example creatures

The example creatures from the README: each one's mesh, its trained joint reader, one AI two-camera clip with the reference frame it started from, and the wolf's two-camera LoRA on MiniMax H3.

tools/fetch_examples.sh downloads them to the right places and checks every file's SHA-256 against tools/examples.txt:

tools/fetch_examples.sh dragon --readers --clips   # one creature: mesh, reader, clip
tools/fetch_examples.sh --readers --clips          # all four
tools/fetch_examples.sh wolf --lora                # the wolf's H3 LoRA (skills/anymove-lora)

file	what	size
wolf.glb dragon.glb spider.glb demon.glb	Tripo H3.1 meshes each assets/<creature>/rig.json was placed on	11-15 MB
reader-<creature>.pt	trained joint readers (DINOv2 ViT-B/14 with registers + heatmap head, 20 epochs on 2,200 renders)	355 MB
clip-<creature>-<move>.mp4, ref-<creature>.png	one MiniMax H3 Max two-camera clip per creature and its reference frame	3-5 MB
lora-wolf-twocam-h3.safetensors	the wolf's two-camera LoRA on MiniMax H3 (1,200 steps, trigger word TWOCAM)	131 MB

Licences: the meshes were made with Tripo and the clips with MiniMax H3 Max, both on fal, so their terms apply to those files. The readers are fine-tuned from Meta's DINOv2 (Apache-2.0). The LoRA is a LoRA on MiniMax H3, so MiniMax H3's community licence applies to it, not this repo's Apache-2.0.