Research map / Video / Video generation
Tea-Adapter: Teacher Adapter for Efficient Conditional Generation
Yinhan Zhang, Yue Ma, Fangqiu Yi, Chenyang Qi, Chi Zhang, Kunyu Feng, Zeyu Wang
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 Mixture of Condition Experts layer (ablation) | Tea-Adapter · Depth map control (Table 2) | FVD: 303.202 -> 292.341 | Improved |
| add Mixture of Condition Experts layer (ablation) | Tea-Adapter · Depth map control (Table 2) | LPIPS: 0.268 -> 0.251 | Improved |
| keep all adapters instead of using half of them (ablation) | Tea-Adapter · Depth map control (Table 2) | FVD: 398.013 -> 292.341 | Improved |
| reduce adapters from 12 to 7 (ablation) | Tea-Adapter · Adapter scaling | Condition fidelity | No clear effect |
| transfer control from a small teacher to the 14B student via adapters instead of DiT-ControlNet on Canny | Wan2.1-14B · Canny edge control | FVD: 425.249 -> 289.565 | Improved |
| use Tea-Adapter instead of full fine-tuning of Wan2.1-14B | Wan2.1-14B · Canny edge control | FVD: 229.186 -> 289.565 | Hurt |
| use Tea-Adapter instead of Ctrl-Adapter on depth | video DiT · Depth map control | FVD: 448.291 -> 292.341 | Improved |
| use Tea-Adapter instead of Ctrl-Adapter for temporal consistency | video DiT · Video generation | Temporal consistency: 0.981 -> 0.984 | Improved |
Closest papers by meaning
- Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model (ICLR 2025 · similarity 0.87)
- DiTCtrl: Exploring Attention Control in Multi-Modal Diffusion Transformer for Tuning-Free Multi-Prompt Longer Video Generation (CVPR 2025 · similarity 0.86)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.85)
- T2V-Turbo-v2: Enhancing Video Model Post-Training through Data, Reward, and Conditional Guidance Design (ICLR 2025 · similarity 0.85)
- Video-STaR: Self-Training Enables Video Instruction Tuning with Any Supervision (ICLR 2025 · similarity 0.85)
- ShotAdapter: Text-to-Multi-Shot Video Generation with Diffusion Models (CVPR 2025 · similarity 0.85)
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.
