MLX-Gen
A local image and video generation runtime for Apple Silicon and MLX, covering text and image-to-image, instruction-based editing, and text, image, and reference-to-video generation and restoration.
MLX-Gen is a local generative image and video runtime for Apple Silicon, forked from the mflux project. It exposes an mlxgen command for text-to-image and image-to-image generation, instruction and reference-guided edits, masked editing, text, image, and reference-to-video generation and editing, and image and video restoration and upscaling through the SeedVR2 and SwiftVR models.
Supported text-to-image models include Qwen Image, FLUX.2 Klein, Z-Image, ERNIE Image Turbo, Bonsai Image, and FIBO, along with optimized quantized variants published specifically for MLX-Gen that keep precision-sensitive layers at higher precision instead of uniformly quantizing every layer. Models are downloaded or prepared locally before generation, so jobs run only against files already on disk, which the project positions as suited to desktop apps and long-running local jobs.
The fork adds tested text-to-image and image-to-image routes, Qwen Image Edit support, Wan2.2 text-to-video and image-to-video support, and reference-guided video generation and editing, while aiming to stay mergeable back into the upstream mflux project.