Load AnimateLCM-I2V Model 🎭🅐🅓②

This is the file loader for AnimateLCM-I2V - a motion module built on Fu-Yun Wang's AnimateLCM-I2V research and ported into ComfyUI form by Kosinkadink, distinct from a regular AnimateDiff motion module in two ways: it's LCM-distilled (so it runs in far fewer steps) and it's image-anchored (a reference image steers the generated sequence, rather than motion coming from noise alone). Like every Gen2 loader in this pack, this node's whole job is reading the model file off disk - it doesn't apply anything by itself. You wire its output into ADE_ApplyAnimateLCMI2VModel to actually put it to work.

Reach for this specifically when you want fast, few-step generation with an image reference anchoring the result, rather than a from-scratch motion clip. The README's honest field note is worth repeating: despite the "I2V" name, it's reported to work best for vid2vid rather than pure image-to-video, at least in the author's own testing.

How it works

It's a straight file loader, the same pattern as ADE_LoadAnimateDiffModel - it reads the AnimateLCM-I2V weights into memory and returns typed objects for downstream nodes to consume. Nothing about applying the model, or handling the reference image, happens here.

The inputs and outputs that matter

Required: model_name - a dropdown of AnimateLCM-I2V model files found in your motion-model folders.

Optional: ad_settings (type AD_SETTINGS) for weight or position-encoding adjustments, the same fine-tuning slot every motion loader in the pack has, left unplugged in most workflows.

Two outputs, both type MOTION_MODEL_ADE but meaning different things: MOTION_MODEL is the full model, for the normal case; encoder_only gives you just the image-encoder half, useful if you want to combine the I2V reference-image mechanism with a different motion module downstream rather than running the full AnimateLCM-I2V model on its own. That pairs with the README's own recommended chaining pattern - using this for the first step or two, then handing off to other motion models via additional Apply nodes.

How to install it

Via ComfyUI Manager: search AnimateDiff Evolved, author Kosinkadink, install, restart. Manually:

cd ComfyUI/custom_nodes && git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved

You need the AnimateLCM-I2V weights themselves - HuggingFace wangfuyun/AnimateLCM-I2V, linked in the README - placed in ComfyUI/models/animatediff_models or the pack's own models folder, same as any other motion module. You'll also want the general AnimateLCM settings on the rest of your graph: autoselect/lcm/lcm[100_ots] beta schedule, an LCM LoRA, the lcm sampler in your KSampler, low CFG (roughly 1.0–2.0), and few steps (minimum around 4).

Common issues & troubleshooting

model_name dropdown is empty. Wrong folder, or ComfyUI hasn't rescanned since you added the file - restart the server rather than refreshing the browser.

Loaded fine but the Apply node complains, or output looks nothing like AnimateLCM. Double-check you're feeding your KSampler the lcm sampler and low CFG per the README's settings above - an AnimateLCM-family module run with ordinary KSampler settings (high CFG, non-LCM sampler, 20+ steps) will produce degraded or broken results, and that's a settings mismatch, not a broken load.

Not sure whether to use MOTION_MODEL or encoder_only. Use the full MOTION_MODEL output unless you're specifically chaining this model's image-anchoring behavior into a separate motion module afterward - that's the one case encoder_only exists for, and it's an advanced pattern most workflows don't need.