AuroraIMG2-100M open-source image model ships with SDXL-level GenEval 2 scores
Tall_Abrocoma_3533 · reddit · 2026-11-05
AuroraAI-Research released AuroraIMG2-100M, a tiny open-source text-to-image model with large improvements over its Preview version — GenEval 2 scores now roughly on par with SDXL despite only 100M parameters. The model card includes detailed training and architecture notes, a diffusers-compatible quick start and samples, all under Apache 2.0.
More from Multimodal
- Isometric SF: a 4B-parameter model paints the entire 1,300-square-mile Bay Area in pixel art — repligate · 2026-11-05
- Image comparison: nano banana looks like a camera roll while ChatGPT Images 2.6 still screams AI slop — flowersslop · 2026-11-05
- GPT Image 2.6 hits Arena: nearly 2x faster, Sunburst climbs to #2 in text-to-image — karinanguyen · 2026-11-05
- Douyin account 'Divine Beast Kitchen' goes viral: AI videos of mythical beasts cooking for gods — xiaohu · 2026-11-05
- AI video imagines year 2126: monopoly, containment or abundance after labor — No_Twist_678 · 2026-11-05
- GPT Image 2.6 complexity parameter tested: low/medium/high GIF generation compared frame by frame — gabrielchua · 2026-11-05