How to prompt GPT Image 2

GPT Image 2 is OpenAI's quality tier in IQON's model picker. Use it when you want a different render bias from Nano Banana Pro — especially edit workflows with reference images.

When to pick GPT Image 2 vs Nano Banana Pro

Both support 1K and 2K on IQON; GPT Image 2 also supports 4K at higher credit cost.

Generate mode: same scene grammar

GPT Image models still prefer narrative prompts:

1. Subject and action
2. Environment
3. Lighting and mood
4. Camera / format

Example: Studio product photo of a matte black pour-over kettle on pale oak. Soft diffused overhead light, subtle shadow beneath the base. Straight-on hero angle, sharp focus on the spout curve. Square format.

Edit mode: references matter

When you attach reference images, be explicit about what to preserve vs what to change:

• "Keep the product shape and logo. Replace background with warm kitchen morning light."
• "Same composition as reference 1. Swap packaging color to match reference 2 palette."

IQON sends storage paths server-side — attach refs in the composer, describe roles in plain language.

Quality tier

GPT Image 2 maps resolution + aspect ratio to low / medium / high quality internally. Higher quality = more credits, finer detail. Draft at 1K medium; ship heroes at 2K or 4K high.

Unsupported combinations fall back in code — if a size fails, IQON retries with a safe aspect ratio.

Practical tips

• One change per iteration — multi-edit prompts confuse edit mode.
• Name references by role — "hero product," "style ref," "background mood."
• Aspect ratio early — pick 1:1, 9:16, or 16:9 before generating; less crop waste.

Pair with brand context

IQON's enricher still runs for GPT models — your brand kit colors, voice, and aesthetic get woven into the prompt. You brief the scene; the system briefs the brand.

AI image generation best practices →