Donut LoRA CivitAI Info (DEPRECATED)

Every LoRA in your models/loras folder is a mystery file until someone tells you what it does. DonutLoRACivitAIInfo solves that by computing the LoRA's SHA256, asking the CivitAI API what model that hash belongs to, and dumping back the official name, version, description, trigger words, and even the preview image - all without you opening a browser tab. Pick a LoRA from the dropdown, hit run, and read the outputs.

It's DEPRECATED, and unlike some deprecations this one is honest: the node is now a thin alias of the unified DonutLoRACivitAILookup node, pinned to "LoRA File" mode with the same 8-output signature and the same widget order, so saved workflows deserialize byte-identically. Same backend engine, same cache, same preview loading. The newer node just adds a Hash lookup mode. If your graph already uses this one, there's zero reason to migrate; if you're starting fresh, use DonutLoRACivitAILookup.

How it works

The pipeline is: get the LoRA's full path from ComfyUI, compute SHA256 (with a local hash cache so it only costs you the first time), check a local metadata cache, and only if it's not cached hit the CivitAI API by hash. Results get cached locally, so repeat lookups are instant and offline-friendly. The description passes through an HTML-stripper that the source notes was "copied verbatim" from the original nodes so behavior stays identical.

The inputs

• lora_name - the LoRA file selector (populated from your loras folder).
• api_key (optional) - for authenticated requests. You don't need one for basic lookups; it helps with rate limits and metadata that requires auth.
• force_refresh (No/Yes) - bypass the cache and re-fetch from CivitAI even if you've looked this file up before.

The outputs

All eight, ready to wire anywhere strings or images go:

• name, version, description, trigger_words - the human-readable metadata.
• model_url - direct link to the CivitAI page.
• hash - the computed SHA256 (useful for CivitAI's own hash search).
• recommended_weight - a FLOAT with the author's suggested strength. This is the sleeper output; plug it into a LoRA stack and you're starting from the creator's own tuning.
• preview_image - an IMAGE you can display or route into a preview node.

It's an output/display node (is_output_node), so it's the end of a branch, not something feeding forward.

Install

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/DonutsDelivery/ComfyUI-DonutNodes.git donutnodes
cd donutnodes
python -m pip install -r requirements.txt

or ComfyUI Manager → "DonutNodes," restart. First run on a big LoRA folder does a lot of hashing, so give it a moment.

The honest caveat

CivitAI lookup by SHA256 is exactly as good as CivitAI's model registry - which, post-2026-domain-split, is still the definitive LoRA index. If the hash isn't found, the node hands you a search URL so you're not stranded. And recommended_weight is only as good as the uploader's honesty, but it's still a far better starting point than a blind 1.0.