wuwukaka/ComfyUI-WanAnimatePlus

ComfyUI‑WanAnimatePlus

What it is – A collection of custom nodes for ComfyUI that extend the WanVideo pipeline. It adds multi‑reference image injection, seamless video stitching, and full support for the newer SCAIL‑2 and Bernini video‑generation models.

Why it matters – The original WanVideo workflow only allowed a single reference image and could not easily continue a video across multiple generation runs. WanAnimatePlus solves these limits, making it possible to:

• Guide a generation with up to 5 reference images (prefix_frames).
• Connect consecutive video segments without visible cuts by feeding the last 21 frames of the previous segment (transition_video).
• Use Bernini models for video‑to‑video, reference‑video‑to‑video, reference‑image‑to‑video, and text‑to‑video tasks.
• Work with SCAIL‑2 checkpoints, handling reference images, pose streams, masks, background images, and two‑phase sampling through a set of dedicated nodes.

Key features

Feature	What it does
prefix_frames	Accepts 1‑5 extra reference images; the node expands the canvas, encodes them, and automatically aligns pose/face offsets.
transition_video	Takes the last 21 frames of a previous clip, writes them into the front of the generation canvas, and trims them after decoding for a seamless hand‑off.
Bernini support	Generates conditioning latents from source video, reference video, or reference images; automatically detects the task (v2v, rv2v, r2v, t2v).
SCAIL‑2 embeds	Pre‑processes reference image, driving pose, colored pose mask, optional prefix/transition latents, and background image into the format expected by SCAIL‑2 models.
Official‑compatible flow nodes	WanAnimatePlus SCAIL_2 Flow Embeds, Flow Sampler, and VAE Decode expose the same MODEL / VAE / CONDITIONING / LATENT / IMAGE interfaces used by vanilla ComfyUI, allowing SCAIL‑2 workflows to be built with standard node wiring.
Easy sampler nodes	WanAnimatePlus Easy Sampler and Easy SamplerSettings hide the many low‑level sampler arguments while still using the full WanAnimatePlus sampler under the hood.
Memory helpers	Options such as force_offload and tiled_vae let you run the heavy video models on modest GPUs.

Typical use‑cases

• Creating long, coherent video sequences with multiple visual references (e.g., a character appearing in several poses).
• Extending an existing clip—e.g., continue an animation or edit the tail of a video.
• Transferring motion from a source video onto a new subject while also injecting reference images.
• Using SCAIL‑2 models for high‑quality pose‑driven animation with optional background replacement.

Installation

# Inside your ComfyUI installation
cd ComfyUI/custom_nodes
git clone https://github.com/wuwukaka/ComfyUI-WanAnimatePlus.git
# Restart ComfyUI so the new nodes are registered

The README notes that the original ComfyUI‑WanVideoWrapper must also be installed, but do not mix WanAnimatePlus nodes with the original WanVideo nodes in the same workflow.

Quick‑start workflow

1. Open ComfyUI; a new category WanAnimatePlus appears.
2. Replace the standard WanVideo chain with the WanAnimatePlus equivalents:
   • WanAnimatePlus ModelLoader
   • WanAnimatePlus VAELoader
   • WanAnimatePlus ContextOptions
   • WanAnimatePlus AnimateEmbeds (now with prefix_frames and transition_video inputs)
   • WanAnimatePlus Sampler / Easy Sampler
   • WanAnimatePlus Decode
3. Connect optional inputs:
   • Prefix frames – up to 5 reference images.
   • Transition video – last 21 frames of the previous segment.
   • For SCAIL‑2: use WanAnimatePlus SCAIL_2 Embeds (or the Flow variant) together with pose images, masks, etc.
4. Run the workflow. If you supplied a transition clip, the output video will be trimmed to remove the prepended frames, giving a smooth continuation.

Example resources

• example_workflows/ – ready‑made JSON workflows demonstrating multi‑reference generation, video continuation, and SCAIL‑2 usage.
• RunningHub links in the README let you try the workflows on a free cloud instance (use the invite code to get 1 000 credits).

Project structure – The repo mirrors a typical ComfyUI custom node layout: core node definitions (nodes.py, nodes_sampler.py), model‑loading helpers, context‑window scheduling, and utility modules. All heavy lifting lives in the wanvideo/ subdirectory, which is a fork of the original WanVideo code.

License & attribution – Apache‑2.0, with a NOTICE file crediting the upstream ComfyUI‑WanVideoWrapper and other helper projects.

---

Bottom line – If you are already using ComfyUI for video generation and need multi‑reference control or seamless stitching, WanAnimatePlus provides a drop‑in set of nodes that extend the WanVideo pipeline while staying compatible with the official ComfyUI node ecosystem.