RL video generation research, sorted and auto-updated

What it does This repository is a curated awesome-list that catalogs research papers applying reinforcement learning—DPO, GRPO, PPO, and policy gradients—to video generation. An automated agent scrapes and analyzes new papers daily at 1:00 AM UTC+8, organizing them into a chronological table that lists base models, training strategies, and affiliations.

The interesting bit The project functions as a living digest rather than a static bibliography. The agent attempts to classify each paper’s contribution and extract base models like Wan2.1 or Qwen2.5-VL, which is ambitious given the weekly deluge of video-generation preprints.

Key highlights

• Tracks methods and benchmarks side-by-side in a single chronological table.
• Auto-updated daily by an agent that collects and analyzes the latest papers.
• Papers are tagged by contribution type: Method, Benchmark, or Survey.
• Includes direct arXiv and project-page links for each entry.
• 546 stars suggest the community was hungry for a centralized tracker.

Caveats

• The README explicitly warns that the automated paper analysis may not be accurate and is for reference only.
• Some table entries contain duplicate rows or minor formatting inconsistencies.
• It is a reading list, not a framework: there is no code, models, or reproducible pipelines here.

Verdict Worth bookmarking if you are actively researching RL fine-tuning for video diffusion or world models. Skip it if you are looking for training scripts or ready-to-run implementations.