Diff-Instruct with Diffused Reward (DIDR)

One-step text-to-image generation with SDXL and Z-Image (NeurIPS 2026). The interactive demo is coming soon.