This subsystem routes typed commands for image generation through GenerativeModels.*. The default recommended image path is FLUX.1 Schnell because it is fast, open-weight, and practical for a server-style _AI worker.

Commands

[text]
generative recommendations
generative flux_schnell generate <prompt> [output_path]
simple image generate <prompt> [output_path]
simple image flux_schnell <prompt> [output_path]
fluz_shnell generate <prompt> [output_path]
image generate <prompt> [output_path]

fluz_shnell is accepted as a typo-compatible alias. If no remote image worker is enabled, the command writes a deterministic local preview artifact so command routing and training pipelines can still be tested without CUDA.

Recommendations

• flux_schnell: black-forest-labs/FLUX.1-schnell; default for fast text-to-image generation.
• sdxl: stabilityai/stable-diffusion-xl-base-1.0; broad compatibility with Diffusers workflows.
• stable_diffusion_3_5: stabilityai/stable-diffusion-3.5-large; higher quality when hardware allows.
• flux_dev: black-forest-labs/FLUX.1-dev; development and non-commercial experimentation.
• local_preview: deterministic fallback for offline tests and command validation.

Optional Remote Worker

Start the worker on the machine that should run image inference:

[powershell]
python scripts\flux_schnell_worker.py –host 0.0.0.0 –port 7720

Then enable remote inference in _AI:

[text]
hardware remote on
generative remote on
generative flux_schnell generate neon minecraft castle

The worker accepts the generative_flux_schnell_sample operation. If Python packages or model weights are not installed, the worker returns a small deterministic PPM preview rather than failing the command path.

---

Filed under: Uncategorized - @ July 18, 2026 6:37 am