feat(recipes): preserve embedded ComfyUI workflow on remote imports

CivitAI serves a re-encoded, metadata-free optimized rendition as the recipe
preview, so the ComfyUI workflow embedded in the original image was dropped:
imported recipes reported has_workflow=false and never offered "Send Workflow
to ComfyUI" even when the source image carried one.

Recover the workflow from the original rendition and carry it to the save step
as data, so the stored preview stays the small optimized image:

- ExifUtils: embed a caller-supplied workflow during optimize_image's single
  encode pass, and add embed_workflow() to patch WebP EXIF in place (used by
  the verbatim skip_optimize branch and as a safety net).
- RecipePersistenceService.save_recipe: embed metadata["workflow"] before
  detecting has_workflow.
- analyze_remote_image: return the workflow recovered from the original
  rendition it already downloads for EXIF parsing.
- RecipeManagementHandler: add _fetch_original_media() and workflow helpers;
  _do_import_from_url reuses them, and _do_import_remote_recipe fetches the
  original only when CivitAI reports a ComfyUI payload (meta.comfy) so
  workflow-less images pay no extra bandwidth.
- Batch URL imports and the import modal forward the recovered workflow.

Verified against the reported image: has_workflow flips from false to true and
the recovered workflow matches the original (25 nodes, same graph id).
This commit is contained in:
Will Miao
2026-09-29 07:20:12 +08:00
parent 0dd8d74032
commit 69691b17a1
11 changed files with 1059 additions and 64 deletions
+38
View File
@@ -117,6 +117,10 @@ class RecipeAnalysisService:
image_info: Optional[dict[str, Any]] = None
is_video = False
extension = ".jpg" # Default
# Workflow recovered from the image. CivitAI's optimized renditions are
# re-encoded and carry no metadata, so for those the workflow only
# exists in the original rendition, fetched below for EXIF extraction.
recovered_workflow: Optional[str] = None
# Diagnostics collected during analysis; surfaced in the payload so
# callers can persist an import_info block explaining empty LoRA lists.
diagnostics: dict[str, Any] = {"channel": "url"}
@@ -238,6 +242,9 @@ class RecipeAnalysisService:
exif_metadata = await asyncio.to_thread(
self._exif_utils.extract_image_metadata, temp_path
)
recovered_workflow = await asyncio.to_thread(
self._read_embedded_workflow, temp_path
)
# Fallback: try the original (non-optimized) image for EXIF data
if not exif_metadata and civitai_image_id and image_info:
@@ -255,6 +262,16 @@ class RecipeAnalysisService:
self._exif_utils.extract_image_metadata,
orig_temp_path,
)
# The original is also the only place a ComfyUI
# workflow survives; carry it so the save step can
# embed it even though the stored preview stays the
# small, metadata-free optimized rendition.
recovered_workflow = (
await asyncio.to_thread(
self._read_embedded_workflow, orig_temp_path
)
or recovered_workflow
)
finally:
self._safe_cleanup(orig_temp_path)
@@ -358,6 +375,8 @@ class RecipeAnalysisService:
diagnostics["is_video"] = is_video
result.payload["diagnostics"] = diagnostics
if recovered_workflow:
result.payload["workflow"] = recovered_workflow
return result
finally:
if temp_path:
@@ -545,6 +564,25 @@ class RecipeAnalysisService:
if not success:
raise RecipeDownloadError(f"Failed to download image from URL: {result}")
def _read_embedded_workflow(self, image_path: Optional[str]) -> Optional[str]:
"""Return a ComfyUI workflow embedded in ``image_path``, if any.
The raw metadata string extractor stops at the generation parameters
(``prompt``/``parameters``), so the UI-format workflow has to be read
through the structured metadata reader. Failures map to ``None``.
"""
if not image_path or not os.path.exists(image_path):
return None
try:
metadata = self._exif_utils._load_structured_metadata(image_path)
except Exception as exc:
self._logger.debug(
"Failed to read embedded workflow from %s: %s", image_path, exc
)
return None
workflow = metadata.get("workflow") if isinstance(metadata, dict) else None
return workflow if isinstance(workflow, str) and workflow else None
def _metadata_not_found_response(self, path: str) -> AnalysisResult:
payload: dict[str, Any] = {
"error": "No metadata found in this image",