Research map / Video / Video generation
Anchor Frame Bridging for Coherent First-Last Frame Video Generation
Xuehan Hou, Meng Fan, Pengchong Qiao, Zesen Cheng, Yian Zhao, Lei Zhu, Kaiwen Cheng, Chang Liu, Jie 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 AFB anchor frame to Wan2.1 first-last frame generation | Wan2.1-FLF2V · Own FLF2V dataset (436 pairs, 5s) | FVD: 413.68 -> 375.12 | Improved |
| add AFB anchor frame to Wan2.1-I2V | Wan2.1-I2V · Own FLF2V dataset (436 pairs, 5s) | PSNR: 32.13 -> 35.41 | Improved |
| add AFB to HunyuanVideo-I2V | HunyuanVideo-I2V · Own FLF2V dataset (436 pairs, 5s) | FVD: 496.32 -> 435.71 | Improved |
| use a single anchor frame rather than two (ablation) | Wan2.1 + AFB · Own FLF2V dataset (436 pairs, 5s) | FVD: 386.94 -> 375.12 | Improved |
| use a single anchor frame rather than three (ablation) | Wan2.1 + AFB · Own FLF2V dataset (436 pairs, 5s) | PSNR: 30.49 -> 35.41 | Improved |
| increase reverse-denoising stop step K from 5 to 50 (ablation) | Wan2.1 + AFB · Own FLF2V dataset (436 pairs, 5s) | FVD: 412.73 -> 375.12 | Improved |
| stop at K=15 instead of full reverse pass (ablation) | Wan2.1 + AFB · Own FLF2V dataset (436 pairs, 5s) | FVD: 388.45 vs 375.12 (+35% time) | Mixed |
| use Qwen-generated prompts instead of a general prompt (ablation) | Wan2.1 + AFB · Own FLF2V dataset (436 pairs, 5s) | FVD: 475.33 -> 375.12 | Improved |
| add AFB under a general text prompt (ablation) | Wan2.1 · Own FLF2V dataset (436 pairs, 5s) | FVD: 486.23 -> 475.33 | Improved |
| add AFB versus FCVG on 5s videos | FCVG · Own FLF2V dataset (436 pairs, 5s) | FVD: 438.83 -> 375.12 | Improved |
| add AFB versus MoG on 5s videos | MoG · Own FLF2V dataset (436 pairs, 5s) | PSNR: 18.41 -> 35.41 | Improved |
| average anchor latent with noise latent at each step instead of direct conditioning (ablation) | Wan2.1 + AFB · qualitative examples | visual quality | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViBiDSampler — ViBiDSampler: Enhancing Video Interpolation Using Bidirectional Diffusion Sampler (ICLR 2025) · better on 3 of 3 shared results
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
- Transforming Image Editors into Video Editors (NeurIPS 2026 · similarity 0.87)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 2026 · similarity 0.86)
- Generative Inbetweening through Frame-wise Conditions-Driven Video Generation (CVPR 2025 · similarity 0.86)
- CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025 · similarity 0.86)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.85)
- Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion (AAAI 2025 · similarity 0.85)
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.
