Dense text and fine detail
OpenAI highlights how the model renders dense text, small type, icons and interface elements. On Deepnia, it is the model offered for infographics, graphic design and text-rich images.
Image model · OpenAI
GPT Image 2 — Text-rich visuals with up to 16 references.
GPT Image 2 is OpenAI’s image generation and editing model, launched on April 21, 2026. On Deepnia it is mainly used for text-rich visuals (infographics, posters, flyers, app mockups), with up to 16 reference images and 1K to 4K output depending on the aspect ratio.
- 1K–4K by aspect ratio
- Up to 16 references
- 8 aspect ratios + Auto
- 4K in 5 aspect ratios
- 20,000-character prompt
The image tool opens with GPT Image 2 already selected. The exact cost shows before every creation; you sign up when you generate.
1:1
3:2
2:3
16:9
9:16
21:9
At a glance
- Maker
- OpenAI
- Resolution
- 1K, 2K or 4K by aspect ratio (1K by default)
- 4K
- In 3:2, 2:3, 16:9, 9:16 and 21:9
- Aspect ratios
- 8 + Auto (9:16 by default)
- References
- Up to 16 images
- Images per generation
- 1
What GPT Image 2 does best
Infographics and learning materials
OpenAI names localized advertising, infographics, visual explainers and educational content among its target uses. Always check figures, arrows and labels: OpenAI says they can still contain errors.
Up to 16 reference images
Add a product, a logo, a character or brand guidelines: up to 16 images, and the tool switches to edit mode. OpenAI says every image you provide is automatically processed at high fidelity.
1K to 4K depending on the aspect ratio
3:2, 2:3, 16:9, 9:16 and 21:9 go up to 4K, 1:1 up to 2K, and Auto, 5:4 and 4:5 come out at 1K. Pick the aspect ratio first: the tool then offers the matching resolutions.
Many styles, several scripts
Photorealism, cinematic stills, pixel art or manga: OpenAI highlights a wide range of styles, and clear progress on non-Latin scripts, including Japanese, Korean, Chinese, Hindi and Bengali.
How to use GPT Image 2 on Deepnia
Open the image tool
The “Try GPT Image 2” button opens Deepnia’s image tool with the model already selected.
Add references if you need them
Under “References”, click “Add an image”: up to 16 images, JPG, PNG or WebP, up to 10 MB each. With at least one reference, the tool switches to “Edit mode”.
Write a designer’s brief
Describe the layout, the heading hierarchy and the exact text in quotation marks; the prompt takes up to 20,000 characters.
Pick the aspect ratio, then the resolution
The “Resolution” options depend on the “Aspect ratio”: 3:2, 2:3, 16:9, 9:16 and 21:9 up to 4K; 1:1 up to 2K; Auto, 5:4 and 4:5 at 1K. Check the cost shown, then click “Generate” or “Edit”.
GPT Image 2 on Deepnia: what you can set
- Inputs
- Text only, or text plus reference images
- Modes
- Generation without references; editing (“Edit mode”) as soon as you add one
- Resolution
- 1K, 2K or 4K (1K by default)
- Resolution by aspect ratio
- 3:2, 2:3, 16:9, 9:16 and 21:9 up to 4K; 1:1 up to 2K; Auto, 5:4 and 4:5 at 1K
- Aspect ratios
- Auto, 1:1, 3:2, 2:3, 16:9, 9:16, 5:4, 4:5, 21:9 (9:16 by default)
- References
- Up to 16 images; JPG, PNG or WebP, up to 10 MB each
- Images per generation
- 1
- Prompt
- Required, up to 20,000 characters
Prompt ideas to try
Copy a prompt, open the tool and adapt it to your project.
Infographic for a fitness coach
- 2:3
- 2K
- Text only
Vertical infographic titled "5 stretches after your run". Five numbered blocks, each with a simple figure showing the stretch, a short title and one sentence of explanation. Cream background, coral and midnight-blue colors, flat design style, all text perfectly readable on a phone.
App screen mockup
- 9:16
- 1K
- Text only
Realistic screenshot of a mobile app for booking yoga classes. Header "My classes", a card reading "Next class: Vinyasa, Thursday 6:30 p.m.", a list of four classes with the time and the teacher’s name, a large green "Book" button. Clean interface, light mode, modern typography.
Flyer for a product launch
- 4:5
- 1K
- Text only
Flyer for the launch of an iced tea range: title "New: peach-mint iced tea", subtitle "Tasting on Saturday, 10 a.m. to 6 p.m.", three bottles lined up on a pale yellow background with mint leaves, the address at the bottom in small readable type. Modern, airy layout.
Campaign from your product photos
- 5 references: 4 photos of the watch and the logo
- 16:9
- 4K
From the reference photos of the watch and the logo, create a campaign visual: the watch resting on a dark stone by the water, low evening light, precise reflections on the dial. Keep the dial, the strap and the logo exactly the same. In the top left corner, the text "Built to last" in thin white letters.
New label on your product photo
- Reference: your photo
- 1:1
- 2K
From your photo of the honey jar, replace the label with a round kraft-paper label that clearly reads "Wildflower honey", "2026 harvest" and "250 g". Keep the jar, the lid, the lighting and the background the same.
Website banner for a conference
- 21:9
- 4K
- Text only
Panoramic banner for a conference website: a large hall with a giant screen and the audience seen from behind, blue and purple lights. On the left, the title "Hybrid Work Summit" in bold white letters, below it "November 12–13 · Convention Center", and a small "Register" button. All text sharp and readable.
Tips for better results
- Put every piece of text in quotation marks, exactly as it should appear, and say where it goes (title at the top, details at the bottom).
- Describe the layout like a designer’s brief: zones, heading hierarchy, colors, space kept for the logo.
- With several references, give each one a role (“photo 1: the product; photo 2: the logo”).
- For an edit, say what changes and what must stay the same: OpenAI notes that characters and brand elements can drift from one creation to the next.
- For 4K, choose 3:2, 2:3, 16:9, 9:16 or 21:9; in Auto, 5:4 and 4:5, the image comes out at 1K.
- Test the layout in 1K and compare versions with “Regenerate”; once the prompt works, move to 2K or 4K with an aspect ratio that allows it.
Good to know
- Pick 9:16 for stories, Reels and TikTok, 1:1 for a feed post, 16:9 for YouTube and 21:9 for a website banner.
- Before you publish, proofread every piece of text and check the figures, labels and arrows of any diagram.
- For a recent product or event, describe it precisely or add a photo as a reference: OpenAI places the model’s knowledge at December 2025.
- Allow up to 2 minutes for a complex request, according to OpenAI.
GPT Image 2 or another model?
Deepnia’s other image models, to pick the right one for your project.
| Model | Resolution | Aspect ratios | References | Best for |
|---|---|---|---|---|
| GPT Image 2This model | 1K–4K | 8 ratios | Up to 16 | Text-rich visuals with up to 16 references. |
| Nano Banana 2 | 1K–4K | 14 ratios | Up to 14 | Precise edits that keep the subject, in 14 aspect ratios and as PNG or JPG. |
| Nano Banana Pro | 1K–4K | 6 ratios | Up to 8 | High-fidelity edits and legible text, up to 4K in every aspect ratio. |
| Nano Banana 2 Lite | 1K | 14 ratios | Up to 10 | Fast drafts in 1K, with up to 10 references. |
| Seedream v5.0 Lite | Automatic | 7 ratios | Up to 10 | Exact pixel sizes set for each aspect ratio, with up to 10 references. |
Frequently asked questions
What is GPT Image 2?
It is OpenAI’s image generation and editing model, launched on April 21, 2026.
Which languages can GPT Image 2 write in?
OpenAI says its models were already more consistent in English and other Latin-script languages, such as French, and highlights clear progress in Japanese, Korean, Chinese, Hindi and Bengali. Write the exact text in quotation marks and proofread accents and spelling before you publish.
What is the maximum resolution on Deepnia?
4K, in 3:2, 2:3, 16:9, 9:16 or 21:9. 1:1 goes up to 2K (if you were at 4K, the tool switches to 2K automatically) and Auto, 5:4 and 4:5 come out at 1K. OpenAI still treats outputs above 2K as experimental, so validate the layout at a lower resolution first.
What resolution does an Auto image come out in?
1K: in Auto, the model picks the proportions itself and the image comes out at 1K, as it does in 5:4 and 4:5. To go up to 2K or 4K, pick a fixed aspect ratio such as 16:9 or 9:16.
How many reference images can I use?
Up to 16, JPG, PNG or WebP up to 10 MB each. As soon as you add a reference, the tool switches to edit mode; say in the prompt what each image is for.
GPT Image 2 or Nano Banana Pro: which one should I use?
GPT Image 2 for infographics, app mockups and text-heavy layouts, with up to 16 references. Nano Banana Pro for high-fidelity edits and posters, up to 4K in every aspect ratio.
How many images do I get per generation?
1 image per generation on Deepnia. Click “Regenerate” to get another version from the same request.
How do I keep the same brand look across several visuals?
Attach the same references to each creation (logo, color palette, a sample visual), give each one a role and describe the style with the same words. OpenAI says every image you provide is processed at high fidelity.
Ready to try GPT Image 2?
The image tool opens with GPT Image 2 already selected. The exact cost shows before every creation; you sign up when you generate.