GPT Image Comes to Cogny: A Second Image Engine Next to Nano Banana
Cogny's built-in image generator has run on Google's Nano Banana since launch. It's fast and cheap and handles product shots well, but like most image models it struggles with text: ask it for a banner reading "Spring Sale −30%" and you'll often get "Spirng Sael".
So the image generator now has a second engine: OpenAI's GPT Image. It's built in, there's nothing to connect, and the agent can use it anywhere Cogny already generates images: in-app chat, loops and scheduled reports, ticket runs, and any agent on app.cogny.com/mcp.
TL;DR
- New tool:
generate_image_openai, next to the existinggenerate_image(Nano Banana 2) in the Image Gen & Media built-in. - Pick it for text in images: headlines, prices, labels, UI and infographic mockups, and prompts with lots of precise details. Nano Banana stays the default for everything else.
- Same inputs as Nano Banana: the same ten aspect ratios, reference images, and your uploaded brand assets. Without references, your brand colors and guidelines are applied automatically.
- Two quality levels:
low(about 10 seconds, for drafts) andmedium(about 30 seconds, the default). - Priced on what you generate: a low-quality image is 1 credit, and a medium square image is about 1 credit. Reference images add a fraction of a credit each.
When to use which engine
Nano Banana 2 (generate_image) | GPT Image (generate_image_openai) | |
|---|---|---|
| Readable text in the image | Unreliable | Strong |
| Many precise details in one prompt | Good | Strong |
| Speed | Fastest | ~10s (low) / ~30s (medium) |
| Brand assets and reference images | Yes | Yes |
| Aspect ratios | 10 (1:1 → 21:9) | The same 10 |
You don't have to choose. Ask for "a LinkedIn ad that says Book a demo in our brand colors" and the agent goes to GPT Image because the image needs words in it. Ask for a lifestyle shot of your product and it stays on Nano Banana. If you have a preference, just say it: "use GPT Image" works. And if one engine turns a prompt down, the agent can try the other.
Brand assets work the same way
Upload logos, founder photos and product shots once (Settings → Brand → Brand Assets, or from any agent with upload_file). Both engines can then use them as references. A GPT Image request with brand_asset_ids can put your actual logo on a mug, a billboard or an ad mockup, and the text around it comes out spelled correctly.
What it costs
GPT Image is billed on what each image actually costs to generate, converted at Cogny's standard credit rate:
- Low quality: 1 credit.
- Medium quality: about 1 credit for a square image. Wide and tall formats cost the same or less.
- Reference images: a fraction of a credit each.
Credits come out of the same balance as the rest of your plan. On Cogny AI, Pro, Cloud and AppSumo plans, that's your included monthly credits first, then top-ups. On Solo and Free, it's your top-up (or signup) credits. Every image is checked against your balance before it's generated, so you're never charged for an image you couldn't afford, and invalid requests are rejected without charge.
Try it
In Cogny chat, or from Claude Code / any MCP client connected to app.cogny.com/mcp:
Make a 4:5 Instagram ad for our spring sale with the headline "Spring Sale −30%" and our logo. Use GPT Image.
The agent calls list_brand_assets, then generate_image_openai with your logo as a reference, and hands back a hosted image URL that's ready for Meta Ads, Google Ads or your CMS.