Apex LoRA Loader

Apex LoRA Loader is a standard LoRA loader with one genuinely nice upgrade: a visual browser. Instead of squinting at a dropdown full of filenames, you click a button and get a full-screen modal with folder navigation and thumbnail previews of every LoRA you own. If you have a few hundred LoRAs in models/loras - and by mid-2026, who doesn't - being able to see the style at a glance beats reading names every time.

Under the hood it's honest about what it is: it calls ComfyUI's own load_lora_for_models with a single strength applied to both the model and the CLIP if you connect one. The clever part is the frontend. A companion HTTP endpoint generates 256px JPEG thumbnails from your LoRA preview images - it grabs the first frame of animated previews, handles PNG/JPG/WebP, and caches everything in a .thumbnails folder at the root of your loras directory. That's the 100–200× load-time improvement the README brags about: your raw preview files can be multi-megabyte; the cached thumbnails are 10–30 KB. It also picks up previews automatically by filename (.preview.png, .preview.jpg, or just the LoRA name itself), which matters because CivitAI downloads frequently ship with confusingly-named previews.

The inputs that matter

• model (MODEL) - your checkpoint, straight from the loader.
• lora_name - the dropdown. It's populated from your models/loras folder at startup, and the node's validation lets the browser hand it any valid path inside that folder.
• strength (−10 to 10, default 1) - one slider for both model and CLIP.

clip is optional - connect it if you want the LoRA's text-side weights applied too (you generally do, unless the LoRA is a "model-only" patch). Outputs are model and clip, which feed the sampler exactly like any other LoRA loader.

Where it sits in the ecosystem

Here's the honest take: this is a nice-to-have, not a must-have. The community's actual workhorse is rgthree's Power Lora Loader, which stacks multiple LoRAs per entry, reads trigger words off files, and is in hundreds of shared workflows - if you grab a workflow from the wild, that's what it'll ask for. Apex's pitch is different: a single-strength loader with a real visual browser, aimed at people with big collections who'd rather browse than type. It's a fine second loader to keep around for discovery. Don't replace your main one unless the thumbnails genuinely save you time.

Installing it

Part of the Apex Artist pack - install once. ComfyUI Manager → search "Apex Artist" → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ApexArtist/comfyui-apex-artist.git

Restart ComfyUI; the node is under Add Node → Apex Artist → Models. No model downloads, no extra pip deps - the thumbnail server uses Pillow, which ComfyUI already ships. Small indie MIT pack with near-zero community footprint, but the source is short and readable.

Gotchas

If a LoRA shows no thumbnail, it has no preview file next to it in models/loras - drop a .preview.png beside it and it'll appear on next refresh. Keep your LoRAs in the standard ComfyUI/models/loras folder, because the browser only navigates from there. One genuine quirk to remember: if strength is 0, the node short-circuits and passes your model straight through untouched - handy as a bypass toggle, mildly confusing the first time you see it. And per the KB's LoRA notes, most LoRAs want a strength around 0.5–0.8, not 1.0, so treat the default as a starting point rather than a target.