Kosinkadink/ComfyUI-AnimateDiff-Evolved

What it solves

This project provides an improved integration of AnimateDiff into ComfyUI, allowing users to generate animations from text or images. It addresses the limitations of basic AnimateDiff implementations by adding advanced sampling options, support for a wide variety of motion modules, and tools for maintaining consistency across longer animations.

How it works

It extends ComfyUI with a set of custom nodes that implement AnimateDiff's motion modules. It uses a sliding context window approach (Context Options) to enable infinite animation length and provides "Evolved Sampling" techniques that can be used both within and outside of AnimateDiff. The system supports various model architectures including SD1.5, SDXL, and LCM, and integrates with other ComfyUI node packs for ControlNet, IPAdapter, and prompt scheduling.

Who it’s for

Digital artists, AI video creators, and ComfyUI users who want high-quality, controllable AI-generated animations with advanced control over motion, noise scheduling, and temporal consistency.

Highlights

• Infinite Animation Length: Uses sliding context windows across the UNet and motion module to bypass frame limits.
• Broad Model Support: Compatible with AnimateDiff v1/v2/v3, HotshotXL, AnimateLCM, and AnimateLCM-I2V.
• Advanced Control: Includes support for Motion LoRAs, prompt travel, and per-frame GLIGEN coordinates.
• ** uma Integration**: Works with ControlNet, SparseCtrl, and IPAdapter for precise structural and visual guidance.
• Consistency Tools: Implements FreeInit, FreeNoise, ContextRef, and NaiveReuse to improve temporal coherence.
• Hardware Compatibility: Supports fp8 for lower VRAM usage and is compatible with Mac M1/M2/M3 chips.