Animate image with Animate_X_Long

Animate image with Animate_X_Long is the long-video variant of Animate-X, and it's the quietest node in the pack - zero impressions, zero fanfare, and it deserves a second look if you've already committed to Animate-X over UniAnimate. Where the base Animate-X node samples a fixed clip, this one runs a sliding context window over an arbitrarily long driving video, chunk by chunk with overlapping frames, so the seams stay invisible and the motion stays coherent past the 32-frame wall. It's the same architecture as UniAnimateImageLong, just pointed at the Animate-X checkpoint.

Structurally it mirrors its UniAnimate twin almost exactly:

• image - your reference photo.
• video - the driving clip; DWPose extraction and alignment happen internally.
• useFirstFrame (default false), dontAlignPose (default false) - the internal pose toggles.
• context_size (16–64, default 32), context_overlap (4–16, default 8), context_stride (1–4, default 1) - the sliding-window controls. Bigger context buys motion coherence per chunk at VRAM and time cost; overlap is what prevents flicker at chunk boundaries.
• max_frames (default 1024000 - effectively unlimited, capped by your driving video), frame_interval (1–16), seed, steps, resolution_x.

Same auto-adjustment guardrails as the UniAnimate Long node: if the context settings conflict, or the clip is under 32 frames, it resets to safe defaults rather than erroring. Outputs are video (the full generated batch) and poses (the internally built Animate-X pose sequence).

Why run this over UniAnimateImageLong?

One reason: motion size. Animate-X's Explicit Pose Indicator holds identity through bigger, more expressive movement than plain UniAnimate does, and that advantage doesn't disappear just because you're generating long-form. If your driving video is a dance routine with real range, this node is the long version of the better-motion model. The costs are the same as the UniAnimate Long node - long clips mean long generations (roughly eight minutes per 32 frames at [512, 768]), and the checkpoint stack is identical, plus the one Animate-X file.

Setup

Pack-standard installation: ComfyUI Manager ("UniAnimate Nodes for ComfyUI") or git clone https://github.com/Isi-dev/ComfyUI-UniAnimate-W into custom_nodes/. Models are the shared ~14GB UniAnimate stack plus animate-x_ckpt.pth in checkpoints/. It's also the most VRAM-forgiving of the Animate-X nodes for long content, because the sliding window keeps any single chunk small - keep context_size at 32 on a 12GB card. One practical note: because it generates its own poses, it pairs naturally with Animate_X_Image_v2's defaults (which favor useFirstFrame on) when you're chaining long generations together.