Recipe Gallery
Pull a model's best Civitai recipes straight into ComfyUI
- image
- info_md
- recipe_params
You've downloaded a checkpoint or LoRA, dropped it in, and gotten mush out. Not because the model's bad - because you're guessing at the prompt, the sampler, the CFG, the LoRA weight, all of it. CivitaiRecipeGallery skips the guessing: point it at a model already sitting in your ComfyUI folders and it pulls up that model's top community images from Civitai, with their full generation recipe attached, right inside the graph. No browser tab, no manually dragging a downloaded PNG onto the canvas to steal its metadata.
That workaround it's replacing is one of ComfyUI's real cultural quirks: a Civitai sample image usually carries its full workflow embedded in the file - prompt, sampler, LoRA stack, all of it - the same PNG-metadata trick that makes "workflow included" a thing people say in this community. Normally getting at that means downloading the image and dragging it onto the canvas yourself. This node automates exactly that step, in the background, so you never have to.
How it works
Feed it which local model you want recipes for - a checkpoint, a LoRA, whatever - and it does the matching by hash rather than filename, the same way Civitai's own site links an uploaded file to its model page. From there it grabs that model's community gallery, filtered and sorted the way you asked, and unpacks the winning image's embedded generation data into something the rest of your graph can use.
The first run is slow - it has to hash your local model files, which the README calls out explicitly, and an earlier version of the pack used to hang ComfyUI's whole startup over a big LoRA folder before that got fixed. After the first pass it's cached, so subsequent loads are fast.
The inputs and outputs that matter
You're setting six things, and most of them are just filters on the gallery:
model_type- which bucket you're browsing: checkpoints, loras, vae, embeddings, diffusion_models, text_encoders, or hypernetworks.model_name- the actual local file. This list only populates from models ComfyUI already knows about in the matching folder, so if it's empty, that's the tell - the model isn't where this model_type expects it.sort- Most Reactions, Most Comments, or Newest.nsfw_level- None, Soft, Mature, or X, matching Civitai's own content-rating tiers.image_limit- 1 to 100, defaults to 32.filter_type- all, image, or video (defaults to image).
Three outputs come back out: image (the selected sample, IMAGE), info_md (a Markdown write-up of the recipe - the README specifically recommends piping this into the pack's own MarkdownPresenter), and recipe_params (a RECIPE_PARAMS bundle you don't use directly - it's built for the pack's companion node, Get Parameters from Recipe, which unpacks it into ckpt_name, positive_prompt, negative_prompt, seed, steps, cfg, sampler_name, scheduler, width, height, and denoise - everything a KSampler wants.
How to install it
Search Civitai Toolkit in ComfyUI Manager and install - that's the pack's current name, even though the class and repo still say "Civitai Recipe." Real gotcha: the pack was renamed from Civitai Recipe Finder to Civitai Toolkit in October 2025, and right after the rename the author was warning users on Reddit that the new name might not show up in Manager search yet. If a manager search comes up empty, go manual:
cd ComfyUI/custom_nodes
git clone https://github.com/BAIKEMARK/ComfyUI-Civitai-Recipe
pip install -r requirements.txt
Restart ComfyUI. You'll find the node under the Civitai category.
Common issues & troubleshooting
model_name is empty. The dropdown only lists models ComfyUI has actually indexed under that model_type. Wrong folder, or a model_type that doesn't match where the file lives, and you get nothing to pick from.
First load is slow, or ComfyUI seems to hang on startup with a big LoRA collection. That was a real bug (fixed in the 4.0.1 update) - hashing used to block startup. If you're on an old version, update; the fix moved model hashing to the background with resume support so it doesn't restart from zero.
Gallery fetches fail or you keep hitting request limits. Civitai's API rate-limits anonymous requests. The pack added its own API Key Management panel in ComfyUI's settings specifically for this - generate a key from your Civitai account page and paste it in, and the node's requests carry proper auth headers from then on.
Coming from the pre-rename version with an old JSON cache. Go to Settings → CivitaiUtils → Migration to move it into the new database format rather than starting your hash cache over.
Nothing NSFW is showing up. Check nsfw_level first - it's a hard filter, not a suggestion, and Civitai's own rating tiers behind it have been genuinely unstable over the past year, so don't assume a thin gallery means the model has no community images.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model_type | COMBO | 7 options: checkpoints, loras, vae, embeddings, diffusion_models, text_encoders, +1 | |
| model_name | COMBO | 0 options: | |
| sort | COMBO | 3 options: Most Reactions, Most Comments, Newest | |
| nsfw_level | COMBO | 4 options: None, Soft, Mature, X | |
| image_limit | INT | 321–100 | — |
| filter_type | COMBO | image | 3 options: all, image, video |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| info_md | STRING | — |
| recipe_params | RECIPE_PARAMS | — |