Text-to-image generation
Describe the subject, composition, lighting, material, style, and intended use to create a new image from a written brief.
GPT Image 2 AI Image Generator & Editor
Generate a new visual from text or edit an existing image with GPT Image 2. Work with high-fidelity reference inputs, choose flexible aspect ratios and 1K–4K output, then refine the details that matter in one focused workspace.
4
OpenAI image generation and editing
Turn prompts and reference images into usable visual work
OpenAI describes GPT Image 2 as its state-of-the-art model for fast, high-quality image generation and editing. It accepts text and image inputs, produces images, supports inpainting, and is designed for flexible sizes and high-fidelity reference inputs.
Start from a clear deliverable, then use references and focused edits to move toward the final image.
Image editing and inpainting
Upload an existing image and direct a full-image or local revision while naming the identity, geometry, layout, and details that must remain unchanged.
High-fidelity image inputs
Use reference images to communicate a subject, product, palette, material, setting, or composition more precisely than text alone.
1K, 2K, and 4K output
Choose the resolution that matches the stage of work: 1K for exploration, 2K for review, or 4K for detail-rich delivery.
Flexible canvas formats
Create square, portrait, landscape, ultra-wide, or ultra-tall images with 15 explicit aspect ratios plus an automatic option in this workspace.
Transparent 1K assets
Select a transparent background at 1K for isolated product elements, icons, stickers, and compositing-ready creative assets.
GPT Image 2 use cases
Create the image, then keep improving it without restarting
GPT Image 2 fits workflows where a strong first composition and controlled revisions both matter—from product campaigns to poster concepts and reference-led edits.
Product imagery
Direct polished product scenes around a real reference
Use a clear product image to preserve shape, color, material, and recognizable details while changing the set, camera framing, light, or campaign mood.
Posters and social
Build layout-aware concepts for the intended canvas
State the destination and aspect ratio before generation, reserve copy space, quote any required wording, and proofread every character before publishing.
Focused image edits
Change one decision while protecting the approved image
Replace a background, adjust a material, remove an object, or rework a region while repeating what must stay fixed and reviewing the complete result.
How to use GPT Image 2
Move from brief to reviewed image in four deliberate steps
The most reliable workflow separates the deliverable, reference roles, requested changes, and final review instead of burying every decision in one long prompt.
01
Choose generation or editing
Start from text for a new composition. Add reference images when an existing subject, product, layout, or visual identity must guide the output.
02
Set size and background
Choose an aspect ratio and 1K, 2K, or 4K output. Use the 1K transparent option when the result needs an alpha channel.
03
Write visible instructions
Describe subject, composition, lighting, materials, exact text, exclusions, and the details that must remain unchanged during an edit.
04
Review and refine
Inspect identity, hands, edges, geometry, small text, background transparency, and unwanted changes. Make the next revision narrow and measurable.
GPT Image 2 prompt guide
Write a compact art-direction brief, not a bag of keywords
Lead with the deliverable, describe what should be visible, assign every reference a purpose, and finish with constraints that protect the result.
Reference-led product prompt
DELIVERABLE — A premium 16:9 hero image for a fragrance launch. REFERENCE — Use @Image 1 for the exact bottle shape, amber glass color, cap proportions, and label placement. SCENE — Dark stone surface at blue hour, soft warm rim light, subtle haze, realistic glass reflections, restrained editorial color grade. COMPOSITION — Bottle centered-right with generous clean negative space on the left for a headline. Eye-level camera, crisp product edges, shallow background depth. CONSTRAINTS — Preserve the product geometry and label position. No added text, duplicate bottles, warped glass, extra logos, or watermark.
Name the deliverable
Say whether you need a product hero, poster, storyboard frame, transparent asset, editorial photograph, or another specific output.
Describe visible relationships
Explain where the subject sits, how the camera and light behave, and how each reference should influence the image.
Separate changes from invariants
For edits, identify the one change to make and repeat every identity, geometry, layout, or lighting detail that must remain stable.
Treat generated text as a draft
Put exact copy in quotes, request it once, and verify spelling, spacing, punctuation, and small type before production use.
When to choose GPT Image 2
A capable balance for generation, references, and controlled edits
Choose GPT Image 2 when you want high-quality image generation and editing with broad canvas options, reference-led direction, and transparent asset support in one workflow.
Use GPT Image 2 for flexible everyday production
Generate or edit with up to 16 JPEG, PNG, or WebP reference images here, choose 1K–4K output, and work across square, portrait, landscape, and extended canvases.
Create with GPT Image 2Read the official model documentation
OpenAI introduced GPT Image 2 on April 21, 2026 and documents text and image inputs, image output, flexible sizes, high-fidelity inputs, generation, editing, and inpainting.
View OpenAI model docsKeep a human review step
Generated images can still introduce spelling errors, identity drift, altered product geometry, or unexpected details. Review every production asset and use a design tool when pixels must remain exact.
GPT Image 2 FAQ
GPT Image 2 questions, answered
Quick answers about the model, generation and editing modes, references, resolution, aspect ratio, and transparency.
What is GPT Image 2?
GPT Image 2 is OpenAI's image generation and editing model introduced on April 21, 2026. OpenAI describes it as a state-of-the-art model for fast, high-quality generation and editing with flexible image sizes and high-fidelity image inputs.
Can GPT Image 2 edit an existing image?
Yes. GPT Image 2 supports image editing and inpainting. Upload a reference, describe the change, and list the identity, composition, geometry, lighting, text, or other details that should stay unchanged.
How many reference images can I use here?
This workspace accepts up to 16 JPEG, PNG, or WebP reference images, with a maximum file size of 30 MB each. Give every reference a clear role and remove images that do not affect the brief.
Which GPT Image 2 resolutions are available?
Choose 1K, 2K, or 4K output. Not every aspect ratio is available at every higher resolution, so the workspace automatically keeps the resolution and canvas combination compatible.
Does GPT Image 2 support transparent backgrounds?
Yes. In this workspace, transparent, opaque, and automatic background options are available for 1K GPT Image 2 output. Inspect hair, glass, shadows, and fine edges before using the asset in a final composite.
How should I prompt GPT Image 2?
Start with the intended deliverable, then describe the subject, composition, style, lighting, materials, and constraints. For edits, separate the requested change from the details that must remain fixed, and refine one decision at a time.
Create with GPT Image 2
Turn a clear visual brief into your next image
Choose a canvas, add purposeful references, protect the details that matter, and create or edit with GPT Image 2.