Flux 2 Edit

Fal.ai's powerful multi-image composition model. Combine up to 9 reference images with advanced control over the generation process.

POST

/api/v1/models/flux-edit/edit

Edit and compose images using Flux 2

Model

Flux 2 Edit (fal.ai)

Cost

15 credits/image

Avg. Time

8-15 seconds

Multi-Image Composition

Flux 2 Edit is unique in its ability to combine up to 9 reference images. Reference them in your prompt as "the person in image 1", "the building in image 2", etc.

Request Body

Name	Type	Description
`prompt`required	`string`	Text instructions for the edit. Reference images as 'image 1', 'image 2', etc.
`imageUrl`required	`string`	Primary source image URL (required)
`imageUrls`	`string[]`	Array of additional reference image URLs (up to 8 more)
`resolution`	`string`Default: `2K`	Output resolution: '1K', '2K', '4K'
`numImages`	`number`Default: `1`	Number of images to generate (1-4)
`aspectRatio`	`string`	Output aspect ratio
`outputFormat`	`string`Default: `png`	'png' or 'jpeg'
`guidanceScale`	`number`Default: `2.5`	How closely to follow the prompt (0-10)
`numInferenceSteps`	`number`Default: `28`	Quality/speed tradeoff (1-50)
`seed`	`number`	Random seed for reproducibility

Single Image Edit

"text">-purple-600">curl "text-blue-600">-X POST https://vydra.app/api/v1/models/flux"text-blue-600">-edit/edit \
  "text-blue-600">-H "Authorization: Bearer YOUR_API_KEY" \
  "text-blue-600">-H "Content">-Type: application/json" \
  "text-blue-600">-d '{
    "imageUrl": "https://example.com/portrait.jpg",
    "prompt": "Place this person on a beach at sunset, photorealistic, golden hour lighting",
    "resolution": "2K",
    "guidanceScale": 3.0
  }'

Multi-Image Composition

Combine multiple reference images into one cohesive output:

"text">-purple-600">curl "text-blue-600">-X POST https://vydra.app/api/v1/models/flux"text-blue-600">-edit/edit \
  "text-blue-600">-H "Authorization: Bearer YOUR_API_KEY" \
  "text-blue-600">-H "Content">-Type: application/json" \
  "text-blue-600">-d '{
    "imageUrl": "https://example.com/living">-room.jpg",
    "imageUrls": [
      "https://example.com/sofa.jpg",
      "https://example.com/lamp.jpg"
    ],
    "prompt": "Redesign this living room. Place the sofa from image 2 in the center and add the lamp from image 3 in the corner. Modern minimalist style.",
    "resolution": "2K"
  }'

Image Numbering

The first image (imageUrl) is "image 1". Additional images inimageUrls are numbered sequentially: "image 2", "image 3", etc.

Response

200 OKjson

{
  "images": [
    {
      "url": "https://fal.media/files/xxx/output_0.png",
      "width": 2048,
      "height": 2048,
      "content_type": "image/png"
    }
  ],
  "description": "Generated with Flux 2 Edit",
  "creditsCharged": 15,
  "durationMs": 12340
}

Response Fields

Name	Type	Description
`images`	`array`	Array of generated image objects
`images[].url`	`string`	Public URL of the generated image
`images[].width`	`number`	Image width in pixels
`images[].height`	`number`	Image height in pixels
`description`	`string`	Generation description
`creditsCharged`	`number`	Credits deducted (15 × numImages)
`durationMs`	`number`	Processing time in milliseconds

Advanced Parameters

Guidance Scale

Controls how closely the model follows your prompt. Higher values = more literal interpretation.

1-2 - More creative freedom
2.5 - Balanced (default)
3-5 - Closer to prompt
5+ - Very literal, may reduce quality

Inference Steps

More steps = higher quality but slower generation.

15-20 - Fast, decent quality
28 - Balanced (default)
35-50 - Maximum quality

Seed

Set a specific seed for reproducible results. Using the same seed, prompt, and parameters will generate the same image.

Use Cases

Product Placement

Place products from one image into different scenes or contexts.

Interior Design

Combine furniture and decor from multiple sources into room photos.

Character Compositing

Place people from different photos into the same scene.

Style Mixing

Combine elements from different styles or time periods.

Tips for Best Results

Be Specific with References

Instead of "add the person", say "place the person from image 2 in the foreground, facing left, at about 2/3 the height of the frame".

Match Lighting

For realistic composites, try to use source images with similar lighting. Or specify "match the lighting of image 1" in your prompt.

Use Higher Steps for Complex Scenes

When combining many images, increase numInferenceSteps to 35-40 for better coherence.