Image generation
Make images on your Mac.
Generate images entirely on-device with FLUX, Stable Diffusion, Qwen Image and more. No cloud, no subscription, no rate limits.
What you can do
- Text-to-image — describe it, generate it locally
- Image-to-image & edits — transform and inpaint
- LoRA and ControlNet — your own styles, and structure control on SD1.5
- Seeds, steps, guidance — reproducible, tunable output + a gallery
Models
The latest open image models, MLX-accelerated.
- FLUX.1 and FLUX.2 Klein
- Stable Diffusion 1.5 and 2, SDXL, SD3.5
- Qwen Image, Chroma
- Krea 2, ERNIE Image, Z-Image
- Ideogram 4, FIBO
- LoRA adapters, SD1.5 ControlNet
Two engines under one interface: mflux for the FLUX and Qwen families, a native engine for the Stable Diffusion ecosystem. Bedrova picks the right one. Architectures it has not proved stay disabled with a reason. See the image docs.
How it works
Generate from the app, or from your own code.
POST localhost:1337/v1/images/generations
{ "model": "flux.1-schnell", "prompt": "a rover dog, warm flat vector", "size": "1024x1024" } Bedrova tracks peak memory and lets you cancel mid-run, so a big batch won't blow your budget. Generated images and metadata save to a local gallery you control.
FAQ
Can I really generate images locally on a Mac?
Yes. Bedrova runs FLUX.1, FLUX.2 Klein, Qwen Image, Krea 2, ERNIE Image, Z-Image, Ideogram 4 and FIBO through its MLX runtime, plus Stable Diffusion 1.5/2, SDXL and SD3.5 and Chroma through a second native engine. All of it on Apple Silicon, entirely on-device.
What do I need to run image models?
Most image models want somewhere between 9 and 24 GB of unified memory. FLUX.2 Klein 4B is comfortable on a 16 GB Mac. The fit-check tells you what fits before you load anything.
Can other apps generate images through Bedrova?
Yes. Call POST /v1/images/generations on the local API, or reach image-to-image and edits through the native routes.
Does it run Stable Diffusion?
Yes, for the SD1.5/2, SDXL and SD3.5 profiles Bedrova has actually proved, through a dedicated native engine rather than a FLUX fallback. Two honest caveats. Most common SD weights carry OpenRAIL or community licences that Bedrova will not redistribute, so you bring your own files rather than picking them from our curated list. And support is profile-specific, not a promise that any checkpoint you find will load. Anything unproved stays disabled with a reason instead of failing halfway through a generation.