What this is: 23 copy-paste prompts for GPT Image 2.0. Three are ones I ran myself and compared against Nano Banana Pro. Twenty more cover branding, infographics, packaging, social, and more. Plus the one thing that makes all of them better.
Made by @deepika.builds · instagram.com/deepika.builds
OpenAI dropped GPT Image 2.0 on April 21, 2026.
Nano Banana Pro (Google Gemini's image model) has been the go-to for anything text-heavy. Infographics, brand kits, layouts where consistency matters. That was the rule.
GPT Image 2.0 has apparently changed it. In LM Arena blind tests, early testers are calling the gap between GPT Image 2.0 and Nano Banana Pro as wide as the gap between Nano Banana Pro and DALL-E. Near-100% text accuracy. Up to 10 images in one shot. Thinking mode for paid users that plans the layout before it renders anything.
I tested it against Nano Banana Pro using the same prompts. Here is what worked.
A reference image is worth 500 words of prompt. Stop describing your aesthetic. Show it.
When you combine a reference image with a text prompt, the output goes from generic AI image to something that actually matches your brand. GPT Image 2.0 is especially strong at reference following. Where it matters, I have marked what kind of reference to use.
1. Full brand kit from moodboard
The most complex one I ran. GPT Image 2.0 handled multi-page output, typography, packaging, and lifestyle direction from a single moodboard reference.
Prompt:
Create a polished multi-page, multiple images brand kit for a vintage luxury sunglasses brand based on this uploaded mood board. Include: logo, typography, color, outer case packaging, art direction, website direction, social media templates, lifestyle shoot examples.
Reference: Yes.