Civitai Images Gallery

There's a specific remix ritual most ComfyUI people know: you see a Civitai image you like, you want its exact prompt, so you open the page, copy the prompt, paste it in, squint at the seed, and quietly give up on the sampler settings. The Civitai Images Gallery automates all of it. Click a card in the graph, hit queue, and out come the positive prompt, negative prompt, the original full-res image, and every generation parameter behind it.

It's the original node in the Firetheft Civitai Gallery pack - the models gallery came later. What you get is a Civitai image browser embedded in your workspace: a masonry waterfall with infinite scroll, filters by NSFW level, sort order (Most Reactions, Newest...), time period, username, and tags, plus a lightbox with zoom/pan and a favorites system that supports custom tags. The pack brands itself SFW-friendly, if that matters to you.

How it works

No inputs, which is the giveaway that this is a UI-first node. All the browsing is frontend - the JS pulls from Civitai's public /api/v1/images endpoint through the pack's own server routes, so browsing needs no API key. When you queue, the selected card's data becomes the outputs. Two details are worth understanding before you wire it up:

• The image downloads lazily. The full-res original isn't fetched when you click the card. When the workflow runs, the node downloads it - but only if the image output is actually connected. The frontend detects the wire and flips a flag; leave the output unconnected and you save the bandwidth. This is the pack's best idea and it works.
• Everything else is JSON wrangling. Prompts come straight from the image's metadata, and info is the raw record, pretty-printed.

The outputs that matter

• positive_prompt / negative_prompt → the two text outputs. Wire them to your KSampler's positive/negative, or into a Show Text node to eyeball them first.
• image → the full-res IMAGE tensor. Wire to Preview Image or Save Image. Remember: only downloaded when connected.
• info → a STRING of JSON with everything else - sampler, steps, seed, CFG, model hashes. Show Text it when you're reverse-engineering someone's settings.

The features that make it worth installing

Three things push this past "fancy prompt copier." First, one-click workflow loading: cards that carry ComfyUI workflow metadata show a 🎁 icon, and clicking it drops the entire workflow into your workspace like dragging in a workflow file. That's the "workflow included" culture from the ecosystem essays, made clickable. Second, video browsing - tick "Show Video" and cards get a ▶️ badge; hover to preview, click to play at original resolution, with autoplay deliberately disabled so ComfyUI doesn't crash. Third, the Edit Prompt checkbox, which lets you rewrite and save prompts, including completing images whose metadata is missing or broken. The models gallery even links into this node - its "View Images" button points the image gallery at one specific model version.

Install

Same pack as the models gallery - one install, both nodes. In ComfyUI Manager, search "Civitai Gallery". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Firetheft/ComfyUI_Civitai_Gallery

Restart and you're done. No pip requirements; it runs on aiohttp, Pillow, numpy, and torch, all already inside ComfyUI. No API key needed for anything this node does.

Common issues

• Empty gallery. The default backend is civitai.work, a mirror aimed at users who can't reach the main site. If you're outside that situation, check the "International" box to route through civitai.com, then hit Refresh. A blank or half-loading gallery is almost always this.
• Black or 1×1 image out. Either the image output isn't connected (then nothing downloads - expected) or the download failed. If the latter, watch the ComfyUI console for "Failed to download or process image."
• Node missing after install. The zip-nesting trap: it must be custom_nodes/ComfyUI_Civitai_Gallery/ with no extra wrapping folder, followed by a full restart.
• Video crash on autoplay. The pack avoids this by design - use the play icon rather than letting cards autoplay.

For the remix workflow specifically - find image, load its workflow, swap the model, re-queue - this is the fastest path I know in ComfyUI. It's fiddly around the mirror-domain edge cases, but the core loop is genuinely good.