fix(recipe): use resources type field to identify checkpoint instead of modelVersionIds[0]

When importing a CivitAI image as a recipe, modelVersionIds[0] was blindly used as the checkpoint version ID. This array mixes checkpoints and LoRAs without ordering guarantees, causing LoRAs to be saved as the recipe checkpoint.

Fix by:
1. Removing the modelVersionIds[0] fallback in _download_remote_media
2. Parsing resources entries with type:"model" as the checkpoint
3. Adding model type validation in populate_checkpoint_from_civitai

Also add 2 tests for the new behavior and fix 3 tests whose mocks lacked the required model.type field.
This commit is contained in:
Will Miao
2026-05-28 15:46:33 +08:00
parent 3f6824eef6
commit 34791c2ad7
6 changed files with 203 additions and 10 deletions

View File

@@ -1293,11 +1293,18 @@ class RecipeManagementHandler:
image_info.get("meta") if civitai_image_id and image_info else None
)
if civitai_image_id and image_info:
# modelVersionId (singular) — the primary version for this
# image on CivitAI. May be absent, or may *not* be the
# checkpoint (e.g. when the image was generated with a LoRA
# as the primary subject). When absent, DO NOT fall back to
# modelVersionIds[0] — that array mixes checkpoints, LoRAs,
# and other model version IDs without ordering guarantees.
# The downstream enrichment flow will find the real
# checkpoint via meta.resources (type:"model" hash) or
# meta.civitaiResources (type:"checkpoint" version ID), so
# leaving model_ver_id as None is safe and avoids the bug
# where a LoRA version ID was treated as the checkpoint.
model_ver_id = image_info.get("modelVersionId")
if not model_ver_id:
ids = image_info.get("modelVersionIds")
if isinstance(ids, list) and ids:
model_ver_id = ids[0]
# Inject root-level modelVersionIds into meta so downstream
# parsers (CivitaiApiMetadataParser) can discover ALL resources