anymove
A Claude plugin of skills for anymove, a repo that does motion capture for any creature from one AI video. A video model films a rigged, limb-painted creature from two cameras in one frame. A DINOv2 joint reader trained on that creature reads every joint. The rig is fitted in both views and exported as a GLB.
What it can touch
| files | write, including .env |
|---|---|
| network | download.pytorch.org, fal.ai, fal.run, github.com, huggingface.co, pypi.org, queue.fal.run, rest.alpha.fal.ai |
| shell | runs commands |
| secrets | FAL_KEY |
What it actually does
The skills are long, consistent workflow guides. They tell the agent to: - clone the anymove GitHub repo if it isn't already in one; - create a Python venv and pip install torch and requirements.txt; - run the repo's shell and Python tools and Blender in headless mode; - download example meshes, trained reader checkpoints (.pt) and a LoRA from the repo's GitHub release, checked against sha256 sums in tools/examples.txt; - read a FAL_KEY from .env or the environment and call fal.ai for paid video generation, mesh generation and LoRA training. mocap/lora.py, the only substantive code shown, zips local videos and captions, uploads them and the reference image to fal, submits training or inference jobs, polls for them, and downloads the resulting video and debug zip into runs/. All outputs are written under runs/<c>/ and assets/<c>/. The ~170 prompt files are plain video-generation prompts. The config files are rig and layout data. No hidden text, injection, credential harvesting, persistence or exfiltration appears in the visible files. Most of the executable code was left out of the bundle and could not be reviewed, including mocap/fal.py, tools/*.sh, tools/fetch_examples.sh, blender/*.py, mocap/solve.py and requirements.txt.
What the check found
mediumDownloads programs or archives while running
Files fetched at run time weren't part of this check, and they can change at any time.
mocap/lora.py:43 fal.download(res["debug_dataset"]["url"], run / "debug_dataset.zip")
mediumShips an archive whose contents weren't checked
Files inside archives aren't unpacked or checked, so they could hide anything.
examples/dragon_roar/det.npz archive, 251772 bytes
mediumUses a fal.ai API key and spends money on paid services
The pipeline reads FAL_KEY and calls paid fal endpoints for video generation (~$0.26–$0.38 per clip), Tripo meshes and LoRA training (~$12 per run). This credential is for the tool's own service, so it isn't malicious. The skills tell the agent never to print the key, to state costs first and to confirm before batch spending, which is good practice. Users should still know the agent can run up charges on their fal account.
skills/anymove/SKILL.md:43 3. For video generation and LoRA training, put `FAL_KEY=...` in `.env` (git ignores it) or export it. Never print it.
skills/anymove/SKILL.md:83 Say what a step will cost before calling fal, and confirm with the person before spending on a batch.
skills/anymove-lora/SKILL.md:39 4. Train. Say the cost first: 1,200 steps is about $12 and about 2 h (rank 32, lr 2e-4, trigger word `TWOCAM`).
mediumClones and runs remote repository code and installs packages
If no checkout exists, the agent is told to git clone the repo, install torch from the PyTorch index and pip install requirements.txt. requirements.txt was not in the reviewed bundle. The agent then runs the repo's shell scripts and Python modules. That code is whatever is on GitHub at clone time, not necessarily what was reviewed here.
skills/anymove/SKILL.md:36 `git clone https://github.com/Patch-gg/anymove && cd anymove`.
skills/anymove/SKILL.md:38 `.venv/bin/pip install torch==2.11.0 --index-url https://download.pytorch.org/whl/cu128` and
skills/anymove/SKILL.md:39 `.venv/bin/pip install -r requirements.txt`. Blender 5.x must be on PATH, or set `BLENDER=/path/to/blender`.
mediumDownloads and loads PyTorch checkpoint files from a GitHub release
Pre-trained reader checkpoints (.pt, ~355 MB each) are fetched from the release and later loaded by mocap.detect. Loading a PyTorch pickle can run code. sha256 sums are pinned in tools/examples.txt, which lowers the risk, but the loader code (mocap/detect.py) and tools/fetch_examples.sh were not available to confirm the checks or a safe load mode.
tools/examples.txt:9 8a797cc9ce7162cdf1a835a96ab06fe7214e1d4d258bc98d970aa59d1ecdad75 355298675 reader-wolf.pt runs/wolf/reader.pt
skills/anymove/SKILL.md:54 - No `runs/<c>/reader.pt`: anymove-reader (about 45 min on one RTX 4090), or `tools/fetch_examples.sh <c> --readers`.
Keep an eye on it
This check is of one commit. Watch anymove and we'll email you if a later commit changes what it can touch.
Made this? Add the badge
[](https://plugsday.com/tools/patch-gg-anymove)About this check
Checked November 2, 2026 at commit c3cb22c26491, 266 files. Rules for known red flags, then two Claude reviews, one of them told to assume the author is malicious. A later commit hasn't been checked. How checks work.