fix(image): grayscale image support for save image advanced

[yousef-rafat]

Adds a channel in case of 2d channel grayscales.
Changes the grayscale format depending on dtype.
Before and after screenshots:

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 4ad507de-daf5-4a04-96c2-af4fcc16071d

📥 Commits

Reviewing files that changed from the base of the PR and between 537e303 and 54f340c.

📒 Files selected for processing (1)

• comfy_extras/nodes_images.py

🚧 Files skipped from review as they are similar to previous changes (1)

• comfy_extras/nodes_images.py

---

📝 Walkthrough

Walkthrough

The change modifies _encode_image in comfy_extras/nodes_images.py to handle grayscale inputs by adding a channel dimension for 2D tensors and deriving frame_fmt and stream_fmt from the numpy dtype. It then uses those derived formats when setting the codec pixel format, creating the VideoFrame, and conditionally reformatting the frame before encoding.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly matches the main change: adding grayscale image support for Save Image (Advanced).
Description check	✅ Passed	The description accurately describes the grayscale channel handling and dtype-based format changes in this PR.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

comfy_extras/nodes_images.py (1)

1105-1106: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Misleading comment: this checks tensor rank, not channel count.

img_tensor.ndim == 2 means the tensor has no channel axis at all (a true 2D (H, W) grayscale), not "2 channels." As written, the comment could easily be misread as referring to a 2-channel (e.g., luma+alpha) tensor.

✏️ Suggested comment fix

-    if img_tensor.ndim == 2: # 2 channel grayscales
+    if img_tensor.ndim == 2:  # (H, W) grayscale with no channel axis
         img_tensor = img_tensor.unsqueeze(-1)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@comfy_extras/nodes_images.py` around lines 1105 - 1106, The inline comment in
the image tensor handling logic is misleading because the `img_tensor.ndim == 2`
check in `nodes_images.py` is about tensor rank, not channel count. Update the
comment near the `img_tensor.ndim` branch to clearly state that this case
handles a 2D grayscale tensor with shape like `(H, W)` and that `unsqueeze(-1)`
adds the missing channel axis; keep the wording aligned with the `img_tensor`
condition so it cannot be read as referring to 2-channel input.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@comfy_extras/nodes_images.py`:
- Around line 1105-1125: The grayscale handling in the image export path still
passes a 3D array with a singleton channel into av.VideoFrame.from_ndarray,
which PyAV grayscale formats do not accept. Update the code in the grayscale
branch around img_tensor/img_np processing so that when num_channels == 1 you
remove the trailing channel dimension before creating the frame, or avoid adding
it earlier; keep the existing format selection logic in _FORMAT_SPECS and the
frame creation flow aligned with a 2D (H, W) grayscale input.

---

Nitpick comments:
In `@comfy_extras/nodes_images.py`:
- Around line 1105-1106: The inline comment in the image tensor handling logic
is misleading because the `img_tensor.ndim == 2` check in `nodes_images.py` is
about tensor rank, not channel count. Update the comment near the
`img_tensor.ndim` branch to clearly state that this case handles a 2D grayscale
tensor with shape like `(H, W)` and that `unsqueeze(-1)` adds the missing
channel axis; keep the wording aligned with the `img_tensor` condition so it
cannot be read as referring to 2-channel input.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: abdb4b42-07bb-4814-b547-cb4037a08a3d

📥 Commits

Reviewing files that changed from the base of the PR and between 77917ed and 537e303.

📒 Files selected for processing (1)

• comfy_extras/nodes_images.py