Model Analyzer (LoRA)

You downloaded a LoRA off Civitai, the model card has a sample gallery, and you're squinting at one screenshot trying to reverse-engineer what checkpoint and CFG it was run at. Model Analyzer (LoRA) skips the squinting. It grabs a real batch of community showcase images generated with that LoRA and tallies what people actually use it with - instead of you guessing from a single lucky post.

What it does

This is the "In-Depth Model Analysis" half of Civitai Toolkit, aimed specifically at LoRAs. It hash-matches your local file against Civitai, pulls a sample of community-posted images made with it, and counts patterns across them: which checkpoint it gets paired with most (a lot of LoRAs are trained for one architecture and quietly fall apart on another), what sampler and CFG range shows up, and which other LoRAs tend to ride along with it - the "golden combination" idea the README talks about. It compresses that into one markdown report instead of you scrolling a gallery by hand.

Worth being clear about scope: this node is about generation settings, not trigger words. If you just want the trigger word, the pack has a separate lightweight node - Lora Trigger Words - for exactly that, pulling from both local metadata and the Civitai API. Use this Analyzer node when you want to know how people run the thing, not what magic word activates it.

The inputs and outputs that matter

Two settings actually change what you're sampling:

• sort - Most Reactions, Most Comments, or Newest. Reactions skews toward the LoRA's all-time greatest hits; Newest skews toward what people are posting this week, which drifts if the LoRA's had a recent trend shift.
• image_limit (1–1000, default 100) - how many showcase images to crunch. Bigger sample, steadier picture, more API calls - no reason to max it out for a LoRA that only has a few dozen public posts anyway.

The rest are filters: nsfw_level (None / Soft / Mature / X) and filter_type (all / image / video) narrow the pool, summary_top_n (1–100, default 10) caps how many top entries land in the report, and force_refresh (no/yes) decides whether you're reading the cached fetch or forcing a fresh one.

Three outputs: full_report_md is the full write-up, meant to feed the pack's Markdown Presenter node (under Display) so it actually renders instead of dumping raw markdown into a text box. fetch_summary is the one-line sanity check - something like "Analyzed 100 entries successfully" - so you don't have to open the full report just to confirm it worked. params_pipe (CIVITAI_PARAMS) is the machine-readable version of the same tally, meant for the companion Get Parameters from Analysis node, which unpacks it into individual KSampler-ready values.

Installing it

Through ComfyUI Manager: search Civitai Toolkit, install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/BAIKEMARK/ComfyUI-Civitai-Toolkit
pip install -r requirements.txt

Restart ComfyUI and you'll find it under Civitai → 📊 Analyzer. The pack also ships a Templates → Custom Nodes → ComfyUI-Civitai-Recipe example if you want to see the node pre-wired.

Troubleshooting

The report comes back thin no matter what you set image_limit to. Most LoRAs, especially niche or recent uploads, just don't have a big public showcase gallery yet. That's a small-sample result, not a broken fetch - check the LoRA's Civitai page directly if you want to sanity-check the count.

Nothing comes back, or fetch_summary reads zero. Usually one of two things: the hash didn't resolve to anything on Civitai (a private or heavily repacked file, or a model that's since been pulled - removals happen there fairly often, not as an edge case), or you're getting rate-limited. For the second one, add a Civitai API key under the plugin's settings panel - the pack added that specifically to raise request limits, and it's worth doing before you ever pull image_limit past a couple hundred.

A filter change doesn't seem to do anything. Flip force_refresh to yes. The node caches fetches aggressively for speed, so a stale cache is the first thing to rule out.

Migrating from the old Civitai Recipe Finder. This pack was renamed from that in October 2025 - same author, same install location, much bigger feature set. If you're upgrading from a pre-4.0 install, go to Settings → CivitaiUtils → Migration to pull your old cached JSON into the new database, or the Analyzer will look like it's starting from scratch.