Tencent-Hunyuan/SRPO

• hightopics#1

  Add relevant topics to the repository

  Why:

  COPY-PASTE FIX

  Copy

  diffusion-models, human-preference, reinforcement-learning-from-human-feedback, rl-from-human-feedback, generative-ai, image-generation, text-to-image, preference-optimization, deep-learning, srpo

  OpenDoneDismiss
• highreadme#2

  Add a concise introductory sentence to the README

  Why:

  CURRENT

  <div align=“center” style=“font-family: charter;”>
  <h1 align="center">Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference </h1>
  <div align="center">
    <a href='https://arxiv.org/abs/2509.06942'></a>  &nbsp;
    <a href='https://huggingface.co/tencent/SRPO/'></a> &nbsp;
    <a href='https://tencent.github.io/srpo-project-page/'></a> &nbsp;
  </div>
  <div align="center">
    Xiangwei Shen<sup>1,2,3*</sup>,
    <a href="https://scholar.google.com/citations?user=Lnr1FQEAAAAJ&hl=zh-CN" target="_blank"><b>Zhimin Li</b></a><sup>1*</sup>,
    <a href="https://scholar.google.com.hk/citations?user=Fz3X5FwAAAAJ" target="_blank"><b>Zhantao Yang</b></a><sup>1</sup>,
    <a href="https://shiyi-zh0408.github.io/" target="_blank"><b>Shiyi Zhang</b></a><sup>3</sup>,
    Yingfang Zhang<sup>1</sup>,
    Donghao Li<sup>1</sup>,
    <br>
    <a href="https://scholar.google.com/citations?user=VXQV5xwAAAAJ&hl=en" target="_blank"><b>Chunyu Wang</b></a><sup>1✝</sup>,
    <a href="https://openreview.net/profile?id=%7EQinglin_Lu2" target="_blank"><b>Qinglin Lu</b></a><sup>1</sup>,
    <a href="https://andytang15.github.io" target="_blank"><b>Yansong Tang</b></a><sup>3,✉️</sup>
  </div>

  COPY-PASTE FIX

  Copy

  <div align=“center” style=“font-family: charter;”>
  <h1 align="center">Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference </h1>
  SRPO (Strong-to-weak Preference Optimization) is a novel framework for aligning diffusion models with fine-grained human preferences, enhancing output quality through direct trajectory optimization.
  <div align="center">
    <a href='https://arxiv.org/abs/2509.06942'></a>  &nbsp;
    <a href='https://huggingface.co/tencent/SRPO/'></a> &nbsp;
    <a href='https://tencent.github.io/srpo-project-page/'></a> &nbsp;
  </div>
  <div align="center">
    Xiangwei Shen<sup>1,2,3*</sup>,
    <a href="https://scholar.google.com/citations?user=Lnr1FQEAAAAJ&hl=zh-CN" target="_blank"><b>Zhimin Li</b></a><sup>1*</sup>,
    <a href="https://scholar.google.com.hk/citations?user=Fz3X5FwAAAAJ" target="_blank"><b>Zhantao Yang</b></a><sup>1</sup>,
    <a href="https://shiyi-zh0408.github.io/" target="_blank"><b>Shiyi Zhang</b></a><sup>3</sup>,
    Yingfang Zhang<sup>1</sup>,
    Donghao Li<sup>1</sup>,
    <br>
    <a href="https://scholar.google.com/citations?user=VXQV5xwAAAAJ&hl=en" target="_blank"><b>Chunyu Wang</b></a><sup>1✝</sup>,
    <a href="https://openreview.net/profile?id=%7EQinglin_Lu2" target="_blank"><b>Qinglin Lu</b></a><sup>1</sup>,
    <a href="https://andytang15.github.io" target="_blank"><b>Yansong Tang</b></a><sup>3,✉️</sup>
  </div>

  OpenDoneDismiss
• mediumreadme#3

  Add a clear statement about the repository's license in the README

  Why:

  COPY-PASTE FIX

  Copy

  ## License
  This project is licensed under [Specify the actual license(s) here, e.g., a custom license, or a combination of licenses if applicable, as found in the LICENSE file]. Please refer to the [LICENSE](LICENSE) file for full details.

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

  How can I improve diffusion model output quality by incorporating fine-grained human preferences?

  you: not recommended
  AI recommended (in order):

  1. OpenAI's TRPO/PPO with Human Feedback (RLHF)
  2. Hugging Face's PEFT (Parameter-Efficient Fine-Tuning) with RLHF
  3. LoRA (Low-Rank Adaptation)
  4. DeepMind's InstructDiffusion
  5. LAION's Aesthetic Predictor
  6. DreamFusion
  7. Score Distillation Sampling (SDS)
  8. SJC (Score Jacobian Chaining)
  9. Krita
  10. Photoshop
  11. InvokeAI
  12. Automatic1111's Stable Diffusion web UI

  AI recommended 12 alternatives but never named Tencent-Hunyuan/SRPO. This is the gap to close.

  Show full AI answer
• CATEGORY QUERY

  What methods exist for aligning generative AI models with detailed human feedback for better results?

  you: not recommended
  AI recommended (in order):

  1. Direct Preference Optimization (DPO)
  2. Constitutional AI (CAI)
  3. Alpaca
  4. Vicuna

  AI recommended 4 alternatives but never named Tencent-Hunyuan/SRPO. This is the gap to close.

  Show full AI answer

• Metadata completeness

  warn

  Suggestion:

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of Tencent-Hunyuan/SRPO?

  pass
  ✓AI named Tencent-Hunyuan/SRPO explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• If a team adopts Tencent-Hunyuan/SRPO in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI named Tencent-Hunyuan/SRPO explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• In one sentence, what problem does the repo Tencent-Hunyuan/SRPO solve, and who is the primary audience?

  pass
  ✓AI named Tencent-Hunyuan/SRPO explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?