Free users generate with a site watermark.
Upgrade to remove watermarkGenerate an image and it will appear here. Your full history is saved in My Creations.
Write a prompt, or upload references, to generate and edit stills. Choose 1K–4K and the ratio for posters, product scenes, character sheets, and wide campaign frames.
Text and image references · 1K–4K · Ultra-wide ratios
Related website: nano banana 2
Open Nano Banana 2
Text-to-image and image-to-image share one model: low-latency generation, subject consistency, in-image type, and production resolution and aspect ratios.
Low-latency generation and editing with stronger lighting, richer texture, and sharper detail. Spatial layout and multi-part instructions are followed more strictly, so product, interior, and editorial stills stay closer to the requested composition.
Generate a still
One workflow can keep up to five characters recognizable and up to fourteen objects visually stable. Scene, pose, and camera may change; input identity and object appearance do not. Image-to-image on this page accepts up to fourteen references.
Edit with references
Renders accurate, legible type in mockups, greeting cards, posters, and labels, and can translate or localize copy already in the frame. Short headlines and structured labels are the intended case.
Set on-image type
Resolutions: 1K, 2K, 4K. Aspect ratios include 16:9, 9:16, 21:9, and ultra-wide 4:1, 1:4, 8:1, 1:8. Generate to the target crop for social, banners, and widescreen frames. Higher resolution increases generation time and credits.
Set output format
Start from a prompt, or upload one to fourteen references and describe the change. The steps match the generator on this page.
Use text-to-image for a new still. Use image-to-image to preserve an existing face, product, or layout. This page locks the model to Nano Banana 2 so parameters stay within range.
State subject, setting, required lettering, crop, and locked elements. Quote on-image copy. If references are uploaded, assign each image a role in the prompt.
Generate the first pass at 1K. Change lighting, copy, or background in the next round, not all three. Confirm composition before exporting 2K or 4K.
| Criterion | Nano Banana 2 | GPT Image 2.5 | GPT Image 2 |
|---|---|---|---|
| Primary use | Rapid drafts, storyboards, short type layouts | Reference edits and multi-step campaigns | Everyday stills and restyles |
| In-image text | Short headlines and localized labels | Posters, UI, mixed type and image | Simple titles; long copy needs review |
| References | Up to 14 images for identity and objects | Up to 16 images, local-edit control | Restyle and background replacement |
| Output | 1K / 2K / 4K; 4:1, 1:4, 8:1, 1:8 | 1K / 2K / 4K; social-ratio set | 1K / 2K / 4K; standard ratios |
| Process | Iterate, then lock crop | Flare for speed; Sunburst for control | Single brief, fewer sequential edits |
Image generation and editing model
Nano Banana 2 is the image model on this page. Write a prompt to generate a still, or upload references to change scene, clothing, or a local detail. Output is a still, not a video. For motion, generate the frame here, then continue in text-to-video or image-to-video.
Use Text to Image for a new frame with no source photo. Use Image to Image when a face, product, or layout must stay recognizable while the background or wardrobe changes. Large identity changes belong in a new Text to Image job, not an edit of the old photo.
Image to Image accepts 1–14 images (JPEG, PNG, or WebP, 30 MB each). In the prompt, say what each image is for and what must stay locked: face, product outline, crop, or existing type. The same set should be reused across a sequence.
Short headlines on posters, cards, and labels are the intended case. Existing on-image copy can also be translated in place. Long paragraphs, small type, and dense tables often misspell or blur. Check prices, names, and translations before you publish.
Resolutions: 1K (13 credits), 2K (20), 4K (30). Ratios include 1:1, 16:9, 9:16, 21:9, and ultra-wide 4:1, 1:4, 8:1, 1:8. Pick the crop in the generator instead of generating square and recropping. Free-plan stills include a site watermark; paid plans can turn it off.
One workflow is designed for up to five recognizable characters and fourteen stable objects. Keep the same references, name each person, and change one variable per round (scene, pose, or camera). Identity is not guaranteed if you change face, age, and wardrobe in one pass.
Stills use the same account, credits, and My Creations library as the video tools. Generate or edit here, download the frame, then continue in text-to-video or image-to-video when motion is required.
Generate with Nano Banana 2
