FLUX 3 Klein API and prices compared

API Pricing Comparison: BFL, fal.ai, Replicate, and Together AI

FLUX 3 Klein (full name FLUX 3 Image Klein) is available through the Black Forest Labs (BFL) API, and fal.ai lists it too. Prices on the BFL API depend on the output size, up to 4K. Image editing costs the same per image as text-to-image generation.

Output size	BFL API	fal.ai (marked Planned)
768 px	$0.008	$0.008
1K	$0.009	$0.009
2K	$0.014	$0.014
4K	$0.036	$0.036

Generating 100 images at 1K resolution on the BFL API costs $0.90 ($0.009 multiplied by 100). At 4K resolution, 100 images cost $3.60. You can verify these numbers on the BFL pricing documentation and the fal.ai model page.

Other platforms handle hosting differently as of October 16, 2026:

• Replicate: Only a community LoRA version (replicate/flux-3-klein-image-lora) was found. It is billed by run time on an NVIDIA L40S GPU rather than a fixed per-image price.
• Together AI: Its “FLUX 3” listing is a video model; no Klein listing was found.

If you prefer a web interface over writing code, Klein Studio (an unofficial site not affiliated with or endorsed by BFL) runs the model through BFL’s API. You can generate 3 free 1K images per day on the home page without signing up, or check our Klein Studio pricing page for details on one-time image packs.

How the BFL API Call Works

Black Forest Labs uses an asynchronous two-step request pattern for FLUX 3 Image Klein. You submit a generation or editing job first, then poll a returned URL until the image is ready.

1. Submit the request: Send an HTTP POST request to https://api.bfl.ai/v1/flux-3-image-klein. Include the header x-key: <your key> for authentication, and pass a JSON body containing your prompt.
2. Get the polling URL: The immediate JSON response from the POST request does not contain the finished image. Instead, the answer contains a polling_url field.
3. Poll for completion: Send HTTP GET requests to that polling_url using the same x-key: <your key> header. Repeat the GET request until the status field in the response is Ready. Once the status is Ready, the result holds a link to the generated image.

Through the model’s feature set, you can generate images from text, edit images by instruction, combine up to 10 reference images, control layout with bounding-box coordinates, and render text inside images. Note that the Hugging Face model card states the model is meant to be used with a prompt upsampler (prompt expander), as short prompts used directly give mediocre results. Whether BFL’s API expands prompts itself is not confirmed.

API Terms vs. Open-Weight Licensing

Using BFL’s hosted API is governed by BFL’s API terms, not the open-weight license (this is our understanding; we have not reviewed the API terms in detail).

If you instead download the open weights to run the 5.6-billion-parameter image model and Qwen3-VL-4B text encoder on your own hardware, your usage falls under the FLUX Kommunity License v1.0. You can read a summary of those rules on our license overview page, and review local GPU requirements on our VRAM and hardware requirements page.