Flux 3 is Black Forest Labs' new generation of image models. Kie AI starts with Flux 3 Klein for fast text-to-image generation and multi-reference image editing up to 4K, with more Flux 3 models to follow.
Describe the image to generate, in any language (up to 5,000 characters).
0/5000Output image width. Both must be passed at the same time; JSON integers, each side ≥256 and ≤4096; aspect ratio between 1:2.5 and 2.5:1. Once passed, aspect_ratio and resolution are ignored.
Output image height. Both must be passed at the same time; JSON integers, each side ≥256 and ≤4096; aspect ratio between 1:2.5 and 2.5:1. Once passed, aspect_ratio and resolution are ignored.
Output aspect ratio.
Output resolution: 1K, 2K or 4K. Higher resolutions cost more and take longer.
Output image format.
Number of images to generate with the same settings.
Random seed. Omit for a random result; with the same seed and parameters the result is reproducible. When generating multiple images, image i uses seed + i. 0-2147483647
Rewrites the prompt into a fuller description before generating, which noticeably improves quality. Enabled by default; pass false to turn it off. Recommended to turn off when your request must be followed literally (exact text edits, "keep X unchanged and only change Y", fixed compositions) or when latency matters.
A configurable parameter. Defaults to true in the Playground.
Explore different use cases and parameter configurations
Describe the edit in plain language: what to change and what to keep. Refer to reference images by their order ("image 1", "image 2"). Up to 5,000 characters, any language.
0/5000Click to upload or drag and drop
Supported formats: JPEG, PNG, WEBP Maximum file size: 30MB; Maximum files: 10
1 to 10 reference images (JPEG, PNG or WebP, up to 30 MB each). The order defines "image 1", "image 2" in your prompt. Reference images are free, but more images take longer.
Output image width. Both must be passed at the same time; JSON integers, each side ≥256 and ≤4096; aspect ratio between 1:2.5 and 2.5:1. Once passed, aspect_ratio and resolution are ignored.
Output image height. Both must be passed at the same time; JSON integers, each side ≥256 and ≤4096; aspect ratio between 1:2.5 and 2.5:1. Once passed, aspect_ratio and resolution are ignored.
Output aspect ratio. auto (default) follows the first reference image.
Output resolution: 1K, 2K or 4K. Higher resolutions cost more and take longer.
Output image format.
Number of images to generate with the same settings.
Random seed. Omit for a random result; with the same seed and parameters the result is reproducible. When generating multiple images, image i uses seed + i. 0-2147483647
Rewrites the prompt into a fuller description before generating, which noticeably improves quality. Enabled by default; pass false to turn it off. Recommended to turn off when your request must be followed literally (exact text edits, "keep X unchanged and only change Y", fixed compositions) or when latency matters.
A configurable parameter. Defaults to true in the Playground.
Explore different use cases and parameter configurations
Complete guide to using Flux 3
Black Forest Labs describes FLUX 3 as one multimodal model for image, video, audio and action prediction. This page offers its open-weights image member, FLUX 3 Image Klein, as two API models on Kie.ai (flux-3/klein-text-to-image and flux-3/klein-edit-image). It generates images from text, edits them from an instruction, and combines several reference images into one.
Both endpoints run on POST /api/v1/jobs/createTask and return results through callBackUrl or the Get Task Details endpoint.
| Parameter | Text-to-image | Edit |
|---|---|---|
| model | flux-3/klein-text-to-image | flux-3/klein-edit-image |
| prompt | Required. Text prompt up to 5000 characters. | Required. An instruction describing how to edit the image, up to 5000 characters. |
| image_urls | Not used. | Required. 1 to 10 reference image file URLs (not file content). Accepted types: image/jpeg, image/png, image/webp. Max 30.0MB per image. |
| aspect_ratio | 1:1 (default), 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21. | auto (default, matches the first input image ratio), 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21. |
| resolution | 1K (default), 2K, or 4K. | 1K (default), 2K, or 4K. |
| width & height | Optional integers from 256 to 4096. Must be passed together, with a width:height ratio between 1:2.5 and 2.5:1. Once passed, aspect_ratio and resolution are ignored. | Optional integers from 256 to 4096. Must be passed together, with a width:height ratio between 1:2.5 and 2.5:1. Once passed, aspect_ratio and resolution are ignored. |
| enhance_prompt | On by default (including when omitted). Rewrites the prompt into a fuller description before generating, which noticeably improves quality. Pass false to turn it off. | On by default (including when omitted). Rewrites the prompt into a fuller description before generating, which noticeably improves quality. Pass false to turn it off. |
| output_format | png (default), jpeg, webp. | png (default), jpeg, webp. |
| num_images | 1 to 4 (default 1). | 1 to 4 (default 1). |
| seed | Optional integer from 0 to 2147483647. | Optional integer from 0 to 2147483647. |
Each image was generated with Flux 3 Klein on Kie.ai in October 2026 and resized for the web.
We asked flux-3/klein-text-to-image for a small corner coffee shop at dusk with a wooden sign reading "NORTHWIND COFFEE" in white painted capitals, a chalk A-frame board on the sidewalk reading "OPEN 7AM", warm window light, a bicycle leaning on the wall, and wet pavement reflecting the lights, using 4K resolution, a 16:9 aspect ratio, and jpeg output. A 4096x2304 px (4K, 16:9) JPEG came back. In our October 2026 test output, both "NORTHWIND COFFEE" on the wooden sign and "OPEN 7AM" on the sidewalk board rendered as written, with a bicycle leaning on the wall, wet pavement reflecting the lights, a dusk sky, and shelves and cups inside the window that stay detailed at 100% crop.
We asked flux-3/klein-text-to-image for a tall watercolor botanical illustration of a single foxglove stem with purple bell-shaped flowers on cream paper with a plain background, setting custom width and height of 1000 x 2400, num_images set to 2, and png output. Two 1000x2400 px PNG images came back from the single request, matching the requested width and height. Both outputs are watercolor foxglove stems on cream paper, and the two images differ in the number and arrangement of flowers.
We passed two reference images to flux-3/klein-edit-image (our orange robot mascot image first, and an illustrated valley landscape banner titled "Explore the Valley" second) at 1K resolution, 16:9 aspect ratio, and jpeg output, asking it to place the robot from the first image on the grass in the second image, waving at the camera, while keeping its design, colors, and proportions the same. A 1360x768 px JPEG came back. In our test output, the robot (orange body, blue eyes, lightning-bolt chest badge, two yellow-tipped antennas) stands on the grass waving and the landscape's "Explore the Valley" title stayed in the image, but the robot kept its smooth 3D-rendered look from the first image instead of matching the illustrated style of the landscape, so the two styles visibly differ.
We passed a photo of a chalkboard menu to flux-3/klein-edit-image at 1K resolution and jpeg output, asking it to change the color of all chalk lettering to yellow while keeping every word, number, and price in the same positions and changing nothing else. An 1184x880 px JPEG came back. Every word and price on the "Today's Specials" board stayed the same and turned yellow; however, the coffee-cup drawing and the decorative flourish also turned yellow even though the prompt mentioned only the lettering.
This page on Kie.ai offers only FLUX 3 Image Klein (flux-3/klein-text-to-image and flux-3/klein-edit-image).
| Model | What Black Forest Labs describes | On this Kie.ai page |
|---|---|---|
| FLUX 3 Image Klein | Open-weights member of the FLUX 3 Image family (5.6B parameters, runs on a single consumer GPU; published in undistilled and distilled versions). Generates images from text, edits images from an instruction, and combines several references into one image. | Yes — as flux-3/klein-text-to-image and flux-3/klein-edit-image |
| FLUX 3 Image | Available under a commercial weights license for companies; Black Forest Labs shows native 2K and 4K generation and up to 10 references. | No |
| FLUX 3 Video | Listed by Black Forest Labs as part of the FLUX 3 family, which it describes as one multimodal model for image, video, audio and action prediction. | No |
| FLUX 3 Max | Listed by Black Forest Labs as part of the FLUX 3 family. | No |
| FLUX 3 Action | Listed by Black Forest Labs as part of the FLUX 3 family. | No |
On both Kie.ai endpoints, enhance_prompt is on by default, including when you omit the parameter. It rewrites your prompt into a fuller description before generating, which noticeably improves quality. Pass enhance_prompt: false when you do not want the prompt rewritten.
You can pass integer width and height values between 256 and 4096 (with a width:height ratio between 1:2.5 and 2.5:1). They must be passed together, and once passed, aspect_ratio and resolution are ignored.
Kie.ai provides this model as an API with a playground; there is no app to download or install. You can try this model in the playground on this page without writing code.
All models in the Kie.ai market use the same authentication method: an Authorization: Bearer YOUR_API_KEY header with the key from your Kie.ai API Key Management page. Your account has one credit balance that all generation services draw from; check it with the Get Remaining Credits API.
Both flux-3/klein-text-to-image and flux-3/klein-edit-image use POST /api/v1/jobs/createTask and support 1K, 2K, and 4K resolution presets, custom width and height from 256 to 4096, and up to 4 images per request (num_images).