# FLUX.1 Kontext Pro

FLUX.1 Kontext Pro introduced multi-turn iterative image editing where each edit builds on the last while preserving identity, style, and distinctive visual features. It delivers fast inference relative to earlier models in its class.

- **Model ID:** `bfl/flux-kontext-pro`
- **Type:** image
- **Providers:** bfl, prodia
- **Context window:** 512
- **Maximum output tokens:** 0
- **Pricing:** $0.04/image
- **Canonical page:** https://vercel.com/ai-gateway/models/flux-kontext-pro

## Supported parameters

- **Parameters:** `aspectRatio`, `seed`
- **Operations:** `text-to-image`, `image-to-image`
- **Input modalities:** `text`, `image`
- **Output modalities:** `image`
- **Aspect ratios:** `3:7`, `2:3`, `9:16`, `3:4`, `1:1`, `4:3`, `16:9`, `7:3`, `21:9`, `9:21`
- **Resolutions:** Not reported
- **Durations:** Not reported
- **Frame rates:** Not reported
- **Maximum outputs per request:** Not reported
- **Maximum input images:** 4
- **Input image formats:** Not reported
- **Maximum input image size:** 20 MB

## Example

```ts
import { experimental_generateImage as generateImage } from 'ai';

const result = await generateImage({
  model: 'bfl/flux-kontext-pro',
  prompt: 'A red balloon on a wooden table.'
});
```

## About

Black Forest Labs released FLUX.1 Kontext Pro on May 29, 2025 as the founding model of the Kontext suite. Black Forest Labs describes it as oriented toward fast iterative image editing: a model that lets you build on previous edits through multiple turns while keeping characters, identities, styles, and distinctive features consistent across scenes and viewpoints.

The architecture accepts both a text instruction and one or more reference images as input. Rather than regenerating a scene from scratch, the model identifies which elements should change and which to preserve. It applies targeted local modifications while leaving unspecified regions intact. Black Forest Labs's KontextBench evaluation covers six editing capabilities: character consistency across environments, local region editing, style reference transfer, text within images, background replacement, and interactive speed. Kontext Pro achieved its highest scores on text editing and character preservation tasks.

The multi-turn loop is what sets Kontext Pro apart operationally. Each edited output becomes the reference image for the next instruction. An editing session can progress through many sequential changes (adjusting clothing, then placing the character in a new scene, then modifying lighting) without losing the identity thread from step one. Black Forest Labs notes that sessions beyond roughly six turns can introduce visual artifacts.

## What to consider

Kontext Pro's iterative editing loop, feeding one output as the next request's input, maps naturally to conversational editing interfaces where users refine an image through a sequence of natural language instructions. Compare $0.04 with other editing tiers.

## When to use

### Best For

- Conversational image editing products where users refine a single image through a sequence of natural language instructions, each building on the previous result
- Character consistency pipelines, placing a specific person or product across multiple distinct scenes while preserving unique visual attributes throughout
- Local editing tasks such as changing a background, swapping clothing, modifying an expression, or adjusting a single object without disturbing the rest of the frame
- Style reference workflows that apply the aesthetic of a reference image to a new scene described in text

### Consider Alternatives When

- Exact prompt adherence and typography precision matter more than editing speed. Kontext Max targets those properties
- The task is inpainting a masked region rather than instruction-based editing. FLUX.1 Fill Pro is built for mask-based fill workflows
- Pure text-to-image generation without a reference image is the primary need. FLUX.2 Pro or FLUX.2 Max may be more appropriate

## Best for

- **Conversational image editing:** Products where users refine a single image through a sequence of natural language instructions, each building on the previous result
- **Character consistency pipelines:** Placing a specific person or product across multiple distinct scenes while preserving unique visual attributes
- **Local editing tasks:** Changing a background, swapping clothing, modifying an expression, or adjusting a single object without disturbing the rest of the frame
- **Style reference workflows:** Applying the aesthetic of a reference image to a new scene described in text

## Consider alternatives

- **Typography and prompt precision:** Kontext Max targets exact prompt adherence and on-image text accuracy beyond Kontext Pro
- **Masked region inpainting:** FLUX.1 Fill Pro is built for mask-based fill rather than instruction-based editing
- **Pure text-to-image generation:** FLUX.2 Pro or FLUX.2 Max may be more appropriate when no reference image is needed

## Frequently asked questions

### What makes Kontext Pro a multi-turn editing model?

The model accepts the output of one editing step as the reference image for the next. Each turn receives a new instruction and the previous result, preserving identity and style across steps. Black Forest Labs notes that sessions beyond roughly six turns can introduce visual artifacts.

### How does Kontext Pro handle character consistency across different scenes?

The model extracts and retains distinctive visual features from the reference image (face, clothing, proportions, recognizable attributes) and reapplies them when generating a new scene. A specific character can be placed in an entirely different environment without reconstructing their appearance from scratch.

### Does Kontext Pro perform both local and global edits?

Yes. Local editing modifies specific elements (an expression, an object, or a piece of text) while leaving surrounding regions unchanged. Global edits transform the entire scene context, moving a subject to a new environment while preserving their visual identity.

### How does Kontext Pro score on KontextBench?

Kontext Pro achieves its highest scores on text editing and character preservation within KontextBench's six real-world editing categories, with measurable scores across all six.

### What happens after many sequential editing turns?

Visual artifacts can accumulate. Black Forest Labs documents this limitation with an example showing degradation after six sequential edits. Shorter, more targeted editing sequences avoid this issue.

### Does Kontext Pro support text-to-image generation without a reference?

Yes. You can use Kontext Pro for text-to-image generation without supplying a reference image.

### What does FLUX.1 Kontext Pro cost per image?

See the pricing section on this page for today's rates. AI Gateway exposes each provider's pricing for FLUX.1 Kontext Pro.

## Links

- [Model page](https://vercel.com/ai-gateway/models/flux-kontext-pro)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://bfl.ai/models/flux-kontext)
- [Provider pricing](https://bfl.ai/pricing)