FLUX 1 [dev] (12 B params, “FluxDev NC” license)

High-quality text-to-image model—slower than Schnell, sharper than SDXL.

• Spec sheet. 12 billion-parameter rectified-flow transformer, trained with guidance distillation; works best at 20-40 denoise steps for 1024 × 1024 frames.

• VRAM needs. Full-precision load is ~22 GB; fp8 / int-8 / 4-bit builds run on 8-12 GB cards (slower but doable).

• tool chain. Native support in Diffusers (FluxPipeline), ComfyUI nodes, ONNX and GGUF exports—from_pretrained("black-forest-labs/FLUX.1-dev") and go.

• License heads-up. Weights are non-commercial; you may sell the pictures you generate, but you can’t resell hosted inference without a separate deal.

Why pick it for Norman AI?

FLUX-1 Dev gives us SDXL-level image quality for a “premium” tier in our graphics pipeline—just keep the license limits in mind before we ship paid endpoints.

response = await norman.invoke(
    {
        "model_name": "stable-diffusion-v1-5",
        "inputs": [
            {
                "display_title": "Prompt",
                "data": "A cat playing with a ball on mars"
            }
        ]
    }
)