comfyui-lora-tag-hash-metadata

ComfyUI LoRA Tag Hash Metadata

ComfyUI custom nodes that credit your generation resources on CivitAI by embedding Name:AUTOV2[:Weight] hash metadata into saved images.

• Civitai Resources To Hash Metadata (v2, primary) — credits local lora tags AND any CivitAI resource (workflows, text encoders, detailers, …) from pasted URLs, hashes, or AIR URNs, with an in-node status list.
• LoRA Tags To Hash Metadata (v1) — converts <lora:name:weight> text into Name:HASH:Weight strings. Still supported.

Created with comfyui-custom-node-template

Install

ComfyUI Manager

1. Open Manager in ComfyUI
2. Click Install Custom Nodes
3. Search for LoRA Tag Hash Metadata or comfyui-lora-tag-hash-metadata
4. Click Install
5. Restart ComfyUI if Manager prompts you to do so

If you use the ComfyUI CLI instead of the Manager UI:

comfy node install comfyui-lora-tag-hash-metadata

Civitai Resources To Hash Metadata (v2)

• Node id: CivitaiResourcesToHashMetadata
• Display name: Civitai Resources To Hash Metadata
• Category: utils/metadata

CivitAI auto-links image resources purely by file hash — including resource types it can't detect from prompts, like whole workflows, text encoders ("Other"), and selectively-applied detailer LoRAs. This node resolves each resource you list to its AutoV2 hash via the CivitAI API and emits it into additional_hashes so uploads credit the creators automatically.

Typical wiring:

Lora Loader (LoraManager).loaded_loras
  -> Civitai Resources To Hash Metadata.loaded_loras
  -> Image Saver Metadata.additional_hashes

Example civitai_resources input:

# workflows, text encoders, detailers — anything on civitai
https://civitai.red/models/1362968/workflow-for-anima-and-sdxl-noobai-xlillustrious-xl
https://civitai.red/models/2598886/anima-text-encoder-qwen3-06b-heretic-abliterated-uncensored
https://civitai.com/models/2767064/anima-detailer?modelVersionId=3114726 0.8
CD64AF8696
urn:air:anima:lora:civitai:2767064@3114726

＋ Add Resource picker

The node's ＋ Add Resource button opens a searchable picker so you never have to hand-copy URLs:

• CivitAI search tab: live search (all content types, no filtering), type chips, sort, load-more; picking a result appends a version-pinned URL line to the textbox. Lora-ish results take an optional weight.
• Local · LoRA Manager tab (shown only when LoRA Manager is installed): pick from every model type LoRA Manager indexes — loras, checkpoints, and embeddings, with kind chips to narrow. Matched files insert their civitai link, unmatched files insert their AutoV2 hash. Weight inputs appear for loras and embeddings only; checkpoints take none. Note: LoRA Manager hashes checkpoints lazily — an unmatched checkpoint it hasn't hashed yet shows a disabled "no hash" button until you hash/match it in LoRA Manager.
• Entries already in the textbox show Remove instead of Add, so the picker can also clean up. The textbox stays fully hand-editable either way.
• Result names link to their civitai page; lora-ish results take an optional weight (the small wt box).
• When a query returns nothing: civitai's public API hides some models (e.g. flagged ones) — the picker links the same search on the civitai site. Pasting such a model's version-pinned URL into the textbox still credits it.

The in-node status list updates live: picker applies, the per-row ✕ (removes that line from the textbox), and the ⟳ Refresh resources button (under ＋ Add Resource) all re-parse civitai_resources through the node's own resolver + cache — no queueing needed. loaded_loras is included live: the preview reads the linked loader's widget state (LoRA Manager's loras panel — honoring per-lora enable toggles, exactly like a run — or any node exposing <lora:…> tag text), falling back to the last run's rows for producers it can't read. Preview renders carry a "queue a prompt to see final resource list" note. Loading a saved workflow re-renders the list the same way. Hover a row (or its status dot) for what the color means — resolved / unverified hash / duplicate / not credited.

Inputs:

• loaded_loras (optional link): <lora:name:weight> text, e.g. from LoRA Manager — handled exactly like v1
• civitai_resources (multiline textbox), one resource per line:
  • model URLs on any civitai domain (.com/.red/.green), optionally pinned with ?modelVersionId=… — unpinned URLs resolve to the model's latest version
  • AutoV2 hashes (10 hex) or full SHA256 hashes (64 hex)
  • AIR URNs like urn:air:anima:lora:civitai:2767064@3114726 (+fileId honored)
  • optional trailing weight: <line> 0.8
  • # comment lines and blanks are ignored
  • bare numeric IDs are rejected (ambiguous)

Outputs:

• additional_hashes: comma-separated Name:AUTOV2[:Weight] entries (weight only when explicit; lora tags keep their weights)
• resolved / missing: comma-separated names for quick display
• resources_json: structured JSON of every entry (status, type, version, hash, links) for downstream tooling

Behavior:

• in-node status list after each run: card rows with a preview thumbnail (civitai preview for URL/hash/AIR entries, local sidecar preview via ComfyUI's stock model-preview route for loras), name linked to civitai, type · version subline, and a status accent (green resolved / amber duplicate / red failed with reason)
• the list is capped at ~320px and scrolls inside the node — it never overflows the node bounds
• executes standalone (output node) — no downstream saver needed to check your list
• persistent on-disk cache under ComfyUI's user directory: version- and hash-pinned lookups cache forever, unpinned model URLs refresh after 24h; cached entries keep working offline
• failures never abort the queue — bad lines land in missing and the status list
• names are sanitized for Image Saver (,/: stripped, never literally vae), and all-decimal hashes get an explicit :1.0 so the hash can't be misparsed as a weight

Image Saver footgun: Name:HASH:Weight (3-part) entries are only parsed when Image Saver Metadata's download_civitai_data is True. With it False, 3-part entries are silently dropped — this affects all v1 output (always weighted) and weighted v2 entries. Keep download_civitai_data=True (the default).

LoRA Tags To Hash Metadata (v1)

• Node id: LoraTagsToHashMetadata
• Display name: LoRA Tags To Hash Metadata
• Category: utils/metadata

Inputs:

• loaded_loras: multiline string containing one or more <lora:name:weight> tags; LoRA Manager's dual-strength form <lora:name:model:clip> is accepted too (the model strength is the credited weight)

Outputs:

• additional_hashes: comma-separated Name:HASH:Weight string
• resolved_loras: comma-separated list of successfully resolved LoRA names
• missing_loras: comma-separated list of unresolved or incompatible names

Behavior:

• supports <lora:name> and defaults weight to 1.0
• resolves path-qualified names before basename fallback
• treats dotted version suffixes such as my_lora_v0.23 as part of the LoRA name unless a known model extension is present
• computes SHA256 locally and emits the first 10 uppercase hex characters
• deduplicates repeated tags with last occurrence winning
• reports comma-bearing names in missing_loras instead of silently dropping them

Local Development

1. Clone/symlink this folder into ComfyUI's custom_nodes/ directory
2. Run pnpm install
3. Run pnpm dev to watch for changes and rebuild dist/

pnpm install    # Install dependencies
pnpm dev        # Watch mode - rebuilds dist/ on change
pnpm build      # Build for production
pnpm test       # EVERYTHING: vitest + pytest + full e2e (live civitai API)

Note: Reload ComfyUI frontend (browser refresh) for JS changes. Restart ComfyUI server for Python changes.

Testing

One suite, no tiers: pnpm test runs frontend unit tests, backend Python tests, and a full end-to-end run against a repo-local pinned ComfyUI (with LoRA Manager + Image Saver + real fixture loras + the live civitai API) — every time, locally and in CI. See docs/TESTING.md for the harness details, manual serve mode (pnpm e2e:serve), and user stories.

uv sync                 # python deps
pnpm test:unit          # fast lanes only
pnpm test:e2e           # provision .e2e/ + playwright

Publishing to ComfyUI Registry

1. Ensure all fields in pyproject.toml are correct
2. Add REGISTRY_ACCESS_TOKEN secret to your repo (from ComfyUI registry)
3. Go to Actions → "Publish to Comfy registry" → Run workflow
4. Select version bump type (patch/minor/major)

License

MIT