Signed Rectified Flow Enables Negativity-Controlled Generative AI.

Runlong Liao, Baiyu Su, Lizhang Chen, Qiang Liu· July 22, 2026 View original

Summary

Researchers introduce Signed Rectified Flow (Signed RF), a new generative model that can promote desired distributions while actively suppressing undesired ones by targeting a signed measure. This framework allows for incorporating negative information and exclusion constraints, improving fidelity-diversity trade-offs and reducing unwanted content in models like Stable Diffusion.

A novel generative modeling technique called Signed Rectified Flow (Signed RF) has been introduced, extending the concept of Rectified Flow to handle signed measures. This method allows generative models to not only learn to produce desired outputs (positive distribution $\pi^+$) but also to actively avoid undesired outputs (negative distribution $\pi^-$) by concentrating probability in positive regions and excluding negative ones. The framework provides a principled way to integrate negative information and exclusion constraints into generative processes. The underlying theory, based on a signed continuity equation and a charged-particle interpretation, explains how "negative mass" creates exclusion barriers. Across various applications, Signed RF demonstrates significant improvements. It enhances the fidelity-diversity trade-off in ImageNet generation, reduces nearest-neighbor similarity in anti-memorization tasks, and effectively mitigates nudity generated by adversarial prompts in Stable Diffusion 3.5 while maintaining CLIP and aesthetic scores.

Why it matters

Professionals developing or deploying generative AI models can use Signed RF to achieve greater control over generated content, improving model safety, reducing bias, and enhancing output quality by explicitly defining what *not* to generate.

How to implement this in your domain

  1. 1Investigate integrating Signed Rectified Flow into your generative AI pipelines to enforce exclusion constraints.
  2. 2Experiment with defining negative distributions ($\pi^-$) to suppress unwanted content or biases in your generated outputs.
  3. 3Apply Signed RF to improve the fidelity-diversity trade-off in your image or data generation tasks.
  4. 4Utilize this technique for content moderation or safety filters in AI-powered creative tools.

Who benefits

Media & EntertainmentAdvertisingE-commerceAI SafetyContent Creation

Key takeaways

  • Signed Rectified Flow allows generative models to actively suppress undesired outputs.
  • It provides a principled way to incorporate negative information and exclusion constraints.
  • The method improves fidelity-diversity trade-offs and reduces unwanted content like nudity.
  • This technique offers enhanced control over generative AI outputs for safety and quality.

Original post by Runlong Liao, Baiyu Su, Lizhang Chen, Qiang Liu

"arXiv:2607.18516v1 Announce Type: new Abstract: We introduce Signed Rectified Flow (Signed RF), a generalization of Rectified Flow that targets the signed measure $\pi^{sign} = (1+\alpha)\pi^+ - \alpha\pi^-$, where $\alpha>0$, $\pi^+$ is the distribution to promote, and $\pi^-$ i…"

View on X

Originally posted by Runlong Liao, Baiyu Su, Lizhang Chen, Qiang Liu on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI Research

New Tool Generates Contamination-Resistant, Labeled Code Datasets for LLMs

Spaghetti Architect is a new open-source tool that generates controlled, multi-language code datasets, addressing issues of contamination and lack of semantic control in existing code corpora. It creates correct-by-construction programs with adjustable "messiness" and difficulty labels, making it ideal for training and evaluating code-generating LLMs.

Yuxiang JiJul 22, 2026
AI ResearchAI Engineering & DevTools

New Method Safely Gates Hazardous LLM Knowledge Without Deletion

Researchers introduce Token Inoculation, a method that allows large language models to retain sensitive "dual-use" knowledge while selectively refusing hazardous queries. This approach uses a special token to condition the model's behavior, improving safety without sacrificing benign domain performance.

Seunghyun Lee, Dongyoon Han, Sangdoo YunJul 22, 2026
AI ResearchAI Engineering & DevTools

GNNAS-TSP Selects Optimal Algorithms for Traveling Salesman Problem

Researchers introduce GNNAS-TSP, a Graph Neural Network (GNN)-based framework for automated algorithm selection (AS) for the Traveling Salesman Problem (TSP). GNNAS-TSP learns TSP instance representations directly from raw graph data, avoiding manual feature engineering, and formulates AS as a joint cost-prediction and ranking task to select the best solver from a portfolio under fixed computational budgets.

Zhaoxuan Li, Jiale Yang, Yifei Lu, Mustafa MisirJul 22, 2026