Text To Image Generation With Rectified Flow Transformers

via “latent-space text-to-image generation with flow matching”

text-to-image model by undefined. 6,84,555 downloads.

Unique: Uses flow-matching formulation instead of traditional DDPM/DDIM noise schedules, enabling faster convergence and better sample quality with fewer steps; implements joint text-image transformer attention rather than cross-attention-only designs, improving semantic alignment and reducing prompt misinterpretation

vs others: Faster inference than Stable Diffusion 3 (2-3x speedup) with comparable or better quality; more open and self-hostable than DALL-E 3 or Midjourney; better prompt following than SDXL due to improved text encoder and flow-matching training