🚫AnimateDiff Loader [DEPRECATED] 🎭🅐🅓

The name isn't subtle: this node's own display label is "🚫AnimateDiff Loader [DEPRECATED]". If you've hit this page because an old workflow uses it, don't panic - it still runs. If you're starting something new, skip it and use the current loader/apply nodes instead. Here's why, and what this one actually did.

What it was

This is the original, all-in-one AnimateDiff node from before the pack grew into what it is today. Look at its inputs and you can see the whole story: a model, a latents tensor, a model_name (which motion module to load), a beta_schedule, and a checkbox called unlimited_area_hack. That's it. No context options, no sample settings, no keyframe scheduling - because none of that existed yet when this node was written. It loads a motion module by name, patches it onto your model, and hands back both the patched MODEL and the LATENT you gave it, unmodified, purely so it slots neatly into a linear model→sampler chain.

unlimited_area_hack is a relic from the same era - a crude pre-context-window trick for pushing frame counts past what the motion module was trained on, before AnimateDiff-Evolved shipped the sliding context window that made that hack unnecessary. It's kept for backward compatibility, not because anyone should reach for it.

Why you shouldn't use it today

The whole point of AnimateDiff-Evolved, per its own README, is Context Options for infinite-length generation, Sample Settings for custom noise scheduling, and a growing pile of Gen2 features (CameraCtrl, PIA, HotshotXL, AnimateLCM) that this loader has no inputs for at all. It's a snapshot of the pack from before any of that shipped. Loading a modern motion module through it will generally still work, but you're locking yourself out of everything that makes AnimateDiff-Evolved worth using over the plain original AnimateDiff.

Inputs and outputs

• model - your base SD checkpoint's MODEL, going in for patching.
• latents - the LATENT batch to animate (frame count = batch size here, since there's no context window to split it).
• model_name - dropdown of motion modules found in your models folder.
• beta_schedule - the noise schedule family; has to match what your motion module expects (see the README's model-specific notes - HotshotXL wants linear (HotshotXL/default), PIA wants sqrt_linear (AnimateDiff), AnimateLCM wants lcm).
• unlimited_area_hack - legacy toggle, leave off unless you're deliberately reproducing an old workflow's behavior.

Outputs are MODEL (patched with the motion module) and LATENT (passed through), feeding straight into a KSampler.

How to install it

You get this node automatically with the whole pack - there's no reason to install for this specifically.

• ComfyUI Manager - search AnimateDiff Evolved, confirm the author is Kosinkadink, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved, then restart.
• Models - download at least one motion module (mm_sd_v15_v2, v3_sd15_mm, etc.) and drop it in ComfyUI/models/animatediff_models or ComfyUI/custom_nodes/ComfyUI-AnimateDiff-Evolved/models.

Common issues & troubleshooting

It's not in the search results, or you can't find it. ComfyUI hides some deprecated nodes from the default add-node menu depending on version; if you're rebuilding an old workflow, loading the workflow JSON directly will still resolve it as long as the pack is installed.

Watermark showing up in outputs. Not specific to this node - it's a known AnimateDiff quirk. mm_sd_v15 was fine-tuned on data with Shutterstock watermarks baked in, and the motion module can reproduce a faint version of that transparency. Switch motion modules (v14, v15_v2, or v3) if you see it.

You inherited a workflow using this node and it errors after an update. Rebuild the loading chain with the current Gen1/Gen2 nodes rather than debugging this one - it's not where new development effort goes, and the current nodes give you context windows and sample settings for free.