Repose image with Animate_X

Repose image with Animate_X is the Animate-X flavor of the pack's still-image pose transfer: give it a photo of your person and a second photo to borrow a pose from, and it returns a single reposed frame that keeps the identity and takes the geometry. On paper that's the same job as Repose image with UniAnimate (ReposeImage), just running through the Animate-X checkpoint instead of the UniAnimate one.

Here's the honest part, straight from the author's release notes: the Animate-X repose nodes did not show improvement over the UniAnimate repose node in image-to-image pose transfer tests. This is the v1 of that experiment. If you're trying one repose node for the first time, ReposeImage is the one the pack's own developer recommends. This node is for when you specifically want Animate-X's character handling - which mostly matters when the pose you're copying is extreme enough that identity drift becomes a problem, since Animate-X's Explicit Pose Indicator conditions on the character more aggressively.

Inputs and outputs

The schema is identical to ReposeImage, which makes the comparison easy:

• image - the person whose identity you keep.
• pose - the photo whose pose you copy. DWPose detection runs internally; you don't pre-extract anything.
• dontAlignPose (default false) - false auto-scales the target pose to your person's proportions; true copies the pose as-is.
• seed, steps (default 25), resolution_x (512 or 768).

Outputs are newPose (the reposed still) and pose (the DWPose skeleton it was generated from).

Setup and the practical take

Installation is pack-standard: ComfyUI Manager ("UniAnimate Nodes for ComfyUI") or git clone https://github.com/Isi-dev/ComfyUI-UniAnimate-W into custom_nodes/. Because it's an Animate-X node it needs animate-x_ckpt.pth in checkpoints/ on top of the shared ~14GB UniAnimate stack, and it's the one extra download that buys you the least here given the author's own verdict.

My take: use ReposeImage unless you hit identity drift on extreme poses, in which case try this one and the v2 sibling to see if Animate-X's conditioning holds better. That's the actual use case - a fallback, not a default. And since it produces a single still rather than a sequence, VRAM stays modest and the ~12GB animation-node budget never comes into play.