How do references guide GPT Image 2?

Generate hero shots and brand scenes from text, or upload references and describe the retouch — background swaps, label updates, and multi-image composites across up to 16 inputs. When uploading several images, say what each one contributes — “product from image 1, backdrop from image 2, lighting mood from image 3.” Supports up to 16 references.

How can I improve text in GPT Image 2 images?

Create posters, packaging concepts, and menus with quoted text. Check spelling, numbers, and line breaks before publishing. Put headlines, prices, and labels in “quotes” and note placement — centered title, price tag bottom-right, small legal line along the footer.

When should I choose GPT Image 2?

Skin, fabric, food, metal, and glass look believable without stacking modifier keywords — a strong default for e-commerce hero shots, brand campaigns, and editorial-style portraits. For product and portrait work, specify material and lighting — brushed steel, matte glaze, soft window light from camera-left — instead of relying on generic “realistic” modifiers.

OpenAI text-to-image · Complex prompts · In-image typography

GPT Image 2 — Image Generation and Editing

Generate images from detailed prompts or edit with up to 16 references. GPT Image 2 supports product visuals, portraits, and layouts with text.

Examples

See what GPT Image 2 can make

What it is good for

GPT Image 2 — Image Generation and Editing

Generate images from detailed prompts or edit with up to 16 references. GPT Image 2 supports product visuals, portraits, and layouts with text.

Create with GPT Image 2

Multi-element creative briefs

Describe several subjects, their relationships, and the camera setup in one prompt — useful for storyboards, campaign key visuals, and scene-heavy concept work.

Marketing layouts with real copy

Draft posters, packaging mocks, menu boards, and social ads where the words on the image need to be spelled correctly, not guessed.

Product and lifestyle photography

Generate hero shots and brand scenes from text, or upload references and describe the retouch — background swaps, label updates, and multi-image composites across up to 16 inputs.

Key capabilities

Why creators pick GPT Image 2

Modern editorial poster demonstrating precision and creative control

OpenAI — AI Image Generator & Editor

An OpenAI image model for text-to-image creation and reference-guided edits in the same workspace.

“Create a polished campaign key visual from this complete art-direction brief while preserving the hierarchy and product silhouette.”

Wildlife magazine spread combining imagery, facts, and structured editorial text

Compose multi-element scenes

Describe subjects, their relationships, and layout in one brief. Review the result for missing objects or changes to required details.

“A busy design studio with exactly five people, two paper prototypes, one red task lamp, and a rainy city visible through the windows.”

Editorial poster showcasing multilingual typography

Typography that reads like real copy

Create posters, packaging concepts, and menus with quoted text. Check spelling, numbers, and line breaks before publishing.

“A premium tea package mockup with the exact label ‘MIST GARDEN’ and the small line ‘SPRING HARVEST’. No other text.”

Photoreal materials out of the box

Skin, fabric, food, metal, and glass look believable without stacking modifier keywords — a strong default for e-commerce hero shots, brand campaigns, and editorial-style portraits.

How to get a strong result

Tips for better results

Start with the subject, then add lighting, style, and anything that must stay the same. Upload a reference whenever you want a more controlled edit.

1
Treat the prompt like a brief
Lead with the subject, then add setting, lighting, lens, and a checklist of objects that must appear. GPT Image 2 is designed for prompts that would overwhelm lighter models.
2
Quote exact on-image copy
Put headlines, prices, and labels in “quotes” and note placement — centered title, price tag bottom-right, small legal line along the footer.
3
Name surfaces and light sources
For product and portrait work, specify material and lighting — brushed steel, matte glaze, soft window light from camera-left — instead of relying on generic “realistic” modifiers.
4
Label each reference’s role
When uploading several images, say what each one contributes — “product from image 1, backdrop from image 2, lighting mood from image 3.” Supports up to 16 references.
5
Iterate at medium, export at high
Use medium quality to explore composition cheaply, then move to high quality and larger aspect ratios — including experimental 4K sizes — once the direction is locked.

Related AI tools

Continue your workflow with other focused Image tools.

BeforeAfter

Image Upscaler

Enlarge low-resolution images while recovering sharper edges and texture.

ImageFeatured

BeforeAfter

Face Restorer

Recover facial detail in old, compressed, or AI-generated portraits.

Image

BeforeAfter

Object Remover

Brush over an unwanted object and reconstruct the background naturally.

ImageFeatured

Series compare

Compare GPT Image versions

Pick the version that fits your workflow — then open its page to create with that model locked in.

Feature	GPT Image 2.5	GPT Image 2Current	GPT Image 1.5
Best for	Fast iteration through premium final creative	Complex prompts, in-image text, photoreal products	Everyday production work and cutouts
Speed	Fastest GPT Image generation with Flare	Balanced	Fast
On-image text	Best in the GPT Image family	Excellent	Very good
Transparent background	Supported	Not supported	Supported
Sizes & framing	Ratios and explicit sizes up to 4K	Wide range, up to 4K experimental	Standard ratios
Editing	Highest precision and reference fidelity	Up to 16 references	Strong prompt following + input fidelity