Open-source text-to-image model AuroraIMG2-100M released, GenEval 2 now near SDXL

Tall_Abrocoma_3533 · reddit · 2026-11-05

A new open-source text-to-image model, AuroraIMG2-100M, has been released with large improvements over its Preview version — GenEval 2 scores are now roughly in line with SDXL's. The model card includes a detailed explanation of training and architecture, a diffusers-compatible quick start, and samples. Licensed under Apache 2.0; the author welcomes feedback and criticism.

Related event: AuroraAI Releases Open-Source 100M-Parameter Text-to-Image Model AuroraIMG2-100M(2 posts)→

Original post →

More from Multimodal

• Kimo: a 15MB CPU-only MCP video editor with no GUI, built for AI agents — LieTop7965 · 2026-11-05
• World Labs demos 'spatial outpainting': extending an image into explorable 3D space — theworldlabs · 2026-11-05
• Fizgig 9.0.4: train Fibo 2.0 on a 12GB card, SAM 3.1 text-to-mask — shootthesound · 2026-11-05
• AI-generated retro sci-fi animation short "little doll" turns heads — superfatbeagle · 2026-11-05
• Graph-N: open-source C++/ImGui desktop client for ComfyUI handles thousands of nodes — elgeminy · 2026-11-05
• Open-source desktop AI studio Unplatform launches, ComfyUI node in the works — jefoce · 2026-11-05

Multimodal channel →