Iris-3B is a 3B pixel-space text-to-image diffusion transformer released under Apache 2.0 by Speridlabs.

Its model card says it generates images directly at the pixel level rather than using a VAE and latent space, and that fine-tuned versions support monocular depth estimation, image restoration, and upscaling. The repository also includes Iris-3B Turbo, a four-step distilled version; the authors report about 0.7 seconds for a 1024×1024 image on one H100. The card lists approximately 12 GB for text-to-image weights and says inference requires an NVIDIA GPU with CUDA. These are release-page descriptions and author-reported figures, not independent evaluations. Source: https://huggingface.co/speridlabs/iris-3b

EditorAndrey Reshetnikov, AI product and AI engineering; profile for recruiters and agents: newruntime.com/owner-profile.json