

ByteDance Model for Images



Subject shots with light as on the set and scenes according to the text.
Choose a version for the task: draft, photo editing or a complex scene.
| Revision | Why | When not to take |
|---|---|---|
| Nano Banana 2 Lite | Quick Drafts | Need 2K, 4K or small text: Lite only gives 1K |
| Nano Banana 2 | Edit photos and pictures for every day | Complex scene with small details or long text: take Pro |
| Nano Banana Pro | Complex scenes, details, long text | Need a cheap draft of the idea: take Lite or Muse Image |
| How they search | That | In our chat |
|---|---|---|
| Nano Banana, Nano Banana 2.5 | First version, Gemini 2.5 Flash Image | None — Nano Banana 2 is the closest |
| Nano Banana 2 | Gemini 3.1 Flash Image | Yes — Nano Banana 2 |
| Nano Banana 2 Lite | Gemini 3.1 Flash Lite Image | Yes — quick drafts, 1K |
| Nano Banana Pro, Nano Banana 3 Pro | Gemini 3 Pro Image | Yes — Nano Banana Pro |
| Nano Banana 2.1 | This is how the version is called on third-party sites | None — there is no such version in the chat |
One selfie and one query: the first result of each version, without selection. Look what's more important to you.

Keep the girl's face with the photo unchanged. Portrait in a Parisian cafe by the window, morning soft light, beige trench coat, cup of coffee in hand, film photo, frame to the waist.



The model is paid. After registration, free pictures are available in Muse Image.

0 ₽at Muse Image

first package













Take a request, insert your photo and repeat in the chat with one click.

Fashion lookbook, flat-lay strictly top-down, vertical 3:4 format. On a solid beige background a summer outfit neatly arranged: white linen shirt, light blue denim shorts, woven straw bag, caramel leather sandals, tortoiseshell sunglasses, straw hat, thin gold bracelet and eucalyptus sprig. Balanced composition with space between items, soft natural shadows, diffused daylight, realistic fabric and weave textures. Product shoot for a magazine, clean warm tones. No text, inscriptions, brand logos or watermarks.

Architectural visualization, vertical 3:4 format. Small modern wooden house with gabled roof and large panoramic windows in a pine forest. Charred dark wood facade, wooden deck, warm interior light. Tall pines, moss, ferns, light mist between trunks. Early evening, soft diffused light, warm interior contrasts cool forest tones. Eye-level angle slightly low, 24mm lens, straight verticals. Photorealistic render, realistic materials. No text, inscriptions, logos or watermarks.

Three friends sit by a campfire in the forest at night laughing, warm firelight on faces, cinematic shot

Street food festival poster: bright illustrations of burgers and tacos, large headline “FOODFEST”, date “June 7–8”

Lookbook: man in light beige linen suit on a Mediterranean street

Plate of ramen from above, steam, chopsticks, dark background, food photography
| What has gone so wrong | What to write | The “way of escape” |
|---|---|---|
| The face changed after a big edit | "Leave your face exactly as in the photo: features, nose shape, eye color, hairstyle. Change only the background and clothes.” | Take one or two clear full-face photos, not ten random ones. |
| In a group shot, people are similar or someone is missing | "Photo 1 — Anya, photo 2 — Oleg, photo 3 — Masha. Put all three next to each other, everyone's face is like in his photo." | Without captions, the model does not know whose face is from which picture. |
| The whole picture changed, and you asked for one thing | “Change only the color of the jacket to blue, leave everything else as it is.” | Quote the result in the chat and write a dot edit. |
| The text in the picture turned out to be crooked | "The inscription is exactly: "Happy Birthday, Mom!", there are no other inscriptions." | Exact text in quotes; for long captions, take Nano Banana 2 or Pro, not Lite. |
| Frame cropped or wrong format | Select the format in the chat list before sending, and not in the request body. | So the price and size are visible in advance. |
| Few pixels to print | Choose 2K or 4K in Nano Banana 2 or Pro. | The price of each quality is visible before shipment. |
Each new attempt is paid. If the generation failed for a technical reason, the tokens will be returned.
Same selfie, same request. The first result of each model, without selection.
You can change the model in the same chat, without a new registration. More neural networks and a task table — on the page "Neural networks for photos".
No. Nano Banana is a model of Google images from the Gemini family; it is officially available in the Gemini app and Google AI Studio. In SelfyStudio, the same model works via API in a shared chat next to GPT Image 2 and Seedream, payment for each picture is in rubles, no Google account is needed.
Lite — drafts, only 1K, ₽9.3 per picture. Nano Banana 2 — for every day: photo editing, 1K/2K/4K, from ₽27.9. Pro — complex scenes, small details and inscriptions, up to 4K, from ₽55.9. All three have up to 14 photos in the request. The first Nano Banana (2.5) and version 2.1 are not in the chat, "3 Pro" is our Pro.
The model itself is paid: from ₽9.3 per picture in Lite and ₽27.9 in Nano Banana 2. Free after registration — 3 pictures in Muse Image and 3 photos in photo shoots, not in this model. The first package of ₽149 is about 5 pictures in Nano Banana 2. Price visible before dispatch, no subscription available.
Attach one or two clear full-face shots, write what to change and what to leave: "change the background for the evening city, face, hairstyle and pose do not change." Select Nano Banana 2, the format is in the chat list. If the traits have floated, click "Edit" and write "return the original face one to one".
Yes, up to 14 photos in one request. Sign who is where: "photo 1 — Anya, photo 2 — Oleg, put them next to the window of the cafe, faces as in the photo." Without this, the model itself will decide who to put where. For two or three people, one clear picture is enough for each.
Short inscriptions — yes: enclose the exact text in quotation marks and write "there are no other inscriptions". It is better to give the long layout of the poster to Nano Banana Pro or GPT Image 2 — they hold a small font. If the letter floats, click "Edit" under the picture and ask to rewrite only the inscription; each attempt is paid separately.
SelfyStudio does not add visible icons. Google can leave an invisible SynthID mark in the images of its models: it is not visible to the eye and does not interfere with printing. The original photo is not overwritten, the result is a separate frame in your account.
Requests pass the rules of the service: we do not make sexual content with real people and children, forgery of documents and the like. A rejected request does not debit the tokens, and if a technical error occurs, they will be returned automatically. If the model refused, rewrite the query without the names of real people and frank details.