FLUX 1 Kontext (dev) (12 B params, FluxDev non-commercial license)

Open-weight, text-guided image editor that keeps your characters and style intact.

• One-prompt edits, zero drift. Specify a change (“add a hat,” “swap background”) and the rectified-flow transformer applies it while preserving character, style and object details—even across multiple successive tweaks.

• Heavy model, light options. Full-precision loads eat ~22 GB VRAM, but fp8 / 4-bit quant tricks bring it under ~10 GB—and community demos show it running on an 8 GB laptop GPU (slower but usable).

• Fast enough for UI. Providers like Fal & Replicate push 1024×1024 edits in < 2 s with 4-step pipelines; raw diffusers flows land in the 5-10 s range on a single A100.

• Drop-in tooling. Native FluxKontextPipeline in 🤗 Diffusers, ComfyUI node, ONNX exports—from_pretrained("black-forest-labs/FLUX.1-Kontext-dev"), pass an image + prompt, done.

• Built-in guardrails. Repository ships a safety checker and mandates content filters under the FluxDev NC license; you can’t resell hosted inference without a separate deal.

Why pick it for Norman AI?

Kontext dev lets us bolt a “magic-erase / quick-edit” feature onto our generative stack with open weights and sub-10 GB hardware—ideal for on-brand tweaks or end-user image editing. Just remember the license is research-only, so we’ll need the pro tier (or a separate agreement) before shipping this in paid products.

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