FLUX 3 Image Klein

FLUX 3 Image Klein is Black Forest Labs' compact, open-weights sibling to FLUX 3 Image. The hosted API uses the same unified contract: text-to-image, edits, multi-reference composition (up to 10 images), and bounding-box layout prompts — without the full model's grounding field. OpenTryOn exposes it as flux3-image-klein on generate and edit (same BFL_API_KEY as FLUX.2 / FLUX 3 Image / FLUX 3 Video).

Prerequisites

BFL_API_KEY=your_bfl_api_key

Sign up and create a key in the BFL dashboard.

CLI

# Text-to-image (distilled default ~16 steps on the API)
opentryon generate --model flux3-image-klein \
  --prompt "Editorial fashion still, soft window light" \
  --aspect-ratio 3:4 --resolution 2k

# Reproducible output
opentryon generate --model flux3-image-klein \
  --prompt "Lookbook flat lay" --seed 42 --steps 12

# Edit with one reference
opentryon edit --model flux3-image-klein \
  --image data/model-1.jpg \
  --prompt "Change the blazer to deep burgundy velvet"

# Dry-run (no API call)
opentryon generate --model flux3-image-klein --prompt "test" --dry-run

Klein vs full FLUX 3 Image

	flux3-image-klein	flux3-image
Endpoint	/v1/flux-3-image-klein	/v1/flux-3-image
Grounding	Not supported	Optional (--no-grounding)
Safety tolerance	0–7	0–4
Extra knobs	--seed, --steps (1–16), --output-format	—
Exact pixels	--width + --height (overrides resolution/ratio)	Resolution tiers only

Layout control uses the same bounding-box JSON appended to the prompt; see the BFL bounding-box tutorial.

Python

from tryon.api import Flux3ImageKleinAdapter

adapter = Flux3ImageKleinAdapter()

images = adapter.generate_text_to_image(
    prompt="Editorial fashion still",
    aspect_ratio="3:4",
    resolution="2k",
    seed=0,
    steps=16,
)
images[0].save("outputs/flux3_klein_t2i.png")

edited = adapter.generate_image_edit(
    prompt="Change jacket to navy wool",
    input_image="data/model-1.jpg",
)
edited[0].save("outputs/flux3_klein_edit.png")

MCP

After the OpenTryOn MCP server loads the registry:

• generate_flux3_image_klein
• edit_flux3_image_klein

Local weights (GPU)

OpenTryOn registers local inference as flux3-image-klein-local on generate and edit (Flux3ImageKleinLocalAdapter, Diffusers ModularPipeline). See FLUX 3 Image Klein (local) for install, VRAM notes, and CLI examples. Default weights: black-forest-labs/flux-3-image-klein; base checkpoint via --model-id black-forest-labs/flux-3-image-klein-base.