AnimTOON-3B: Token-Efficient Vector Animation Generation
AnimTOON-3B is a specialized language model developed by srk0102200, built upon the Qwen2.5-3B-Instruct base model with a LoRA merge. Its core innovation is a compact, plain-text animation format that drastically reduces the number of tokens required to describe Lottie animations. This model achieves a 98.8% token reduction compared to raw Lottie JSON, using 3-4x fewer tokens than models like OmniLottie for similar animation tasks.
Key Capabilities & Features
- Token Efficiency: Generates animations using significantly fewer tokens (e.g., 166-597 tokens for complex animations vs. thousands).
- Lottie Output: The generated plain-text format can be converted into complete animated
.lottie files. - SVG Integration: Accepts existing SVG files as input for shapes, allowing the model to focus purely on animation logic without hallucinating shapes.
- Character Animation: Version 3 (v3) introduces support for character animation, including idle/walk cycles, trained on Spine and DragonBones skeletal data.
- Resource-Light: Operates with lower VRAM requirements (5GB) and achieves higher FPS (30) compared to alternative methods.
- Separation of Concerns: Follows an architecture where SVG provides perfect shapes, the model generates animation, and a converter merges them, ensuring valid and precise output.
Ideal Use Cases
- Icon and Logo Animation: Generating dynamic effects like pulse, bounce, spin, fade, and wobble for UI elements.
- Character Movement: Creating basic character idle and walk cycles for games or interactive applications.
- SVG Animation: Adding complex animations to existing SVG assets, such as multi-part SVG animations with coordinated movements.
- Developers seeking efficient animation generation from text prompts, especially when working with Lottie files and vector graphics.