Research map / Image generation / Text-to-image diffusion
TraDiffusion: Trajectory-Based Training-Free Image Generation
Mingrui Wu, Oucheng Huang, Jiayi Ji, Jianzhuang Liu, Xiaoshuai Sun, Liujuan Cao, Rongrong Ji
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 distance-aware control energy instead of prior structure guidance (ablation) | TraDiffusion · COCO2014 | DTL: 0.0077 -> 0.0119 | Improved |
| use control energy instead of trajectory expanding (ablation) | TraDiffusion · COCO2014 | DTL: 0.0080 -> 0.0119 | Improved |
| add movement loss (ablation) | TraDiffusion · COCO2014 | DTL: 0.0119 -> 0.0156 | Improved |
| add movement loss (ablation) | TraDiffusion · COCO2014 | FID: 64.68 -> 68.53 | Hurt |
| increase lambda from 10 to 20 (ablation) | TraDiffusion · COCO2014 | DTL: 0.0156 -> 0.0158 | No clear effect |
| increase lambda from 10 to 20 (ablation) | TraDiffusion · COCO2014 | FID: 68.53 -> 72.80 | Hurt |
| increase lambda to 100 (ablation) | TraDiffusion · COCO2014 | FID: 68.53 -> 129.91 | Hurt |
| use trajectory control instead of box control (BoxDiff) | TraDiffusion · User study | Controllability: 3.22 -> 4.04 | Improved |
| use trajectory control instead of mask control (DenseDiffusion) | TraDiffusion · User study | User-friendliness: 1.07 -> 2.87 | Improved |
| use trajectory control instead of Backward Guidance | TraDiffusion · COCO2014 | FID: 69.04 -> 68.53 | Improved |
Closest papers by meaning
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.84)
- MotionPro: A Precise Motion Controller for Image-to-Video Generation (CVPR 2025 · similarity 0.84)
- Motion Prompting: Controlling Video Generation with Motion Trajectories (CVPR 2025 · similarity 0.84)
- LeviTor: 3D Trajectory Oriented Image-to-Video Synthesis (CVPR 2025 · similarity 0.84)
- Trajectory attention for fine-grained video motion control (ICLR 2025 · similarity 0.84)
- PoseTraj: Pose-Aware Trajectory Control in Video Diffusion (CVPR 2025 · similarity 0.84)
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.
