional.softmax(logits, dim=-1)

comfy/utils.py

@@ -1313,6 +1313,7 @@ def reshape_mask(input_mask, output_shape):
     dims = len(output_shape) - 2
 
     if dims == 1:
+        input_mask = input_mask.reshape((-1, 1, input_mask.shape[-1]))
         scale_mode = "linear"
 
     if dims == 2:

comfy_api_nodes/util/request_logger.py

@@ -9,6 +9,7 @@ from typing import Any
 import folder_paths
 
 logger = logging.getLogger(__name__)
+_SENSITIVE_HEADERS = {"authorization", "x-api-key"}
 
 
 def get_log_directory():
@@ -73,6 +74,10 @@ def _format_data_for_logging(data: Any) -> str:
     return str(data)
 
 
+def _redact_headers(headers: dict) -> dict:
+    return {k: ("***" if k.lower() in _SENSITIVE_HEADERS else v) for k, v in headers.items()}
+
+
 def log_request_response(
     operation_id: str,
     request_method: str,
@@ -101,7 +106,7 @@ def log_request_response(
         log_content.append(f"Method: {request_method}")
         log_content.append(f"URL: {request_url}")
         if request_headers:
-            log_content.append(f"Headers:\n{_format_data_for_logging(request_headers)}")
+            log_content.append(f"Headers:\n{_format_data_for_logging(_redact_headers(request_headers))}")
         if request_params:
             log_content.append(f"Params:\n{_format_data_for_logging(request_params)}")
         if request_data is not None:

comfy_extras/nodes_color.py

@@ -16,23 +16,30 @@ class ColorToRGBInt(io.ComfyNode):
             ],
             outputs=[
                 io.Int.Output(display_name="rgb_int"),
-                io.Color.Output(display_name="hex")
+                io.Color.Output(display_name="hex"),
+                io.Float.Output(display_name="alpha"),
             ],
         )
 
     @classmethod
     def execute(cls, color: str) -> io.NodeOutput:
-        # expect format #RRGGBB
-        if len(color) != 7 or color[0] != "#":
-            raise ValueError("Color must be in format #RRGGBB")
+        # expect format #RRGGBB or #RRGGBBAA
+        if len(color) not in (7, 9) or color[0] != "#":
+            raise ValueError("Color must be in format #RRGGBB or #RRGGBBAA")
         try:
             int(color[1:], 16)
         except ValueError:
-            raise ValueError("Color must be in format #RRGGBB") from None
+            raise ValueError("Color must be in format #RRGGBB or #RRGGBBAA") from None
+
+        alpha = 1.0
+        if len(color) == 9:
+            alpha = int(color[7:9], 16) / 255.0
+            color = color[:7]
+
         r, g, b = hex_to_rgb(color)
 
         rgb_int = r * 256 * 256 + g * 256 + b
-        return io.NodeOutput(rgb_int, color)
+        return io.NodeOutput(rgb_int, color, alpha)
 
 
 class ColorExtension(ComfyExtension):

requirements.txt

@@ -1,6 +1,6 @@
 comfyui-frontend-package==1.45.20
 comfyui-workflow-templates==0.11.2
-comfyui-embedded-docs==0.5.6
+comfyui-embedded-docs==0.5.7
 torch
 torchsde
 torchvision

tests-unit/comfy_test/mask_test.py

@@ -0,0 +1,336 @@
+"""
+Tests for mask handling across arbitrary spatial dimensions.
+
+Covers resolve_areas_and_cond_masks_multidim (conditioning masks) and
+reshape_mask (denoise masks) for 1D (audio), 2D (image), and 3D (video)
+spatial dims, including edge cases around batch size, spurious unsqueeze
+from upstream nodes, and size mismatches.
+"""
+
+import torch
+
+from comfy.samplers import resolve_areas_and_cond_masks_multidim
+from comfy.utils import reshape_mask
+
+
+def make_cond(mask):
+    """Create a minimal conditioning dict with a mask."""
+    return {"mask": mask, "model_conds": {}}
+
+
+def run_resolve(mask, dims, device="cpu"):
+    """Run resolve on a single condition and return the resolved mask."""
+    conds = [make_cond(mask)]
+    resolve_areas_and_cond_masks_multidim(conds, dims, device)
+    return conds[0]["mask"]
+
+
+# ============================================================
+# 1D spatial dims (audio models like AceStep v1.5)
+# dims = (length,), expected mask output: [batch, length]
+# ============================================================
+
+class Test1DSpatial:
+    """Tests for 1D spatial models (e.g. audio with noise shape [B, C, L])."""
+
+    def test_correct_shape_same_length(self):
+        """[B, L] mask with matching length — should pass through unchanged."""
+        mask = torch.ones(2, 100)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (2, 100)
+
+    def test_correct_shape_resize(self):
+        """[B, L] mask with different length — should resize via linear interp."""
+        mask = torch.ones(1, 50)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (1, 100)
+
+    def test_bare_spatial_mask(self):
+        """[L] mask (no batch) — should get batch dim added."""
+        mask = torch.ones(50)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (1, 100)
+
+    def test_spurious_unsqueeze_from_hooks(self):
+        """[1, B, L] mask (from set_mask_for_conditioning unsqueezing a [B, L] mask)
+        — should squeeze back to [B, L]."""
+        # Simulates: mask is [B, L], hooks.py does unsqueeze(0) -> [1, B, L]
+        mask = torch.ones(1, 2, 100)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (2, 100)
+
+    def test_spurious_unsqueeze_batch1(self):
+        """[1, 1, L] mask (batch=1, hooks added extra dim) — should become [1, L]."""
+        mask = torch.ones(1, 1, 50)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (1, 100)
+
+    def test_batch_gt1_same_length(self):
+        """[B, L] mask with batch=4 and matching length — no changes needed."""
+        mask = torch.rand(4, 100)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (4, 100)
+        torch.testing.assert_close(result, mask)
+
+    def test_batch_gt1_resize(self):
+        """[B, L] mask with batch=4 and different length — should resize each batch."""
+        mask = torch.rand(4, 50)
+        result = run_resolve(mask, dims=(100,))
+        assert result.shape == (4, 100)
+
+    def test_values_preserved_no_resize(self):
+        """Mask values should be preserved when no resize is needed."""
+        mask = torch.tensor([[0.0, 0.5, 1.0]])
+        result = run_resolve(mask, dims=(3,))
+        torch.testing.assert_close(result, mask)
+
+    def test_linear_interpolation_values(self):
+        """Check that linear interpolation produces sensible values."""
+        mask = torch.tensor([[0.0, 1.0]])  # [1, 2]
+        result = run_resolve(mask, dims=(5,))
+        assert result.shape == (1, 5)
+        # Should interpolate from 0 to 1
+        assert result[0, 0].item() < result[0, -1].item()
+
+    def test_set_area_to_bounds_skipped_for_1d(self):
+        """set_area_to_bounds should be skipped for 1D (no crash)."""
+        mask = torch.zeros(1, 100)
+        mask[0, 10:50] = 1.0
+        conds = [{"mask": mask, "model_conds": {}, "set_area_to_bounds": True}]
+        resolve_areas_and_cond_masks_multidim(conds, (100,), "cpu")
+        assert "area" not in conds[0]
+
+
+# ============================================================
+# 2D spatial dims (image models) — regression tests
+# dims = (H, W), expected mask output: [batch, H, W]
+# ============================================================
+
+class Test2DSpatial:
+    """Regression tests for standard 2D image models."""
+
+    def test_correct_shape_same_size(self):
+        """[B, H, W] mask matching dims — pass through."""
+        mask = torch.ones(1, 64, 64)
+        result = run_resolve(mask, dims=(64, 64))
+        assert result.shape == (1, 64, 64)
+
+    def test_bare_spatial_mask(self):
+        """[H, W] mask — should get batch dim added."""
+        mask = torch.ones(64, 64)
+        result = run_resolve(mask, dims=(64, 64))
+        assert result.shape == (1, 64, 64)
+
+    def test_resize_different_resolution(self):
+        """[B, H1, W1] mask with different size than dims — should bilinear resize."""
+        mask = torch.ones(1, 32, 32)
+        result = run_resolve(mask, dims=(64, 64))
+        assert result.shape == (1, 64, 64)
+
+    def test_4d_mask(self):
+        """[B, C, H, W] mask (4D) — should resize via common_upscale 4D path."""
+        mask = torch.ones(1, 1, 32, 32)
+        result = run_resolve(mask, dims=(64, 64))
+        assert result.shape == (1, 64, 64)
+
+    def test_batch_gt1(self):
+        """[B, H, W] mask with batch > 1."""
+        mask = torch.rand(4, 64, 64)
+        result = run_resolve(mask, dims=(64, 64))
+        assert result.shape == (4, 64, 64)
+
+    def test_batch_gt1_resize(self):
+        """[B, H, W] mask with batch > 1 and different resolution."""
+        mask = torch.rand(4, 32, 32)
+        result = run_resolve(mask, dims=(64, 64))
+        assert result.shape == (4, 64, 64)
+
+    def test_set_area_to_bounds(self):
+        """set_area_to_bounds should work for 2D masks."""
+        mask = torch.zeros(1, 64, 64)
+        mask[0, 10:20, 10:30] = 1.0
+        conds = [{"mask": mask, "model_conds": {}, "set_area_to_bounds": True}]
+        resolve_areas_and_cond_masks_multidim(conds, (64, 64), "cpu")
+        assert "area" in conds[0]
+
+    def test_non_square_resize(self):
+        """[B, H1, W1] mask resized to non-square dims."""
+        mask = torch.ones(1, 16, 32)
+        result = run_resolve(mask, dims=(64, 128))
+        assert result.shape == (1, 64, 128)
+
+
+# ============================================================
+# 3D spatial dims (video models)
+# dims = (T, H, W), expected mask output: [batch, T, H, W]
+# ============================================================
+
+class Test3DSpatial:
+    """Tests for 3D spatial models (e.g. video with noise shape [B, C, T, H, W])."""
+
+    def test_correct_shape_same_size(self):
+        """[B, T, H, W] mask matching dims — pass through."""
+        mask = torch.ones(1, 8, 64, 64)
+        result = run_resolve(mask, dims=(8, 64, 64))
+        assert result.shape == (1, 8, 64, 64)
+
+    def test_bare_spatial_mask(self):
+        """[T, H, W] mask — should get batch dim added."""
+        mask = torch.ones(8, 64, 64)
+        result = run_resolve(mask, dims=(8, 64, 64))
+        assert result.shape == (1, 8, 64, 64)
+
+    def test_resize_hw(self):
+        """[B, T, H1, W1] mask with different H, W — should resize last 2 dims."""
+        mask = torch.ones(1, 8, 32, 32)
+        result = run_resolve(mask, dims=(8, 64, 64))
+        assert result.shape == (1, 8, 64, 64)
+
+    def test_set_area_to_bounds_skipped_for_3d(self):
+        """set_area_to_bounds should be skipped for 3D (no crash)."""
+        mask = torch.zeros(1, 8, 64, 64)
+        mask[0, :, 10:20, 10:30] = 1.0
+        conds = [{"mask": mask, "model_conds": {}, "set_area_to_bounds": True}]
+        resolve_areas_and_cond_masks_multidim(conds, (8, 64, 64), "cpu")
+        assert "area" not in conds[0]
+
+
+class TestNoMask:
+    """Conditions without masks should pass through untouched."""
+
+    def test_no_mask_key(self):
+        """Condition with no mask key — untouched."""
+        conds = [{"model_conds": {}}]
+        resolve_areas_and_cond_masks_multidim(conds, (64, 64), "cpu")
+        assert "mask" not in conds[0]
+
+    def test_empty_conditions(self):
+        """Empty conditions list — no crash."""
+        conds = []
+        resolve_areas_and_cond_masks_multidim(conds, (64, 64), "cpu")
+        assert len(conds) == 0
+
+
+# ============================================================
+# Area resolution (percentage-based)
+# ============================================================
+
+class TestAreaResolution:
+    """Test that percentage-based area resolution works for different dims."""
+
+    def test_percentage_area_2d(self):
+        """Percentage area for 2D should resolve to pixel coords."""
+        conds = [{"area": ("percentage", 0.5, 0.5, 0.25, 0.25), "model_conds": {}}]
+        resolve_areas_and_cond_masks_multidim(conds, (64, 64), "cpu")
+        area = conds[0]["area"]
+        assert area == (32, 32, 16, 16)
+
+    def test_percentage_area_1d(self):
+        """Percentage area for 1D should resolve to frame coords."""
+        conds = [{"area": ("percentage", 0.5, 0.25), "model_conds": {}}]
+        resolve_areas_and_cond_masks_multidim(conds, (100,), "cpu")
+        area = conds[0]["area"]
+        assert area == (50, 25)
+
+
+# ============================================================
+# reshape_mask — mask reshaping for F.interpolate
+# ============================================================
+
+class TestReshapeMask1D:
+    """Tests for reshape_mask with 1D output (e.g. audio with noise shape [B, C, L])."""
+
+    def test_4d_input_same_length(self):
+        """[1, 1, 1, L] input (typical from pipeline) — should reshape and expand channels."""
+        mask = torch.ones(1, 1, 1, 100)
+        result = reshape_mask(mask, torch.Size([1, 64, 100]))
+        assert result.shape == (1, 64, 100)
+
+    def test_4d_input_resize(self):
+        """[1, 1, 1, L1] input resized to different length."""
+        mask = torch.ones(1, 1, 1, 50)
+        result = reshape_mask(mask, torch.Size([1, 64, 100]))
+        assert result.shape == (1, 64, 100)
+
+    def test_3d_input(self):
+        """[1, 1, L] input — should work directly."""
+        mask = torch.ones(1, 1, 100)
+        result = reshape_mask(mask, torch.Size([1, 64, 100]))
+        assert result.shape == (1, 64, 100)
+
+    def test_2d_input(self):
+        """[B, L] input — should reshape to [B, 1, L]."""
+        mask = torch.ones(1, 50)
+        result = reshape_mask(mask, torch.Size([1, 64, 100]))
+        assert result.shape == (1, 64, 100)
+
+    def test_1d_input(self):
+        """[L] input — should reshape to [1, 1, L]."""
+        mask = torch.ones(50)
+        result = reshape_mask(mask, torch.Size([1, 64, 100]))
+        assert result.shape == (1, 64, 100)
+
+    def test_channel_repeat(self):
+        """Mask with 1 channel should repeat to match output channels."""
+        mask = torch.full((1, 1, 1, 100), 0.5)
+        result = reshape_mask(mask, torch.Size([1, 32, 100]))
+        assert result.shape == (1, 32, 100)
+        torch.testing.assert_close(result, torch.full_like(result, 0.5))
+
+    def test_batch_repeat(self):
+        """Single-batch mask should repeat to match output batch size."""
+        mask = torch.full((1, 1, 1, 100), 0.7)
+        result = reshape_mask(mask, torch.Size([4, 64, 100]))
+        assert result.shape == (4, 64, 100)
+
+    def test_values_preserved_no_resize(self):
+        """Values should be preserved when no resize is needed."""
+        values = torch.tensor([[[0.0, 0.5, 1.0]]])  # [1, 1, 3]
+        result = reshape_mask(values, torch.Size([1, 1, 3]))
+        torch.testing.assert_close(result, values)
+
+    def test_interpolation_values(self):
+        """Linear interpolation should produce sensible intermediate values."""
+        mask = torch.tensor([[[[0.0, 1.0]]]])  # [1, 1, 1, 2]
+        result = reshape_mask(mask, torch.Size([1, 1, 4]))
+        assert result.shape == (1, 1, 4)
+        # Should interpolate from 0 to 1
+        assert result[0, 0, 0].item() < result[0, 0, -1].item()
+
+
+class TestReshapeMask2D:
+    """Regression tests for reshape_mask with 2D output (image models)."""
+
+    def test_standard_resize(self):
+        """[1, 1, H, W] mask resized to different resolution."""
+        mask = torch.ones(1, 1, 32, 32)
+        result = reshape_mask(mask, torch.Size([1, 4, 64, 64]))
+        assert result.shape == (1, 4, 64, 64)
+
+    def test_same_size(self):
+        """[1, 1, H, W] mask with matching size — no resize needed."""
+        mask = torch.rand(1, 1, 64, 64)
+        result = reshape_mask(mask, torch.Size([1, 4, 64, 64]))
+        assert result.shape == (1, 4, 64, 64)
+
+    def test_3d_input(self):
+        """[B, H, W] input — should reshape to [B, 1, H, W]."""
+        mask = torch.ones(1, 32, 32)
+        result = reshape_mask(mask, torch.Size([1, 4, 64, 64]))
+        assert result.shape == (1, 4, 64, 64)
+
+
+class TestReshapeMask3D:
+    """Regression tests for reshape_mask with 3D output (video models)."""
+
+    def test_standard_resize(self):
+        """[1, 1, T, H, W] mask resized to different resolution."""
+        mask = torch.ones(1, 1, 8, 32, 32)
+        result = reshape_mask(mask, torch.Size([1, 4, 8, 64, 64]))
+        assert result.shape == (1, 4, 8, 64, 64)
+
+    def test_4d_input(self):
+        """[B, T, H, W] input — should reshape to [1, 1, T, H, W]."""
+        mask = torch.ones(1, 8, 32, 32)
+        result = reshape_mask(mask, torch.Size([1, 4, 8, 64, 64]))
+        assert result.shape == (1, 4, 8, 64, 64)