xiejunjie524
Featured
handdraw-story-video
A tool that converts 7–9 hand-drawn story keyframes into a 35–45 second vertical video with progressive line art and color animation, configurable via JSON and renders with HyperFrames and GSAP.
DistillAlign
DistillAlign is a method that balances mode-covering and mode-seeking objectives in autoregressive video distillation, achieving superior results with a 1.3B teacher compared to baselines using a 14B teacher.
Stars
54
7-day growth
No data
Forks
1
Open issues
0
License
Apache-2.0
Last updated
2026-07-30
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It demonstrates that a carefully aligned smaller teacher can outperform larger teachers in video distillation, and provides a fully open-source release including checkpoints, training data, and an evaluation pipeline.
Who it is for
- Researchers working on video generation and diffusion models
- Practitioners interested in model distillation and compression
- Developers of autoregressive video architectures
- Engineers seeking efficient video synthesis pipelines
Use cases
- Compressing large video generation models (e.g., Wan2.1-14B) into smaller student models
- Benchmarking distillation techniques for autoregressive video models
- Generating high-quality videos with reduced computational cost
- Reproducing and extending state-of-the-art video distillation results
Strengths
- Outperforms baselines refined with a 14B teacher while using only a 1.3B teacher
- Provides complete open-source release: code, checkpoints, training data, and evaluation tools
- Includes a teacher-normalized distribution evaluation pipeline using V-JEPA2 features
- Supports both single and batch inference with clear commands
Considerations
- Requires substantial GPU memory (24GB for 1.3B student, 80GB for native 14B teacher) and 128GB system RAM for loading the 14B teacher
- Currently focused on the Wan2.1 architecture; generalization to other video models is not demonstrated
- The evaluation protocol relies on a fixed prompt/seed grid and V-JEPA2 encoder, which may not capture all aspects of video quality
README quick start
DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation
Jiaxing Li1,2*, Kai Zou1*, Cindy Zhou1,3, Kaichen Huang1,2, Junyao Gao1, Zile Wang1,
Yang Liu1, Bin Liu1, Bo An2, Yangguang Li1†
1Riemann Dynamics 2Nanyang Technological University 3Wellington College, UK
*Equal contribution †Corresponding author
DistillAlign aligns and balances the mode-covering and mode-seeking objectives of the multi-stage autoregressive video distillation pipeline — using only a 1.3B DMD teacher, it already surpasses baselines refined with a 14B DMD teacher.
Columns · Wan2.1-1.3B | Self-Forcing | Causal-Forcing | Ours (1.3B teacher) | Ours (14B teacher)
🔥 News
- [2026-07] Initial release: inference code, released checkpoints, training data, and the full teacher-normalized distribution-evaluation pipeline.
🤗 Release Table
All released generators are Wan2.1-1.3B students; the size in each name refers to the teacher used during training.
Checkpoints
| Model | Checkpoint | Description |
|---|---|---|
| Initializer (1.3B teacher) | 🤗 Huggingface | Pre-DMD initializer, trained toward a Wan2.1-T2V-1.3B teacher |
| Initializer (14B teacher) | 🤗 Huggingface | Pre-DMD initializer, trained toward a Wan2.1-T2V-14B teacher |
| Distilled (1.3B teacher) | 🤗 Huggingface | Final joint-distilled generator, Wan2.1-T2V-1.3B DMD teacher |
| Distilled (14B teacher) | 🤗 Huggingface | Final joint-distilled generator, Wan2.1-T2V-14B DMD teacher |
Distillation data (prompts curated from VidProM, 25K samples each):
| Data | Download | Description |
|---|---|---|
| Wan2.1-14B data | 🤗 Huggingface | 25K distillat |
Description
DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation
Related repositories
Similar projects matched by category, topics, and programming language.
bytec-ai
Featured
book-video-factory
A portable multi-account short video production skill for books that manages accounts, scripts, images, voiceovers, subtitles, and final exports through a unified workspace.
Krishnagangwal
Featured
CS-Fundamentals
A curated collection of Computer Science fundamentals (PDFs, notes, cheatsheets, interview question banks) for placement preparation, covering seven core subjects plus general resources.