Research map / Video / Video generation
QWERTY: Training-Free Motion Control via Query-Warped Video Diffusion Transformers
Kyobin Choo, Youngmin Kim, Hyunkyung Han, Geunrip Park, Chanyoung Kim, Sunyoung Jung, Seong Jae Hwang
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 STCD semantic channel selection (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.297 -> 0.237 | Improved |
| add STCD semantic channel selection (ablation) | Qwerty on CogVideoX · VIPSeg object control | FTD: 0.542 -> 0.473 | Improved |
| warp queries only instead of keys only (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.293 -> 0.237 | Improved |
| warp queries only instead of queries and keys (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.290 -> 0.237 | Improved |
| warp queries instead of attention output (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.288 -> 0.237 | Improved |
| also warp temporal RoPE (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.237 -> 0.285 | Hurt |
| add spatial RoPE warping (ablation) | Qwerty on CogVideoX · VIPSeg object control | FVD (x10^3): 1.578 -> 0.668 | Improved |
| add self-guided latent optimization to query-warped denoising (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.274 -> 0.237 | Improved |
| add self-guided latent optimization to query-warped denoising (ablation) | Qwerty on CogVideoX · VIPSeg object control | FTD: 0.501 -> 0.473 | Improved |
| add query-warped denoising to latent optimization (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.355 -> 0.237 | Improved |
| add phase-consistency loss (ablation) | Qwerty on Wan · DL3DV camera control | FTD: 0.285 -> 0.237 | Improved |
| apply query warping to Wan without latent optimization | Wan 2.2 TI2V-5B · DL3DV camera control | FTD: 0.367 -> 0.274 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SG-I2V — SG-I2V: Self-Guided Trajectory Control in Image-to-Video Generation (ICLR 2025) · better on 4 of 4 shared results
- FreeTraj — FreeTraj: Tuning-Free Trajectory Control via Noise Guided Video Diffusion (IJCV 2026) · better on 1 of 1 shared result
- MotionClone* — MotionClone: Training-Free Motion Cloning for Controllable Video Generation (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Go-with-the-Flow: Motion-Controllable Video Diffusion Models Using Real-Time Warped Noise (CVPR 2025 · similarity 0.90)
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.90)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.90)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.90)
- Motion-Zero: A Zero-Shot Trajectory Control Framework of Moving Object for Diffusion-Based Video Generation (AAAI 2025 · similarity 0.90)
- Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model (ICLR 2025 · similarity 0.90)
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.
