Open source
FreeApache-2.0The pipeline is free; you pay fal for video clips (about $0.26 per 5-second clip) and meshes, and supply your own GPU or rent one. Official source 1 ↗
Reviewed tool profile
anymove
Open-source motion capture for any 3D creature from one AI video, exported as a game-ready GLB.
Sources checked
- Pricing
- The code is free and open source under Apache-2.0. The AI services it calls on fal cost money: about $3 for a usable move all in, and about $17 for a new creature with about 5 moves (about $11 of AI services on your own GPU). Official source 1 ↗
- Available as
- Runs locally with an NVIDIA CUDA GPU (12 GB to train a reader, 24 GB safer), Blender 5.x, Python 3.11 and a fal key; there is no cloud version. · Works on Linux and on Windows through Git Bash. · Installs as a Claude Code plugin, and its skills are plain Agent Skills other agents can load. Official source 1 ↗ Official source 1 ↗ Official source 1 ↗
- Primary group
- Image & Design
Also consider
See all 4 similar tools ↓- Atlas CloudOne API key for 400+ image, video, audio, 3D and language models, with an MCP server, CLI and agent skills.Paid
- Hi3DImage-to-3D model generator with split-to-print, multi-colour prep, print service, API and an MCP server; Pro from $19.90/month.Free plan
- MaketAI home design studio: floor plans, 3D views, renders and custom furniture from text or a photo.Free plan
What it does
anymove is an open-source (Apache-2.0) pipeline from Patch that turns one AI-generated video into a game-ready animation for any rigged creature, including ones with wings, tails or eight legs. A video model films the creature from a front and a side camera in one frame, a joint reader trained on that creature finds every joint in both views, the two views pin each joint in 3D, and the rig is fitted frame by frame and exported as a GLB. It runs locally on an NVIDIA GPU with Blender and Python, uses fal for the AI video and mesh steps, and ships agent skills for Claude Code and other agents.
Official source 1 ↗Features
- Two-camera AI video capture. A video model (MiniMax H3 Max on fal by default) animates the creature from a front and a side camera in one frame, so each joint can be pinned in 3D. Official source 1 ↗
- Per-creature joint reader. A DINOv2 vision model learns each creature from about 2,200 random poses with known joints; on the dragon its average miss fell from 40 px to 6 px. Official source 1 ↗
- Rig fit and GLB export. The rig is fitted to the joints frame by frame and exported as a GLB any engine can play; the best takes become one file with seamless loops and root motion. Official source 1 ↗
- Any body plan. Big single gestures work on any body plan (roars, slams, rearing, taking off), and walks and runs work on two and four legs. Official source 1 ↗
- Mesh from a picture. If you have no model, one picture becomes a textured mesh at real size on fal. Official source 1 ↗
- Agent skills. Seven skills walk Claude Code or another agent through rigging, filming, training the reader, capture and packing. Official source 1 ↗
- Ready-made creatures. Nine creatures with 4 to 7 moves each are ready to use in Patch's asset library. Official source 1 ↗
Pricing & free limits
Billing details
Most clips fail the checks: a usable move took about 7 clips on average and 58% of captured clips passed, which is why a move costs about $3 rather than the price of one clip. Official source 1 ↗
Free access & limits
Free access and its limits were not established by the checked official sources.
Best for & limitations
Give a game creature such as a dragon or a giant spider roar, slam or take-off animations without a mocap suit or hand keyframing. Official source 1 ↗
Use one of the nine ready-made animated creatures from Patch's asset library in a browser game. Official source 1 ↗
- Needs an NVIDIA GPU. Without an NVIDIA card it doesn't run, and there is no cloud version. Official source 1 ↗
- Weak moves. Eight-legged walking and bird wings don't work well, and feet slide 1 to 2 cm a step. Official source 1 ↗
- One creature per reader. An animation only fits the creature it was made for, and each new model needs a new reader and a hand-placed skeleton. Official source 1 ↗
- Heavy meshes. Generated creatures are 55,000 to 60,000 triangles with 4K textures, heavy for browser games, with no lighter versions yet. Official source 1 ↗
Similar tools
Tools like anymove
Published tools that share its categories.
Atlas CloudOne API key for 400+ image, video, audio, 3D and language models, with an MCP server, CLI and agent skills.
Visit Hi3DImage-to-3D model generator with split-to-print, multi-colour prep, print service, API and an MCP server; Pro from $19.90/month.
Visit MaketAI home design studio: floor plans, 3D views, renders and custom furniture from text or a photo.
Visit 
MeshyAI 3D model generator that turns text, photos or sketches into textured, rigged, print-ready 3D models.
Visit