BREAKING
OmniLottie Turns Prompts Into Motion
Three Inputs, One Vector Output
1
Provide prompt
↓
2
Interpret input
↓
3
Generate tokens
↓
4
Decode JSON
Tokenizer Lifts Generation Success
Raw JSON
13.4
Lottie tokens
97.3
0
B
model size
0
M
animations
Open Access, Real Constraints
Available
Open release
●
Weights and Gradio demo
●
ComfyUI integration
Limits
Research stage
●
Training code unreleased
●
Invalid sequences remain possible
Editable Motion, Still Experimental
AI NEWS BLITZ
OpenVGLab's 4B model generates editable Lottie JSON from multimodal prompts.