Skip to content
Bedrova bedrova
FeaturesCompatibilityDevelopersCompareDocs Coming soon

Docs

Image generation

Generate images locally with FLUX, Stable Diffusion, Qwen Image and more on Apple Silicon.

Bedrova generates images entirely on your Mac — nothing is sent to the cloud.

Screenshot: the image generation view with the gallery — added at launch.

Models

Bedrova runs image models through two engines behind one interface, and picks the right one for the model you load:

  • MLX / mflux — FLUX.1, FLUX.2 Klein, Qwen Image, Krea 2, ERNIE Image, Z-Image, Ideogram 4 and FIBO.
  • A native diffusion engine — the Stable Diffusion ecosystem: SD1.5 and SD2, SDXL, SD3.5, and Chroma. This engine also brings SD1.5 ControlNet and image-to-image and inpainting on the profiles that support them.

LoRA adapters work where the adapter matches the exact base architecture and variant. Supplying an adapter a provider cannot apply is an error, never something Bedrova silently ignores.

What “supported” means here

Support is profile-specific, not a generic checkpoint promise. A given SD family is enabled only once its exact composition, tensor and config signatures, licence chain, cancellation and unload behaviour have been proved on real hardware. Anything unproved stays visible but disabled, with the reason shown, rather than failing mid-generation.

You bring your own Stable Diffusion weights. Most common SD checkpoints ship under OpenRAIL, community or noncommercial terms that Bedrova will not redistribute or recommend, so they are user-governed imports rather than curated picks. FLUX and Qwen Image families are the curated defaults.

Architectures Bedrova recognises but does not run: Hunyuan Image, PixArt, Kolors, Lumina, Stable Cascade, AuraFlow and SeedVR2. Each needs its own adapter and licence study before it could be enabled, and none enters through an optimistic fallback.

Controls

Prompt and negative prompt, dimensions, steps, guidance, scheduler, seed (for reproducibility), and batch count. Bedrova shows the text-encoder/transformer retention policy and tracks peak memory so a big run won’t blow your budget; you can cancel mid-run.

API

Generate from your own apps via POST /v1/images/generations (and image-to-image / edits via the native /api/v1/images/* routes). Generated images and their metadata are saved to a local gallery you control.