An offline alternative to Nano Banana
Nano Banana made editing photos with a sentence feel normal. Kokon does the same kind of edit on your own computer: free, private and without daily limits. Here is what's the same, what's different, and when the cloud is still the better choice. Updated October 21, 2026.
What Nano Banana is
"Nano Banana" is the nickname of Google's Gemini image model. You give it a photo and an instruction in plain words, and it returns the edited photo. ChatGPT does something similar with OpenAI's image model. Both run in the cloud: the photo is uploaded, processed on Google's or OpenAI's servers, and sent back.
What Kokon does the same way
- Edit with a sentence: "remove the people", "make it golden hour", "change the sign to say…".
- Combine photos: refer to them as image 1, image 2: "put the person from image 1 on the beach in image 2".
- Text in images that is spelled correctly.
- Create from text as well as edit.
- No settings to learn. You choose Fast, Balanced or Quality; Kokon chooses the rest.
What's different
| Kokon | Nano Banana / ChatGPT | |
|---|---|---|
| Where it runs | Your own GPU | Google's or OpenAI's servers |
| Your photos | Never leave your computer | Uploaded |
| Limits | None | Daily limits on free plans |
| Cost | Free, open source | Free tier, then a subscription |
| Internet | Only for setup and updates | Always |
| Paint the exact area to change | Yes: Change, Remove, Replace, Add | Partly |
| Hardware | 12 GB GPU or Apple Silicon | Any device with a browser |
| Commercial use of the images | No (model license) | Depends on the plan's terms |
When the cloud is still better
To be fair: cloud models run on data-center GPUs. If your computer has no dedicated graphics card, or you need many images per minute, a cloud tool will be faster. Some complex edits are also stronger on the largest cloud models. Kokon uses Qwen-Image 2.1, the open model that leads the open-weights image-editing rankings, and runs it on the computer you already have.
Try the same prompts
Prompts you'd write for Nano Banana work in Kokon too. A few that work well:
- "Remove all the people from the beach"
- "Make it a snowy winter night, warm light glowing from the windows"
- "Change the gold sign text to 'KOKON CAFÉ' in the same lettering"
- "Turn it into a hand-painted watercolor illustration"
Only have a few words? Turn on Prompt+ and Kokon writes the detailed prompt for you, in any language, using the model it already loaded. More tips in the prompt guide.
What you need
| Works | Recommended | |
|---|---|---|
| Windows or Linux + NVIDIA | 12 GB of VRAM (8 GB in beta) | 16 GB or more |
| Windows or Linux + AMD / Intel Arc | 12 GB of VRAM, Vulkan (beta) | 16 GB or more |
| Mac | Apple Silicon, 16 GB, macOS 15 (slow) | 32 GB or more, Pro/Max/Ultra chip |
| Disk | 15 GB free | 25 GB free |
Speed with the default model: on an RTX 5070 Ti, about 24 seconds for an edit at 1.5K and about 36 seconds for a new image at native 2K. 12 GB cards are a little slower, 8 GB cards take 1 to 2 minutes per edit, and an M2 Pro Mac takes about 4 minutes per image.
Questions
Can I run Nano Banana locally?
No. Nano Banana (Gemini's image model) only runs on Google's servers. Kokon is a free app that does the same kind of edit, from a sentence, with an open model on your own GPU.
Is the quality the same as Nano Banana?
It's close on everyday edits like removing objects, relighting, changing text and restyling, and it can differ on complex edits. Kokon uses Qwen-Image 2.1, the open model that leads the open-weights image-editing rankings.
Does it have limits like the free Gemini or ChatGPT plans?
No. Kokon runs on your own graphics card, so there are no credits, queues or daily limits.
Can I use the images commercially?
Not with the default model: Qwen-Image 2.1 is under the Qwen Research License, which doesn't allow commercial use. The app itself is open source (AGPL-3.0).
Try Kokon
Free and open source for Windows, macOS and Linux. One installer; Kokon sets up the model for your graphics card.