Signed Rectified Flow enables negativity-controlled AI generation

Researchers have introduced Signed Rectified Flow (Signed RF), a novel generative modeling technique that extends Rectified Flow by targeting a signed measure. This method allows for the promotion of desired distributions while simultaneously suppressing unwanted ones, offering a principled way to incorporate negative information and exclusion constraints. Signed RF has demonstrated improvements in image generation fidelity and diversity on ImageNet, reduced similarity in anti-memorization tests, and decreased unwanted content generation in Stable Diffusion 3.5, all while maintaining aesthetic and CLIP scores. AI

IMPACT Introduces a new method for controlling generative models, potentially improving content filtering and data diversity.

RANK_REASON The cluster contains a research paper detailing a new generative modeling technique. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Hugging Face
• ImageNet
• Rectified Flow
• Signed Rectified Flow
• Stable Diffusion 3.5

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →