AtlasML Releases Open Weights for High-Res Image Diffusion Model 'Atlas Fine'

AI · 5 min read

Atlas Fine supports image generation up to 8K and emphasizes color accuracy and halftone control for print workflows. By releasing weights under a permissive license, AtlasML aims to let creative studios run the model on-premises and integrate it into DTP pipelines.

The team also released a companion 'print-aware' sampler that preserves spot colors and avoids banding commonly introduced by naive upscaling. Sample notebooks show workflows for producing CMYK-ready files and embedding color profiles for commercial printers.

Privacy-conscious designers welcomed the open weights, but AtlasML cautioned that running Atlas Fine on commodity hardware requires significant VRAM or model sharding. The company partnered with several inference vendors to provide turnkey hosting for teams that prefer cloud-hosted instances.