Research map / Video / Video generation
TITAN-Guide: Taming Inference-Time Alignment for Guided Text-to-Video Diffusion Models
Christian Simon, Masato Ishii, Akio Hayakawa, Zhi Zhong, Shusuke Takahashi, Takashi Shibuya, Yuki Mitsufuji
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 |
|---|---|---|---|
| use TITAN-Guide forward-gradient guidance instead of backprop-based guidance | AnimateDiff guided T2V · VGG-Sound audio-video alignment | GPU memory: 67.2GB (DOODL) -> 20.6GB | Improved |
| sample gradient guesses from a subset of frames instead of random Gaussian guesses (ablation) | TITAN-Guide · VGG-Sound audio-video alignment (AnimateDiff) | FVD: 348.26 -> 331.11 | Improved |
| use score-based guesses instead of random guesses (ablation) | TITAN-Guide · VGG-Sound audio-video alignment (AnimateDiff) | FVD: 348.26 -> 333.42 | Improved |
| increase guidance iterations from 20 to 50 or 100 (ablation) | TITAN-Guide random guesses · VGG-Sound audio-video alignment | FVD: 348.26 -> 340.21 (50), 350.52 (100) | No clear effect |
| increase resolution from 256x256 to 384x384 (ablation) | TITAN-Guide · VGG-Sound audio-video alignment (AnimateDiff) | all metrics: FVD 331.11 -> 317.05 (sampled guesses) | Improved |
| traverse to t=0 for clean estimate instead of one-step Tweedie estimate | classifier guidance for T2V · VGG-Sound audio-video alignment | AV-Align | Improved |
| apply TITAN-Guide on CogVideoX backbone with TFG as baseline | CogVideoX · VGG-Sound audio-video alignment | FVD: 343.28 -> 318.99 | Improved |
| TITAN-Guide on style guidance ED vs TFG | AnimateDiff · TFG-1000 style guidance | ED: TFG 8.92 vs TITAN 9.16 | Hurt |
| TITAN-Guide on image guidance tasks | CAT-DDPM / CelebA-DDPM · Super resolution, CelebA, Deblurring | FID/LPIPS/Acc: SR FID 75.9 vs TFG 65.9; CelebA Acc 77.2 vs 75.2 | Mixed |
| TITAN-Guide needs extra time to traverse to t=0 | TITAN-Guide · 16 frames 256x256 on H200 | guidance time: about 40 seconds -> 2 minutes | Hurt |
Closest papers by meaning
- REG: Rectified Gradient Guidance for Conditional Diffusion Models (ICML 2025 · similarity 0.89)
- VideoGuide: Improving Video Diffusion Models without Training Through a Teacher's Guide (CVPR 2025 · similarity 0.89)
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.89)
- TI2V-Zero: Zero-Shot Image Conditioning for Text-to-Video Diffusion Models (CVPR 2024 · similarity 0.89)
- Gradient Guidance for Diffusion Models: An Optimization Perspective (NeurIPS 2024 · similarity 0.89)
- Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models (AAAI 2025 · similarity 0.88)
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.
