Turn Your Photo into an AI Figurine — The 2026 Prompt and Tool Guide

Last updated: 2026-08-25 (prices and free-tier limits below were verified on that date)

Summary: what actually makes an AI figurine in 2026?

• The "1/7 scale figurine" prompt that went around in 2025 still works word for word. What changed is not the prompt but the model running it.
• Trap one — "Nano Banana" now means something else. In 2025 posts it refers to Gemini 2.5 Flash Image. As of August 2026 it means Nano Banana 2 (Gemini 3.1 Flash Image) or Nano Banana Pro (Gemini 3 Pro Image). Follow an old guide and you will hunt for a menu item that no longer exists.
• Trap two — turning the watermark off does not remove the watermark. Since 2026-08-14 Google lets you disable the visible one, but invisible SynthID and C2PA provenance metadata are still embedded, and cropping does not strip them.
• The cheapest route is the Gemini app: free users get roughly 20 images a day with Nano Banana 2, capped at 1K resolution.
• Labeling rules for synthetic media are arriving jurisdiction by jurisdiction — Korea's AI Framework Act took effect on 2026-01-22, and it won't be the last. Captioning an image as AI-generated costs nothing and covers most of it.

The name changed twice

When this trend started, Nano Banana was the nickname for Gemini 2.5 Flash Image. Not anymore.

Upload a photo in the Gemini app today and Nano Banana 2 handles it by default. Reach for Pro when you need higher resolution or tighter detail. A 2025 guide telling you to pick "2.5 Flash Image" is describing an option that is gone.

The rest of the lineup shifted too. Google's Imagen family entered staged shutdown on 2026-08-17. On the OpenAI side, DALL·E 3 was retired in May 2026 and GPT Image 2 took over — that one comes from industry coverage rather than a primary announcement, so treat it as reported rather than confirmed. Midjourney is on V8.2, Adobe Firefly on Image 4 / 4 Ultra. If you self-host, the current names are FLUX.2, Qwen-Image 2.0, and Seedream 4.5.

From one photo to one figurine

Pick the photo. Face roughly front-on or slightly turned, framed from the chest up rather than cropped at the chin, and no hard shadow cutting across the face. Those three choices decide about half the result. If several people are in the frame, the model has to guess which one you meant.

Upload and paste. The flow is identical in Gemini, ChatGPT, and Grok: attach the image, then paste one of the prompts below. Keep them in English — translated versions are noticeably less stable.

Iterate. A first pass rarely lands. Follow up with short corrections: make the face closer to the original, make the base clear acrylic, move the packaging box to the left. Nano Banana Pro and Grok Imagine 2.0 also let you mark a region and edit only that area.

The three prompts

1. The original 1/7 scale commercial figurine

Create a 1/7 scale commercialized figurine of the character in the picture,
in a realistic style, in a real environment. The figurine is placed on a
computer desk. The figurine has a round transparent acrylic base. The
content on the computer screen is a 3D modeling process of this figurine.
Next to the computer screen is a toy packaging box, designed in a style
reminiscent of high-quality collectible figures, printed with original
artwork. The packaging features two-dimensional flat illustrations of
the figurine.

2. PVC anime figure

A masterpiece 1/7 scale PVC anime figure of the person in the photo.
The figure is highly detailed with dynamic flowing clothing, intricate
paintwork, and a flawless matte skin finish, standing on an elaborate
decorative circular base. Professionally photographed in a lightbox with
a macro lens, realistic plastic and resin textures, masterpiece
collector's item.

3. Boxed action-figure style

1/7 scale PVC figure of the person in the photo, resin kit, high gloss
finish, detailed painting, plastic blister-pack toy box with printed
artwork and window, collectible packaging design, dynamic pose,
showcase lighting, glass reflection, premium collectible.

All three are unchanged from 2025. Multiple 2026 write-ups report them working in Nano Banana 2 and Pro, GPT Image 2, and Grok Imagine 2.0.

Tools and what they cost

The free ceiling is the Gemini app: around 20 images a day for free users on Nano Banana 2, up to 1K resolution. Google AI Plus is 10 a day, Pro 100 a day, and so on by tier. These numbers move often — read them as verified on 2026-08-25.

For volume or scripted generation, go to the API. Nano Banana Pro is reported at roughly $0.13 per image at 1K–2K and about $0.24 at 4K, and as of July 2026 it sits outside the free API tier. A hundred images is about thirteen dollars, which means scripting pays for itself sooner than most people expect.

When it comes out wrong

If the face doesn't look like the person, suspect the source resolution first — a 512px profile picture is the usual culprit. Busy backgrounds make the model misread the subject's outline. Glasses, masks, and hats get dropped during stylization, so name them explicitly if you want them kept. Attempts at celebrity likenesses are filtered at the model level.

Watermarks and disclosure

Your output already carries a watermark. Since 2026-08-14 Google lets you switch off the visible mark, but only the visible one. Invisible SynthID and C2PA provenance metadata stay embedded and survive cropping and resizing. "I removed the watermark, so nobody can tell" does not hold.

Disclosure rules are catching up. Korea's AI Framework Act took effect on 2026-01-22 with its enforcement decree, and it is a useful preview of where labeling requirements are heading. Under it, generative AI output must be labeled: ordinary output can use either a visible label or watermark/metadata, while deepfake-style content that realistically reproduces a real person's face reportedly requires a clearly visible label. Korea has set a grace period of at least a year before fines apply, with a ceiling of 30 million won. If you publish to a Korean audience, that is the regime you are in; elsewhere the details differ but the direction is the same. A one-line caption saying the image is AI-generated is the cheap answer everywhere.

Someone else's photo is a different question. Likeness and publicity rights apply independently of any AI statute. Turning your own selfie into a figurine for your own use is low risk. Doing it to a friend or coworker without asking, then posting it, is not. Asking first remains the cheapest safeguard.

This is a summary of published effective dates and reporting, not legal advice. For commercial use or anything with real dispute potential, check with a professional.

Sources

• Google Gemini image-generation help