Load Anima XPred LoRA

Load Anima XPred LoRA is the quiet middle child of the leafmoone/ComfyUI-RUM-Anima-XPred pack: model in, model out, one slider. It takes the RUM_ANIMA_XPRED bundle from Load Anima XPred Model, merges in a LoRA from ComfyUI's models/loras/ folder, and hands you a fresh bundle with the LoRA baked in. In a workflow it belongs exactly where the README puts it: Load Anima XPred Model → Load Anima XPred LoRA → Sample Anima XPred.

How it works

Under the hood it re-creates the pack's sd-scripts adapter with the LoRA attached and reloads the model with it merged - which is why it returns a new bundle rather than mutating the one you fed it. If your LoRA uses the newer diffusion_model.*.lora_A/B key format, the node converts it to the sd-scripts lora_unet_* keys it expects, on the fly.

The three inputs

• model - wire it from Load Anima XPred Model. This is what tells the node which checkpoint to merge onto.
• lora - a dropdown of everything in models/loras/. If you haven't dropped a file in yet, the placeholder literally reads put_anima_lora_in_models_loras.safetensors, which is the author being friendly rather than mysterious.
• strength - default 1.0, range −20 to 20. For Anima-style LoRAs you're almost always fine between ~0.6 and 1.2. Negative values work and are how you'd push against a baked-in style; past a couple of notches the range is just a playground.

The gotcha that bites

The LoRA has to be in a format this node recognizes: sd-scripts lora_unet_* keys, or diffusion_model.*.lora_A/B keys it can convert. Drop in a LoRA with some other key layout and you'll get the node's own error: "teacher_lora must use sd-scripts lora_unet_ keys or diffusion_model..lora_A/B keys." If your Anima LoRA was trained with sd-scripts - which is how most of them are, since that's the ecosystem's default training stack - you're fine.

It's also a single-LoRA node. Chain two of these and the second one re-merges from the base checkpoint, effectively replacing the first instead of stacking. Keep it to one.

Install

Same as the rest of the pack - it's all one repo:

cd ComfyUI/custom_nodes
git clone https://github.com/leafmoone/ComfyUI-RUM-Anima-XPred

or grab it via ComfyUI Manager (search "ComfyUI-RUM-Anima-XPred") and restart. No extra pip installs; the pack vendors the sd-scripts code it needs, and your LoRA just sits in models/loras/.

If strength goes too far, the usual LoRA failure modes apply - melted faces, style bleached over everything. There's no community consensus yet on what this specific experiment likes, because the repo is days old and barely anyone's run it. Start at 1.0 and move in 0.1 steps.