Research map / Video / Video generation
Free2Guide: Training-Free Text-to-Video Alignment using Image LVLM
Jaemin Kim, Bryan Sangwoo Kim, Jong Chul Ye
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 GPT-4o image LVLM guidance | LaVie · VBench text alignment (LaVie) | Avg. score: 0.3722 -> 0.3827 | Improved |
| add GPT-4o image LVLM guidance | VideoCrafter2 · VBench text alignment (VideoCrafter2) | Avg. score: 0.4129 -> 0.4425 | Improved |
| ensemble GPT-4o with CLIP reward | LaVie + CLIP · VBench text alignment (LaVie) | Avg. score: 0.3780 -> 0.3868 | Improved |
| ensemble GPT-4o with ImageReward | VideoCrafter2 + ImageReward · VBench text alignment (VideoCrafter2) | Avg. score: 0.4435 -> 0.4483 | Improved |
| use weighted sum ensemble rather than consensus (ablation) | LaVie + CLIP + GPT · VBench ensemble methods (LaVie + CLIP) | Avg. score: 0.5679 -> 0.5738 | Improved |
| use weighted sum ensemble rather than normalized sum (ablation) | LaVie + CLIP + GPT · VBench ensemble methods (LaVie + CLIP) | Avg. score: 0.5734 -> 0.5738 | No clear effect |
| use 1-9 scale scoring instead of yes/no logits (ablation) | VideoCrafter2 + GPT · VBench text alignment (VideoCrafter2) | Avg. score: 0.4358 -> 0.4425 | Improved |
| increase samples per step from 5 to 10 (ablation) | LaVie + CLIP · VBench text alignment (LaVie) | Avg. score: 0.3780 -> 0.3705 | Hurt |
| extend guidance range to t in [T,T-10] (ablation) | LaVie + CLIP · VBench text alignment (LaVie) | Avg. score: 0.3780 -> 0.3635 | Hurt |
| add GPT-4o guidance on general video quality | VideoCrafter2 · VBench general video quality (VideoCrafter2) | Avg. score: 0.7617 -> 0.7537 | Hurt |
| ensemble GPT-4o with ImageReward on general quality | LaVie + IR · VBench general video quality (LaVie) | Avg. score: 0.7666 -> 0.7699 | Improved |
| use open-source Qwen2.5-VL 3B image LVLM instead of GPT-4o (ablation) | LaVie · VBench text alignment (LaVie) | Avg. score: 0.3722 -> 0.3816 | Improved |
Closest papers by meaning
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.90)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.89)
- Confidence-aware Reward Optimization for Fine-tuning Text-to-Image Models (ICLR 2024 · similarity 0.88)
- Align-A-Video: Deterministic Reward Tuning of Image Diffusion Models for Consistent Video Editing (CVPR 2025 · similarity 0.88)
- VideoDPO: Omni-Preference Alignment for Video Diffusion Generation (CVPR 2025 · similarity 0.87)
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · similarity 0.87)
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.
