FLUX 3 Image

FLUX 3 Image is Black Forest Labs' unified still-image endpoint: one model handles text-to-image, edits, and compositions from up to 10 reference images. OpenTryOn exposes it as flux3-image on the generate and edit services (same adapter class, same BFL_API_KEY). For the faster distilled sibling, see FLUX 3 Image Klein (flux3-image-klein). For video, see FLUX 3 Video (flux3-video).

Prerequisites

BFL_API_KEY=your_bfl_api_key

Sign up and create a key in the BFL dashboard.

CLI

# Text-to-image
opentryon generate --model flux3-image \
  --prompt "Editorial fashion still, soft window light" \
  --aspect-ratio 3:4 --resolution 2k

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

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

Notable flags

Flag	Default	Notes
--aspect-ratio	auto	Many ratios plus auto (follows first reference or prompt)
--resolution	1k	768sq, 1k, 1.5k, 2k, 4k
--safety-tolerance	2	0 (strictest) through 4
--no-grounding	grounding on	Skip web/image search pre-step for faster runs
--reference-images	—	Edit only — extra refs beyond --image (10 total)

Layout control uses bounding boxes in the prompt (JSON on a 0–1000 grid); see the BFL bounding-box tutorial.

Python

from tryon.api import Flux3ImageAdapter

adapter = Flux3ImageAdapter()

images = adapter.generate_text_to_image(
    prompt="Minimal product photo on warm stone",
    aspect_ratio="16:9",
    resolution="1k",
)

images[0].save("outputs/flux3_t2i.png")

edited = adapter.generate_image_edit(
    prompt="Replace the background with a runway set",
    input_image="data/model-1.jpg",
    resolution="2k",
)
edited[0].save("outputs/flux3_edit.png")

MCP

After registry changes, restart the MCP server. Tools are generated from the registry:

• generate_flux3_image
• edit_flux3_image

Both unlock when BFL_API_KEY is set in the OpenTryOn .env (Studio Connect passthrough).

API details

• Submit: POST https://api.bfl.ai/v1/flux-3-image with x-key
• Poll: use the returned polling_url until status is Ready
• Download: result.sample (signed URL; download within about one hour)
• Rejected fields: FLUX.2-only fields such as width, height, and seed return HTTP 422 on this endpoint

Related

• FLUX.2 Image Generation — separate PRO/FLEX endpoints with pixel width/height
• Virtual try-on remains on the dedicated flux-vto model (vton service); see CLI.