Load Anima LLLite Model

This is a deliberately tiny node with a specific job: list the Anima ControlNet-LLLite weight files sitting in your controlnet folder and hand you back the one you picked as a filename. That filename is exactly what the Anima LLLite Tiled ControlNet Sampler (its big sibling in the same pack) wants on its lllite_name input. So instead of reaching into the sampler to change models each time, you pick the file once here and route it into one or more samplers - or park the loader in a loaders group, like you already do with checkpoint and LoRA loaders.

Why it exists

Anima - Circlestone's 2B anime base on Cosmos-Predict2 - famously has no conventional ControlNet. The only structural control option is kohya-ss's ControlNet-LLLite port, a lightweight LoRA-style module injected at attention and MLP points rather than a copied encoder. That port is genuinely experimental: the controlnet KB notes that the Anima LLLite models are reported to barely move the image, with the community often falling back to training a control LoRA instead. So don't expect this node to magically fix Anima's control gap - expect it to make the LLLite path you do have slightly less awkward to operate.

The vsLinx sampler builds the LLLite module itself, so this loader only outputs the filename; it has zero dependency on any other node pack. That's also why it's safe to leave in a workflow you share.

The one input and one output

• lllite_name - a dropdown populated from everything in models/controlnet. If you've just dropped a new file in, refresh the ComfyUI folder listing (the dropdown reads the folder on load).
• Output lllite_name - the filename string, wired into the sampler's lllite_name slot.

That's the whole node. It doesn't load weights into VRAM, doesn't apply anything, doesn't need a model input. It's a combo-list selector with a wire.

Getting the weights

Nothing is bundled - you download the LLLite .safetensors yourself and drop them in ComfyUI/models/controlnet. The canonical source is kohya-ss's Anima-LLLite repo (also linked from the pack README's credits). There are community tile/repair and regional-conditioning LLLite files floating around too; one known workable recipe for the tile model is strength 0.8 with denoise 0.5.

Install

It ships in the vsLinx pack. ComfyUI Manager → search "ComfyUI vsLinx Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/vslinx/ComfyUI-vslinx-nodes.git comfyui-vslinx-nodes

Then restart ComfyUI. No extra Python packages, no other node packs needed - the sampler's LLLite apply logic is vendored inside this pack.

Honest assessment

If you're on SDXL or Illustrious, you don't need this node at all - your models have real ControlNets. If you've committed to Anima and want its one control path, this is the clean way to manage it. Just keep expectations calibrated: LLLite on Anima is experimental by the author's own admission, and "barely moves the image" is the community's most common verdict. Use it as the picker it is, don't expect a revolution.