🔍 LoRA Explorer Stack

Stacking LoRAs by chaining three LoraLoader nodes works, but it makes your graph tall and you can't see any previews while you do it. 🔍 LoRA Explorer Stack is the "we fixed that" version from the anima-loras-explorer pack: eight LoRA slots in a single node, each with its own model and CLIP strength, all applied in one box. For Anima specifically this earns its keep - the community consensus is that three LoRAs stack fine without collapsing (something its rival Z-Image can't do), so you'll actually be running character + style + artist at once.

How it works

Every populated slot does the same thing the single LoraExplorerLoader does - comfy.sd.load_lora_for_models - and feeds its output MODEL/CLIP into the next slot. Slots are applied top to bottom, so the second LoRA lands on a model the first one already modified. Order matters, and it's the only order you get: there's no per-slot layer-blend control, just the sequence. The node returns one MODEL and one CLIP at the end, plus previews for every loaded slot so you can see the stack you built.

The inputs

• model / clip - in from your checkpoint, out to conditioning and sampler.
• fetch_civitai - applies to any slot missing a preview, pulling metadata + preview from Civitai by hash.
• Eight slots, each with lora_1..8, strength_model_1..8, and strength_clip_1..8.

The Explorer button (same 🔍 UI as the single loader) fills the first empty slot when you hit "Use LoRA" or "Download & Use" from the search - handy for building a stack as you find things. The only off switch per slot is setting it to None; slots with both strengths at 0 are skipped too, but don't rely on that as a mute button.

Installing it

Same pack, same install: ComfyUI Manager, search "Anima LoRA Explorer", or:

cd ComfyUI/custom_nodes
git clone https://github.com/fulletLab/anima-loras-explorer.git

Restart and it appears as 🔍 LoRA Explorer Stack under loaders/lora explorer. No models to download, one dependency (aiohttp) that ComfyUI already has. It shares the pack's gotchas: remote Civitai search sits behind an opt-in "Internet Access Required" gate, and downloads need a Civitai API key stored in config.json (don't commit it).

The trap to avoid

Every slot defaults both strengths to 1.0. Drop a third LoRA into a stack that already has two at 1.0 and you're not getting "more style" - you're getting bleed. The same Anima advice that applies to the single loader applies harder here: pull style slots down toward 0.6–0.8. And if fetch_civitai is on with several fresh files, first run does one SHA256 hash plus a network round-trip per missing preview - leave it on, let it cache, then turn it off for normal runs.