Research map / Video / Video generation
Consistent and Controllable Image Animation With Linear Motion Diffusion Transformers
Xin Ma, Yaohui Wang, Genyun Jia, Xinyuan Chen, Tien‐Tsin Wong, Cunjian Chen
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 motion residual learning instead of direct frame prediction (ablation) | MiraMo · VBench I2V | I2V Average Consistency: 92.62 -> 94.02 | Improved |
| add motion residual learning instead of direct frame prediction (ablation) | MiraMo · VBench I2V | Average Consistency: 93.62 -> 94.53 | Improved |
| add motion residual learning instead of direct frame prediction (ablation) | MiraMo · VBench I2V | Dynamic Degree: 32.68 -> 35.12 | Improved |
| add DCTInit noise refinement on top of MRL (ablation) | MiraMo · VBench I2V | I2V Average Consistency: 94.02 -> 95.47 | Improved |
| add FFT noise refinement on top of MRL (ablation) | MiraMo · VBench I2V | I2V Average Consistency: 94.02 -> 94.93 | Improved |
| use DCT instead of FFT for noise refinement (ablation) | MiraMo · VBench I2V | I2V Average Consistency: 94.93 -> 95.47 | Improved |
| use DCT instead of FFT for noise refinement (ablation) | MiraMo · VBench I2V | Dynamic Degree: 33.90 -> 35.93 | Improved |
| add FFT noise refinement on top of MRL (ablation) | MiraMo · VBench I2V | Dynamic Degree: 35.12 -> 33.90 | Hurt |
| use synthetic (Wan2.1/HunyuanVideo) videos instead of real data for post-training (ablation) | MiraMo T2V · VBench T2V | Multiple Object: 52.91 -> 45.50 | Hurt |
| use synthetic videos instead of real data for post-training (ablation) | MiraMo T2V · VBench T2V | average VBench score: over a 3% improvement | Improved |
| use dependent residuals z_i - z_{i-1} instead of residuals to first frame (ablation) | MiraMo · VBench I2V | video quality | Hurt |
| use frame-wise temporal attention instead of full 3D attention over video tokens (ablation) | MiraMo T2V · T2V | temporal consistency | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VideoCrafter2 — VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models (CVPR 2024) · better on 1 of 1 shared result
- LaVie — LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models (IJCV 2024) · 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
- Consistent and Controllable Image Animation with Motion Diffusion Models (CVPR 2025 · similarity 0.91)
- AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning (ICLR 2024 · similarity 0.90)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.89)
- Animate Anyone: Consistent and Controllable Image-to-Video Synthesis for Character Animation (CVPR 2024 · similarity 0.89)
- MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model (CVPR 2024 · similarity 0.88)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 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.
