Generate from a direct prompt
The workflow starts with the OpenAI GPT Image 2 model selected, so you can describe a scene, product shot, poster, or interface concept and submit it without changing models.
Create a new image from text or edit a reference image with GPT Image 2 selected in the first-screen generator. Use it for product shots, posters, UI mockups, image text, and precise creative changes.
GPT Image 2 AI Image Generator
Use GPT Image 2 for text-to-image, reference editing, text-heavy posters, and product visuals.
Reference Image
Describe your idea
0/5000Model
Aspect Ratio
Output Format
Recent Jobs
Features
The workflow starts with the OpenAI GPT Image 2 model selected, so you can describe a scene, product shot, poster, or interface concept and submit it without changing models.
Upload one reference image and describe the exact change. Replicate documents the model with an input_images field for editing or composing from references.
What it is
GPT Image 2 AI Image Generator is a Uni-1 image workflow that calls Replicate model openai/gpt-image-2. The page keeps GPT Image 2 selected in the tool so you can start with a prompt or upload a source image for editing.
Replicate describes GPT Image 2 as an OpenAI image generation and editing model for creating images from text and editing existing images with precise instructions. Its documented inputs include prompt, input_images, aspect_ratio, quality, number_of_images, output_format, background, moderation, and an optional OpenAI API key field.
This implementation uses the confirmed Replicate model id and the documented GPT Image 2 input names. It does not promise transparent PNG backgrounds, because Replicate notes that GPT Image 2 does not support transparent backgrounds.
How to use it
The generator is visible before the guide, so you can test GPT Image 2 AI Image Generator immediately and then refine the prompt.
01
Use Text-to-Image for a new visual, or Image Edit when you want GPT Image 2 to transform an uploaded reference.
02
Describe the subject, material, lighting, camera angle, text copy, and the parts that must remain unchanged.
03
Select 1:1, 3:2, or 2:3, choose JPG, PNG, or WEBP, then start the job and download the result when it is ready.
Use cases
Use GPT Image 2 when the prompt needs detailed instruction following, reference-aware edits, or readable visual layout.
Examples
Browse real Uni-1 image cases, copy a prompt, and test it with the GPT Image 2 model already selected.
Yes. The generator sends jobs with modelKey gpt_image_2, and the server maps that to Replicate model openai/gpt-image-2 with the documented input_images and number_of_images fields.
Yes. Replicate documents GPT Image 2 for creating images from text descriptions. Use Text-to-Image, write the scene, choose an aspect ratio, and start the job.
Yes. Use Image Edit, upload a source image, and describe the change. The implementation sends the source URL through Replicate input_images for GPT Image 2.
The page shows 1:1, 3:2, and 2:3 for GPT Image 2 because those are the aspect_ratio options listed on the Replicate model page.
Replicate notes that GPT Image 2 does not support transparent backgrounds. Use opaque images here, or choose a different model if a transparent PNG is required.
For GPT Image 2, the UI offers JPG, PNG, and WEBP. The server sends JPG to Replicate as jpeg because that is the documented output_format value.
Replicate positions GPT Image 2 around sharp text rendering and complex layouts. Keep copy short, put exact text in quotes, and inspect the generated image before using it commercially.
Open the generator, write one specific prompt or upload a reference image, then refine from the result.
Use quoted copy, hierarchy, and typography direction when asking for posters, UI mockups, infographics, labels, or social graphics that need sharper image text.
For edits, write what must remain stable, such as the subject identity, product shape, camera angle, clothing, lighting, or background structure.
The workflow exposes the Replicate-supported aspect ratios and output formats for GPT Image 2 instead of showing incompatible widescreen or vertical ratios.
When the job finishes, preview the output, download it, or load the result back into the editor for another targeted GPT Image 2 pass.