Sperid Labs Open-Sources Iris-3B, Challenging Pixel-Space Diffusion

Sperid Labs released Iris-3B, a 3B-parameter text-to-image model trained from scratch in pixel space without a VAE, open-sourced under Apache 2.0. Their experiments show pixel-space diffusion does not outperform latent-space models.

2026-10-09 ~ 2026-10-09 · 2 related posts

• Iris-3B: pixel-space diffusion offers no edge over latent models, study finds — speridlabs · 2026-10-09
• Iris-3B: a 3B pixel-space text-to-image model with no VAE, fully open under Apache 2.0 — zhenjun_zhao · 2026-10-09