The developers also applied the model's generative prior to detail-critical vision tasks, specifically image restoration and depth estimation. These methods are detailed in a research paper titled 'Going Beyond the Latent with Pixel-Space Diffusion Training, Conversion and Fine-Tuning' released alongside the model.
Wednesday, October 21, 2026
Story
Sperid Labs Releases Iris-3B Open 3B Parameter Image Model Without VAE
AI12 days ago
A new framework from Sperid Labs produces visual data by bypassing traditional latent stages. Known as Iris-3B, the 3B parameter text-to-image model was pre-trained from scratch and is available under an Apache 2.0 license. The organization has published the model's weights, source code, and a demonstration for public use.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.