OpenAI - GPT Image 2

GPT Image 2

Name: GPT Image 2
Brand: OpenAI
Availability: InStock

OpenAI image generation for crisp typography, photoreal product shots, UI mockups, infographics, and controlled reference edits.

Best for text-heavy images, product photography, UI mockups, and controlled edits.

Text->ImageImage->ImageQuality controlPNG / JPEG / WEBPMask URL

Launch GPT Image 2 Open the Image Lab

Pay-as-you-go · Price shown before you generate

GPT Image 2 AI video example: GPT Image 2 demo still from MaxVideoAI — Price$0.20/image
Resolution optionsUp to 4K canonical sizes
FormatPNG / JPEG / WEBP

Best use cases

Readable text in ads, posters, and infographicsProduct photography with sharp labelsUI and app-screen mockupsReference edits with optional masks

Why GPT Image 2 stands out

Readable text (Use it for posters, labels, infographics, and UI screens that need clean typography.)
Product fidelity (Strong fit for packaging, catalog shots, and e-commerce images with visible labels.)
Fal controls (Image size preset or custom width/height, quality, output format, and mask URL map directly into the provider request.)
Unified image workflow (Switch between Create and Edit without leaving the Image Lab.)

Specs

The limits that shape your renders.

Price / imagelandscape_4_3 $0.20/imagesquare_hd $0.29/imagesquare $0.29/imageportrait_4_3 $0.20/imageportrait_16_9 $0.22/imagelandscape_16_9 $0.21/imagecustom $0.20/image

Text-to-Image

Image-to-Image

Resolution optionsUp to 4K canonical sizes

Aspect ratiosauto / 16:9 / 4:3 / 1:1 / 3:4 / 9:16

Output formatPNG / JPEG / WEBP

Release dateApr 2026

How to write a strong GPT Image 2 prompt

Use structured prompts with scene, subject, exact text, visual details, use case, and constraints.

Tip: put exact words in quotation marks and describe where each text block should appear.

Quick prompt (fast iteration)

Use 1–2 sentences when you want variations.

Quick = variations. Use for fast iteration.

Template (copy/paste)

[Style] + [Subject] + [Setting] + [Lighting] + [Mood / palette]

Structured prompt (best for reliable results)

Separate information so the model can follow it consistently.

Structured = consistency. Use when you need reliable results.

Template (copy/paste)

Subject:
[Who/what + 1 clear action or pose.]

Setting:
[Where + time of day + 2–3 visual anchors.]

Composition:
- Framing: [wide / medium / close-up]
- Angle: [eye-level / low / high]
- Lens feel: [e.g., 35mm natural, shallow DOF]

Lighting + palette:
[Key light + 2–3 palette anchors]

Style constraints:
No logos. No readable text. Clean background.

Pro prompt (ultra-specific "studio brief")

Use this when you need a very specific look or continuity across shots.

Pro = continuity. Use for precise, repeatable looks.

Template (copy/paste)

Project / intent:
[One-line goal for the still.]

Subject:
[Who/what + wardrobe/materials + 2–3 distinctive traits.]

Location / set:
[Where + time of day + weather. Add 3 visual anchors.]

Composition:
- Framing + angle
- Lens feel + depth of field
- Look: [clean digital / subtle film grain / soft bloom]

Lighting & grade:
- Key light
- Contrast
- Palette anchors (3–5)

Constraints:
No logos. No readable text. No overlays.

Before you generate

Build the reference before the edit

If the still needs a reusable character reference or a better viewpoint first, solve that before Nano Banana.

Build a reusable character reference

Lock identity, outfit, and reference quality.

Change the viewpoint before the edit

Change the viewpoint before you spend video credits.

Open Image

Build or clean the source still first.

Tips and limits

GPT Image 2 performs best when the prompt separates subject, composition, exact text, and constraints.

What works best

Use High quality for typography and product-final renders.
Use Low or Medium for cheaper layout exploration.
For edit prompts, write what should change and what must remain unchanged.
When using a mask URL, keep the mask public and aligned with the source image.

Hard limits to keep in mind

Still images only.
The MaxVideoAI composer currently allows up to 4 outputs per request.
Edit mode requires at least one reference image.
Mask input is URL-based.

Safety and likeness

No sexual content or content involving minors.
No hateful, harassing, or graphic-violence requests.
Do not impersonate real people or public figures without consent.
Do not include private personal data or sensitive documents.

FAQ

Does GPT Image 2 support text-to-image and image-to-image?

Yes. MaxVideoAI exposes one GPT Image 2 model with text-to-image mapped to openai/gpt-image-2 and image-to-image mapped to openai/gpt-image-2/edit.

Which Fal controls are available?

The Image Lab exposes Fal image_size presets, canonical sizes 1024x768, 1024x1024, 1024x1536, 1920x1080, 2560x1440, and 3840x2160, a Custom width/height mode, quality, num_images, output_format, reference image inputs for edits, and optional mask_url.

How much does GPT Image 2 cost?

Fal pricing follows the quality x size table: 1024x768 costs $0.01 / $0.04 / $0.15 for Low / Medium / High; 3840x2160 costs $0.02 / $0.11 / $0.41 before MaxVideoAI account pricing rules.