GPT Image 2

Overview

GPT Image 2 (snapshot gpt-image-2-2026-04-21) is OpenAI's state-of-the-art image generation + editing model. Replaces GPT Image 1 with higher fidelity, multi-image edit support, and stronger character consistency.

Key Strengths

• Photorealism — significantly improved over DALL-E 3 + GPT Image 1
• Text rendering — carries DALL-E 3's lead in legible signs, labels, typography
• Image editing — native inpainting and multi-image composition via v1/images/edits
• Resolution flexibility — not locked to 1024×1024; multiple aspect ratios
• Character consistency — stronger across iterations of the same subject

API Endpoints

• POST /v1/images/generations — text-to-image
• POST /v1/images/edits — inpaint and multi-image edit
• Available via Chat Completions (/v1/chat/completions) and Responses (/v1/responses) too

Parameters (key)

• quality — low | medium | high
• size — square / landscape / portrait variants
• n — number of images per request
• response_format — url | b64_json
• transparent_background — true / false (PNG output)
• output_format — png | jpeg | webp

Prompting Tips

GPT Image 2 still rewards descriptive natural-language prompts — same playbook as DALL-E 3. Don't over-specify camera technicals; the model interprets style words better:

GOOD:  "A misty alpine valley at dawn, watercolor illustration, soft pastel palette, hand-drawn ink outlines"
LESS:  "alpine, ISO 100, f/2.8, 35mm, golden hour, bokeh"

For text in images, name the exact string and font hint:

"A movie poster reading 'NIGHT TRAIN' in bold red sans-serif at the top, subtitle 'a Marek Komosa film' in elegant serif below"

Pricing (snapshot 2026-04-28)

Token-based — actual per-image cost depends on resolution + quality.

Modality	Input (per 1M)	Cached input	Output (per 1M)
Image	$8.00	$2.00	$30.00
Text	$5.00	$1.25	—

Use OpenAI's image generation calculator for per-image estimates.

Limitations

• No streaming, function calling, structured outputs, or fine-tuning
• Pricing is token-based — harder to estimate per-image without the calculator
• Tier 1–5 rate limits: 5–250 images/min depending on usage tier

When to choose GPT Image 2 vs alternatives

Use case	Pick
Photorealism / detail	GPT Image 2
Inpaint / edit workflow	GPT Image 2 (DALL-E 3 doesn't support)
Text rendering in images	GPT Image 2 ≈ DALL-E 3 (slight edge to GPT Image 2 for layout)
Cost-sensitive bulk generation	Consider GPT Image 1 or DALL-E 3 (cheaper for simple prompts)
Open-weights / local runtime	Flux, SDXL — GPT Image 2 is API-only