Research map / Architectures and efficiency / Pruning and efficient tuning
Prompt2Effect: Training-Free LoRA Synthesis for Controllable Video Effects
Xiaomeng Yang, Yanyu Li, Gordon Qian, Ivan Skorokhodov, Viacheslav Ivanov, Avalon Vinella, Xuan Zhang, Yanzhi Wang, Sergey Tulyakov, Anil Kag
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 weight-driven base-weight token input (vs noise embeddings) (ablation) | Prompt2Effect · IID effects | VLM Overall: 46.14 -> 56.93 | Improved |
| add weight-driven base-weight token input (vs noise embeddings) (ablation) | Prompt2Effect · OOD effects | VLM Overall: 43.58 -> 48.87 | Improved |
| use one joint hypernetwork instead of per-block ensemble (ablation) | Prompt2Effect · IID effects | VLM Overall: 51.29 -> 56.93 | Improved |
| use one joint hypernetwork instead of per-block ensemble (ablation) | Prompt2Effect · OOD effects | VLM Overall: 45.11 -> 48.87 | Improved |
| predict SVD-canonicalized LoRA factors instead of raw LoRA (ablation) | Prompt2Effect · training NMSE | NMSE convergence | Improved |
| compress weight tokens to n=256 instead of n=128 (ablation) | Prompt2Effect · IID effects | VLM Overall: 49.75 -> 56.93 | Improved |
| compress weight tokens to n=256 instead of n=512 (ablation) | Prompt2Effect · IID effects | VLM Overall: 56.22 -> 56.93 | No clear effect |
| increase tokens from n=256 to n=512 (ablation) | Prompt2Effect · OOD effects | Aesthetic Quality: 53.44 -> 54.02 | Mixed |
| use predicted weights as init, 100 steps, vs LoRA from scratch 100 steps | Prompt2Effect · OOD effects | VLM Overall: 52.28 -> 56.33 | Improved |
| use predicted weights as init, 50 steps, vs LoRA from scratch 100 steps | Prompt2Effect · OOD effects | VLM Overall: 52.28 -> 54.89 | Improved |
| fast adaptation Init-100 vs 1000-step LoRA | Prompt2Effect · OOD effects | CLIP Score: 25.73 -> 25.82 | Improved |
| zero-shot Prompt2Effect vs base I2V model | Prompt2Effect · IID effects | VLM Overall: 42.57 -> 56.93 | Improved |
Closest papers by meaning
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.87)
- LoRA-X: Bridging Foundation Models with Training-Free Cross-Model Adaptation (ICLR 2025 · similarity 0.87)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.87)
- LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025 · similarity 0.87)
- Activation Steering of Video Generation Models via Reduced-Order Linear Optimal Control (NeurIPS 2026 · similarity 0.86)
- Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model (ICLR 2025 · 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.
