FLUX 3 Image Klein API Guide: Five Resolution Tiers, Parameters and Examples

What FLUX 3 Image Klein is

FLUX 3 Image Klein is the lightweight image model in Black Forest Labs' FLUX 3 family. On HiAPI it has a single model ID, flux-3-image-klein, for both text-to-image and reference-based editing. A prompt alone gives you a new image. Add image_urls and it edits from the references or blends the styles of several images.

Its strengths are low cost and a wide range of tiers. Five resolutions from 768 up to 4K make it a good fit for batch generation and for drafting small, then finishing large.

Pricing: per image, by resolution tier

• 768: $0.0092 per image
• 1K (default): $0.0103 per image
• 1.5K: $0.0126 per image
• 2K: $0.016 per image
• 4K: $0.0412 per image

Reference images are free, so one reference costs the same as ten. Lowering steps doesn't change the price either; only resolution does. Failed tasks are refunded in full.

For example, 50 style drafts at 768 cost $0.46, and 5 finals at 4K cost $0.206.

Parameters

Everything except model and prompt is optional.

• prompt: required, 1–5,000 characters.
• resolution: 768, 1K, 1.5K, 2K or 4K. Defaults to 1K.
• aspect_ratio: 16 values: 21:9, 2:1, 16:9, 3:2, 7:5, 4:3, 5:4, 1:1, 4:5, 3:4, 5:7, 2:3, 9:16, 1:2, 9:21 and auto (default).
• image_urls: 1–10 publicly reachable reference images. Omit it for pure text-to-image.
• steps: 1–16, default 16. Fewer steps are faster but may lose some detail.
• seed: 0–4294967295. A fixed seed with the same parameters aims to reproduce a similar result; omit it for a random seed.
• safety_tolerance: 0–4, default 2. Higher values moderate less strictly.
• output_format: jpeg (default) or png.

Making a request

Image tasks are asynchronous: you create a task, then poll for the result. model goes at the top level and everything else goes inside input.

Text-to-image at 2K, 4:5 portrait, with a fixed seed:

curl -X POST https://api.hiapi.ai/v1/tasks \
  -H "Authorization: Bearer $HIAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "flux-3-image-klein",
    "input": {
      "prompt": "A ceramic coffee cup on a wooden table in morning light, shallow depth of field, warm tones, product photography",
      "resolution": "2K",
      "aspect_ratio": "4:5",
      "seed": 42
    }
  }'

That image costs $0.016.

Editing with two references to blend styles:

{
  "model": "flux-3-image-klein",
  "input": {
    "prompt": "Keep the building structure from the first image and repaint it with the watercolor strokes and palette of the second",
    "image_urls": [
      "https://example.com/building.jpg",
      "https://example.com/watercolor.jpg"
    ],
    "resolution": "1.5K",
    "output_format": "png"
  }
}

Once the task is created, poll GET https://api.hiapi.ai/v1/tasks/{taskId} until it succeeds, then take the image URL from the output.

Tips

• Draft small, finish large: test prompts at 768 or 1K with a fixed seed. Once you like the result, change only resolution to 2K or 4K and run it again.
• Batch variations: keep the prompt and change seed to get a set of candidates with different compositions and a consistent style.
• When editing, say what to keep: with references, stating what stays and what changes is more reliable than describing only the target image.

Current limits

• Each call returns one image.
• Up to 10 reference images, which must be public http(s) URLs.
• There is no transparent-background option; png is simply a lossless format.