Back to Model
M

MOSS-Image-SFT MLX

Model Oct 18, 2026

A community Apple Silicon port of MOSS-Image-SFT with BF16, 8-bit and 4-bit MLX weights and a fork-based mflux workflow.

Visit primary link 2 sources attached
Author-generated MOSS-Image-SFT watchmaker-workbench samples compare BF16, 8-bit and 4-bit MLX output side by side.
Contributor-published generated samples, with the full three-column comparison preserved. Memory labels describe the image Transformer, not the complete pipeline; this is not an AITrending reproduction or proof of general quality parity. Source: vanch007 / MOSS-Image-SFT MLX
multimodalself-hosted

What happened lately

MOSS-Image-SFT MLX timeline

This community conversion brings MOSS-Image-SFT to Apple MLX through a contributor’s mflux fork. Its model card labels the weights Apache-2.0 and offers BF16, 8-bit and 4-bit options. The implementation proposal opened October 17, and the 4-bit model repository was created October 18.

The implementation was still an open, unmerged mflux pull request when reviewed. Users are directed to the contributor’s feature/moss-image-sft branch. The advertised 2.66GB figure covers the 4-bit image Transformer, not the separate text encoder, VAE or complete application’s unified-memory use. Performance and quality comparisons are the contributor’s reports; AITrending has not run the port.