anymove
Creature animation from one AI-generated video
- Best for
- Technically confident developers who need animations for non-human creatures such as dragons, spiders or quadrupeds
- Makes
- 24 fps GLB animations for a rigged creature, captured from an AI-generated two-view video
- Made by
- Patch (patch-gg)
- Where you use it
- CLI
- Engines
- Blender
- Pricing
- Free and open source (Apache-2.0); you pay for video generation on fal, about $0.26 per 5-second clip per the README (October 2026)
- Free plan
- Yes
- Open source
- Yes (Apache-2.0)
- Export
- A GLB animation at 24 fps of the rigged creature, according to the README
- Status
- Available
anymove is a free, open-source tool from Patch (patch-gg on GitHub) that turns one AI-generated video into an animation for a rigged creature, including ones with wings, tails or eight legs. It is released under the Apache-2.0 license.
The method, which the README credits to @CanItCode (with the MiniMax H3 LoRA step credited to @Graalitoo), has a video model film the creature from the front and the side in a single frame. anymove then renders 2,200 random poses of that one creature (about 25 minutes) and trains a joint reader on them, based on Meta's DINOv2 (about 20 minutes on one GPU); the README reports its average miss falling from 40 pixels to 6. The reader finds each joint in both views, the two views fix it in 3D, and the rig is fitted frame by frame and exported as a 24 fps GLB with no hand edits, in under a minute per clip.
The code is free, but generation isn't. The README's example costs about $0.40 for a Tripo H3.1 mesh (from a FLUX 3 concept image) and about $0.26 for each 5-second MiniMax H3 Max clip on fal. Four example creatures (a wolf, a dragon with a 42-bone rig, a giant spider and a fire demon) can be downloaded as starting points. You need an NVIDIA GPU with CUDA and 24 GB of memory to train a reader, Python 3.11, Blender 5.x and a fal key; a CPU demo fit lets you try the last step without that hardware. Skeletons are placed by hand, and eight-legged walking and lying-down poses don't fit well.
anymove also ships as a Claude Code plugin with Agent Skills that walk an agent through each step. It suits developers comfortable with Python and Blender more than beginners. AIGamer hasn't tested it.
Release history
- October 27, 2026: first public version and the examples-v1 release of example creatures published on GitHub under Apache-2.0 (releases)
Limitation: Training the per-creature joint reader needs an NVIDIA CUDA GPU with 24 GB, and the README says eight-legged walking and lying-down poses don't fit well.
FAQ
Is anymove free?
Yes, the code is free under the Apache-2.0 license. You pay for the AI video clips on fal, about $0.26 for a 5-second MiniMax H3 Max clip according to the README.
What hardware does anymove need?
An NVIDIA CUDA GPU with 24 GB to train the joint reader, plus Python 3.11, Blender 5.x and a fal API key. Reading and fitting need much less, and a CPU demo fit is included.
Can anymove animate any creature?
It's built for non-human rigs such as dragons, spiders, wolves and demons, but the README says eight-legged walking and lying-down poses don't fit well; big single gestures work best.
What format does anymove export?
A 24 fps GLB animation of the rigged creature.
Sources
Compiled by AIGamer from anymove's own pages and the sources below, checked 29 October 2026. We haven't tested anymove hands-on.
- patch-gg: anymove README on GitHub
- patch-gg: anymove examples-v1 release on GitHub
Image: GitHub / patch-gg
Written in our own words; quotes are marked and credited. How we report