Skip to content

GPT Image 2 (via Fal.ai)

OpenAI's GPT Image 2 model for image editing via Fal.ai.

Quick Example

from tarash.tarash_gateway import generate_image
from tarash.tarash_gateway.models import ImageGenerationConfig, ImageGenerationRequest

config = ImageGenerationConfig(
    provider="fal",
    model="openai/gpt-image-2/edit",
    api_key="YOUR_FAL_KEY",
)

request = ImageGenerationRequest(
    prompt="Add a dramatic rainbow arching over the elephant",
    image_list=[
        {"image": "https://example.com/photo.jpg", "type": "reference"},
    ],
    n=1,
    quality="high",
)

response = generate_image(config, request)
print(response.images)  # → list of image URLs

Supported Models

Model Description Image Input Notes
openai/gpt-image-2/edit Image editing with reference images ✅ Up to 16 reference images; optional mask for targeted edits

Parameters

Parameter Required Notes
prompt ✅ Text description of the desired edit
image_list (reference) ✅ One or more reference images to edit
n — Number of images to generate (1–4), maps to num_images
quality — "low", "medium", or "high" (default: "high")
size — "auto", "square_hd", "portrait_4_3", etc. (default: "auto")
extra_params.output_format — "jpeg", "png", or "webp" (default: "png")
extra_params.mask_url — URL of a mask image — white areas indicate the region to edit

Image Editing with Mask

config = ImageGenerationConfig(
    provider="fal",
    model="openai/gpt-image-2/edit",
    api_key="YOUR_FAL_KEY",
)

request = ImageGenerationRequest(
    prompt="Replace the sky with a stormy night",
    image_list=[
        {"image": "https://example.com/landscape.jpg", "type": "reference"},
    ],
    extra_params={
        "mask_url": "https://example.com/sky-mask.png",
        "output_format": "png",
    },
)

response = generate_image(config, request)
print(response.images[0])