It's the successor to the 1.5 release — a fresh visual model that pushes resolution up to 4K, fixes most in-frame typo issues, and accepts up to 16 source photos for reference editing in one job. On this page it runs through Kie credits, so you don't need a personal OpenAI API key to try it.

The older release topped out around 2K and struggled with long words inside the frame. The new model hits 4K on the long edge, lays out wordmarks and short paragraphs cleanly, and lets you stack 16 references instead of just a few. Most teams notice the win first on posters and product mockups.

Switch to reference mode in the panel above, drop in up to 16 source pictures, then describe the change you want. The model will keep the layout and identity cues you give it and apply your edit. Re-uploading the previous result as a reference is a good way to iterate.

Yes. This page wraps the model behind Kie credits, so you can run a render without touching the OpenAI API directly. New accounts ship with a small free balance to test before you commit, which is handy if you just want to compare it against another AI image generator.

Two limits: auto aspect ratio is capped at 1K, and the 1:1 ratio is capped at 2K. To unlock the full 4K render, pick a non-square ratio like 9:16, 16:9, 4:3 or 3:4 and then choose 4K in the resolution selector.

Yes — that's one of the headline upgrades. Headlines, UI labels, brand wordmarks and short paragraphs land cleanly the first time, so you don't burn renders fixing typos. Long paragraphs of body copy are still risky, so keep blocks of text short for best results.

Up to 16 in reference mode. That gap matters: it lets you lock a character, outfit or product across many shots in a single batch instead of redoing the prompt for every variant. It also helps when you need a consistent brand look across a campaign.

ChatGPT uses a related family of models, but on this page you get direct control over aspect ratio, resolution and reference count. If you need a specific 4K size or precise reference editing, the standalone tool is more flexible than the chat experience.

Nano Banana 2 leans into stylized art and quick character edits. The OpenAI model is stronger when you need crisp wordmarks, photoreal detail and a true 4K file. Plenty of teams keep both around and pick per job — try both with the same prompt to feel the gap.

Six in total: auto, 1:1, 9:16, 16:9, 4:3 and 3:4. Note again that 1:1 caps at 2K and auto caps at 1K, so pick a non-square ratio whenever you need a print-ready render.

Yes. The Kie job for this model accepts an optional callback URL. Submit the task, walk away, and your backend will get pinged the moment the render is ready — no polling required. The async flow is identical to every other model in the platform.

Posters, ads, social hero shots, product mockups, slide decks and storefront covers. Anywhere you need readable in-frame text, sharp 4K detail or a consistent character across many variants, this model holds up well — even against fast-moving rivals like Nano Banana.

It accepts up to 20,000 characters per prompt, which is more than almost anyone needs. Stick to clear, scene-first phrasing — describe subject, environment and mood in that order — and you'll get a closer match than padding the brief with adjectives.

You'll need a free Kie account to start, but no credit card and no OpenAI API setup. New accounts get a small credit balance, so you can run several renders, compare it against other tools, and decide whether to top up — all without any commitment.