[Bug] Image generation metadata not embedded in output | Preview pane shows low-res image | Preview pane buttons not functional

[cuter-doc0]

Environment

• RookieUI version: 0.8.0
• ComfyUI version: 0.22.0
• Backend: ComfyUI (standalone Docker container)

---

Issue 1: Generation metadata not saved to images

Images generated through RookieUI do not appear to contain embedded generation metadata (prompt, steps, sampler, CFG, seed, etc.). This makes it impossible to review or reproduce generation parameters after the fact by inspecting the image file.

In A1111, metadata is written directly into the PNG as EXIF/tEXt chunks and is readable by tools like image viewers or by dragging the image back into the UI. This behaviour is expected here but is not occurring.

Steps to reproduce:

1. Generate an image through RookieUI - file saved as expexted, eg RookieUI_00123_.png
2. Inspect the output PNG metadata (e.g. exiftool output.png or drag back into ComfyUI)
3. No generation parameters are present

Expected: PNG contains embedded prompt and generation parameters
Actual: PNG contains no metadata

---

Issue 2: Preview pane renders low-resolution image instead of full output

After generation completes, the preview pane sometimes displays a low-resolution version of the image rather than the full resolution output. The full image appears to exist on disk but is not being displayed correctly in the preview.

---

Issue 3: Preview pane buttons not functioning as expected

Buttons beneath the preview pane do not behave as expected compared to A1111. Specific expected behaviours that are not working:

• [describe specific buttons if you want to add detail, e.g. Save, Send to img2img, etc.]

---

Additional context

Coming from an A1111 workflow. These three issues together make post-generation review and iteration significantly harder than expected. Happy to provide logs or further details if helpful.

[cuter-doc0]

regarding the metadata; I asked openCode to clone this repo and said this is an extension/node for comfyui to minic the ui of a1111 - currently images generate fine, but they don't have any metadata. can you try and figure out why that might be?

Response was:

Here's what was wrong and what I changed:

Root Cause

The main txt2img/img2img generation path uses ComfyUI's built-in SaveImage node, which only writes ComfyUI-specific metadata (prompt and workflow JSON chunks). It does not write the A1111-style parameters text chunk (which stores generation infotext like steps, seed, CFG, prompt, etc.). RookieUI already had the infrastructure to parse A1111 infotext (in pnginfo.py) and to write PNG metadata chunks (in asset_store.py::save_output_image), but save_output_image was only used for XYZ Plot and Extras, not the main generation pipeline.

Changes Made

8 files modified:

1. rookieui/services/infotext.py (new) — Builds A1111-style infotext strings from NormalizedTxt2ImgRequest / NormalizedImg2ImgRequest objects. Format: <prompt>\nNegative prompt: <neg>\nSteps: X, Sampler: Y, CFG scale: Z, Seed: N, Size: WxH, Model: M, ...

2. rookieui/nodes.py — Added RookieUISaveImage custom node that saves images to {ComfyUI_output}/RookieUI/ and embeds a parameters PNG text chunk containing the infotext. Registered in NODE_CLASS_MAPPINGS.

3. rookieui/services/workflow_builders/output.py — Changed _append_save_node to use RookieUISaveImage instead of SaveImage, with a parameters input. Added infotext and request params to _build_decode_and_save.

4. rookieui/services/workflow_builders/adetailer.py — _append_decode_adetailer_and_save now builds infotext from the request and passes it to _append_save_node.

5. rookieui/services/workflow_builders/non_sd_templates.py — All 15 calls to _build_decode_and_save now pass request=request.

6. tests/test_txt2img_translation.py — Updated to expect RookieUISaveImage class type.

7. tests/test_img2img_translation.py — Same.

8. tests/test_adetailer_chain.py — Same.

I replaced the changed files in my comfyui and images generated now do appear to have a1111 metadata - validation via the rookieui PNG info tool validates the that too.

after validation, I asked for a write-up.

Here's a writeup on the issue:

---

Problem: Generated images from RookieUI's txt2img/img2img pipeline don't have A1111-style generation metadata (the parameters PNG text chunk). Without it, dragging images back into PNG Info or other tools that parse A1111 infotext yields nothing.

Root cause: The main generation path builds workflows terminating in ComfyUI's stock SaveImage node, which only writes ComfyUI-specific chunks (prompt JSON, workflow JSON). RookieUI already has the machinery to write PNG metadata (asset_store.py::save_output_image uses PngImagePlugin.PngInfo()) and to parse A1111 infotext (pnginfo.py), but save_output_image is only called for XYZ Plot grids and Extras, never for the primary generation flow.

Fix: Replace SaveImage with a custom RookieUISaveImage node that embeds an A1111-style parameters text chunk, and construct the infotext string at workflow-build time from the normalized request parameters.

Changes:

1. rookieui/services/infotext.py (new) — build_infotext(request) constructs an A1111-compatible infotext string from either NormalizedTxt2ImgRequest or NormalizedImg2ImgRequest. Format:

   positive prompt
   Negative prompt: negative prompt
   Steps: 20, Sampler: Euler, CFG scale: 7.0, Seed: 42, Size: 512x512, Model: model.safetensors, ...
   

2. rookieui/nodes.py — Added RookieUISaveImage node:

   • Inputs: images (IMAGE), filename_prefix (STRING), parameters (STRING)
   • Saves to {ComfyUI_output}/RookieUI/ with the same naming convention as the original SaveImage
   • Writes the parameters string as a parameters PNG text chunk via PngImagePlugin.PngInfo().add_text("parameters", ...)
   • Returns {"ui": {"images": [...]}} matching the expected ComfyUI output format for queue/history consumption
   • Category: RookieUI/image, registered in NODE_CLASS_MAPPINGS

3. rookieui/services/workflow_builders/output.py:

   • _append_save_node() now emits "class_type": "RookieUISaveImage" with a "parameters" input holding the infotext
   • _build_decode_and_save() accepts optional request and infotext kwargs; if request is provided, it builds the infotext automatically

4. rookieui/services/workflow_builders/adetailer.py — _append_decode_adetailer_and_save() now calls build_infotext(request) and passes the result to _append_save_node()

5. rookieui/services/workflow_builders/non_sd_templates.py — All 15 call sites of _build_decode_and_save pass request=request

Tests updated: test_txt2img_translation.py, test_img2img_translation.py, test_adetailer_chain.py — changed "SaveImage" expectations to "RookieUISaveImage".

[rookiestar28]

regarding the metadata; I asked openCode to clone this repo and said this is an extension/node for comfyui to minic the ui of a1111 - currently images generate fine, but they don't have any metadata. can you try and figure out why that might be?

Response was:

Here's what was wrong and what I changed:

Hi @cuter-doc0,

Thanks for the detailed issue report and for contributing PR #5. The reproduction notes were very helpful.

For Issue 1, the root cause was that RookieUI correctly submits native ComfyUI workflows, but its custom backend submission path also needs to provide the metadata contract that ComfyUI normally receives through the frontend queue path. ComfyUI workflow metadata and A1111 parameters infotext are different formats, so generated PNGs need an explicit raw A1111-style parameters chunk for PNG Info and external metadata tools.

One important distinction: the fix should not replace the save path with a simplified standalone saver. The current dev implementation keeps ComfyUI-compatible save behavior, including filename/path handling, hidden prompt / extra_pnginfo inputs, metadata-disable behavior, and future compatibility with ComfyUI save semantics, while adding the missing raw parameters metadata contract.

Because of that, PR #5 should not be merged as-is. Its direction is useful, but the implementation has been superseded by the current dev metadata path. We did not merge PR #5 directly. Instead, we implemented the fix on dev using the existing RookieUI architecture, then carried over the useful intent of the PR: generated PNGs now include raw A1111-style parameters metadata, but through the current metadata/save pipeline rather than the standalone RookieUISaveImage implementation.

Issue 2 was investigated separately and appears to be related to hires-fix redraw/denoise behavior rather than the preview pane selecting a lower-resolution output.

Issue 3 has a separate root cause and is not fixed by metadata embedding alone. Preview buttons need A1111-style output action behavior: they must use the selected final output artifact, not a stale live preview image, and transfer image + generation parameters into the target pane. Img2Img/Inpaint should receive prompt, negative prompt, steps, sampler, scheduler, CFG, seed, size, model, VAE, and clip skip; Inpaint also switches mode to inpaint. Extras keeps selected-image handoff semantics, while PNG Info remains an inspect/import action. Tests should verify actual destination field values for both toolbar buttons and dropdown actions.

Thanks again for the contribution. Further issues and PRs are very welcome :)