Research map / Video / Video generation
LottieGPT: Tokenizing Vector Animation for Autoregressive Generation
Junhao Chen, Kejun Gao, Yuehan Cui, Mingze Sun, Mingjin Chen, Shaohui Wang, Xiaoxiao Long, Fei Ma, Qi Tian, Hao Zhao, Ruqi Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the Lottie tokenizer instead of raw Lottie JSON text (ablation) | Qwen2.5-VL-7B · LottieBench animation, text-only | valid rate (%): 22.61% -> 78.35% | Improved |
| add the Lottie tokenizer instead of raw Lottie JSON text (ablation) | Qwen2.5-VL-7B · LottieBench animation, text-only | CLIP: 0.7456 -> 0.8156 | Improved |
| add the Lottie tokenizer instead of raw Lottie JSON text (ablation) | Qwen2.5-VL-7B · LottieBench animation, text-only | DINOv2: 0.4789 -> 0.7715 | Improved |
| add Stage 2 animation training after static Stage 1 (ablation) | LottieGPT-7B · LottieBench animation, text-only | valid rate (%): 78.35% -> 96.96% | Improved |
| add Stage 2 animation training after static Stage 1 (ablation) | LottieGPT-7B · LottieBench animation, text-only | LPIPS: 0.1925 -> 0.0366 | Improved |
| add Stage 2 animation training after static Stage 1 (ablation) | LottieGPT-7B · LottieBench animation, text-only | JSON structure: 0.2347 -> 0.8062 | Improved |
| add the Lottie tokenizer instead of QwenVL tokenizer on Lottie JSON | tokenizer · MMSVG-icon | avg tokens (k): 1.6 k -> 1.3 k | Improved |
| add numeric quantization to the Lottie tokenizer | Lottie tokenizer · LottieAnimation | avg tokens (k): 17.4 k -> 6.6 k | Improved |
| use Lottie JSON instead of SVG code as training text | QwenVL tokenizer · MMSVG-icon | avg tokens (k): 2.6 k -> 1.6 k | Improved |
| simplify Lottie JSON (prune fields, round values) | Lottie dataset · LottieAnimation | avg JSON size: 60.72 KB -> 39.87 KB | Improved |
| joint training on mixed static and dynamic data instead of two stages (ablation) | LottieGPT · training stability | convergence | Hurt |
| train on Lottie JSON rather than SVG-based SVG generator (LottieGPT vs OmniSVG) | LottieGPT-7B-Stage1 · LottieBench static, text-only | CLIP: 0.8321 -> 0.9331 | Improved |
Closest papers by meaning
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.87)
- Animation2Code: Evaluating Temporal Visual Reasoning in Video-to-Code Generation (NeurIPS 2026 · similarity 0.86)
- Parallelized Autoregressive Visual Generation (CVPR 2025 · similarity 0.86)
- MotiF: Making Text Count in Image Animation with Motion Focal Loss (CVPR 2025 · similarity 0.86)
- AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning (ICLR 2024 · similarity 0.86)
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
