## Basic model info

Model name: krea/krea-3
Model description: Krea 3 is a fast image editing model that follows plain-language instructions. It also generates images from text.


## Model inputs

- prompt (required): Instruction describing the image to generate, or the edit to make. For edits, say what should change and what should stay the same. (string)
- images (optional): Up to 8 input images to edit or combine. Leave empty for text-to-image generation. When using several images, refer to each by what it shows in the prompt. (array)
- thinking_mode (optional): How much planning happens before generation. 'fast' is quicker and works for most requests. 'high' plans complex or multi-step edits more carefully, but is slower and costs more. (string)
- resolution (optional): Output resolution. (string)
- aspect_ratio (optional): Aspect ratio of the output. 'auto' picks the framing from your input images and prompt. (string)
- regional_edit (optional): Experimental. Change only the parts of the image that need to change and keep the rest unchanged. Needs at least one input image. Quality may be lower and generation slower than a regular edit. (boolean)
- negative_prompt (optional): What to avoid in the output. (string)
- seed (optional): Random seed. Reusing a seed does not guarantee identical results, because the prompt is expanded before generation. (integer)


## Model output schema

{
  "type": "string",
  "title": "Output",
  "format": "uri"
}

If the input or output schema includes a format of URI, it is referring to a file.


## Example inputs and outputs

Use these example outputs to better understand the types of inputs the model accepts, and the types of outputs the model returns:

### Example (https://replicate.com/p/v6x750vv99rmy0d1869t0hjtz0)

#### Input

```json
{
  "images": [],
  "prompt": "koi fish swimming in a pond",
  "resolution": "2k",
  "aspect_ratio": "auto",
  "regional_edit": false,
  "thinking_mode": "fast"
}
```

#### Output

```json
"https://replicate.delivery/xezq/OHWy5JzUnhLqPBFlUO21ObdoEH6zitiTe63r6CfAMxT06vXXA/tmpjss2yesd.png"
```


## Model readme

> # Krea 3
> 
> Krea 3 is a fast image editing model from Krea. Tell it what should change and what should stay the same, and it returns the edit in seconds. It also generates images from text.
> 
> This is the Berry version of Krea 3, which is tuned for image quality. A faster version called Jam is coming from Krea.
> 
> ## What it can do
> 
> - Edit an image with a plain-language instruction.
> - Combine up to 8 input images into one result.
> - Generate an image from a text prompt when you don't pass any input images.
> - Change only part of an image and keep the rest unchanged, with regional edits.
> - Output at 1K or 2K resolution.
> 
> ## How to use it
> 
> - `prompt`: describe the edit or the image you want. Lead with the change, then add one sentence about what should stay the same. With several input images, refer to each one by what it shows, like "Put the sneaker from the second image on the person in the first image."
> - `images`: up to 8 input images. Leave empty for text-to-image.
> - `thinking_mode`: `fast` is quicker and works for most requests. `high` plans complex or multi-step edits more carefully, but it takes longer and costs more.
> - `resolution`: `1k` or `2k`.
> - `aspect_ratio`: pick a ratio, or leave it on `auto` to let the model choose the framing from your images and prompt.
> - `regional_edit`: experimental. Use it for local changes like recoloring an object or fixing a detail. It needs at least one input image. Expect lower quality and slower generation than a regular edit. Don't use it for global changes like restyling or relighting.
> - `negative_prompt`: what to avoid.
> - `seed`: reusing a seed doesn't guarantee identical results, because the prompt is expanded before generation.
> 
> Around 1 megapixel is enough for most input images. Send the highest-resolution version you have when small details like text or faces matter. Input images larger than 4 megapixels are resized.
> 
> ## Pricing
> 
> You pay per output image, based on resolution and thinking mode, plus a charge for input images based on their size in megapixels.
> 
> | Output | Fast thinking | High thinking |
> | - | - | - |
> | 1K | $0.03 | $0.05 |
> | 2K | $0.06 | $0.08 |
> 
> Input images cost $0.02 per megapixel.
> 
> ## Credits
> 
> Krea 3 is made by [Krea](https://www.krea.ai/). Learn more in the [Krea 3 API docs](https://www.krea.ai/docs/developers/krea-3/overview).

