AI image prompt guide · 30 September 2026

AI Figurine Prompt: Turn a Photo Into a Collectible 3D Figure

Turn a permitted reference photo into a collectible figurine concept with prompts for boxed toys, desk displays, anime variants and brand-safe character design.

GPTPrompts.AI Editorial

Practical image-prompt workflows and source checks · Last updated September 30, 2026

A good AI figurine image is a product brief, not just a request to make someone into a toy. You need to describe the subject, pose, material, scale, base, camera angle and packaging separately. If you upload a real person, get permission and preserve the distinction between a playful concept and a real product. ChatGPT Images supports image uploads and edits; Google’s current Nano Banana 2 model supports image input and image generation. Neither makes a generated figure a manufactured object, a licensed character or an accurate commercial package.

What to prepare before you prompt

Decide whether you are generating from words, editing an uploaded image, or combining both. For a person or pet, use a clear reference and permission. For a product, remove private order information and decide whether accuracy or mood matters more. For a social post, choose the final aspect ratio before you start. A square prompt is not the same as a 9:16 story prompt: composition, hands and empty space all change.

Write the brief in five parts: subject, composition, physical style, constraints and output shape. “A Polaroid photo of me” or “make me a figurine” covers only the subject and trend label. The prompts below add the missing production decisions. Copy one, replace the bracketed detail, then change one variable at a time. If you change the subject, camera angle, lighting, material and background in one turn, you will not know which instruction fixed or broke the image.

How to use these prompts

Open the image feature in the tool you have access to. In ChatGPT, you can ask for an image in a conversation or open Images; in Gemini, use the current image-generation surface or API model available to your account.
Upload only the reference you are allowed to use. Say what must remain unchanged before describing the transformation.
Paste one prompt and inspect the first result for identity, anatomy, text, logos, crop and the requested visual cues.
Make a surgical follow-up: “Keep everything else unchanged; fix only the lower border,” or “keep the face, change only the jacket.”
Save the prompt with the final image and disclose that it is AI-created when a viewer could mistake it for a real photograph or product.

Copy-paste prompt library

1. The boxed collector figure

Use this as the neutral base before adding a profession or costume.

Using the uploaded reference only for the consenting adult’s general likeness, create a premium collectible figurine in a clear retail display box. Stylised but recognisable face, simplified toy proportions, matte painted resin, standing on a small circular base, neutral pose, soft studio lighting, three-quarter camera angle. Add blank packaging panels with no brand, logo, name or readable text. Show the whole box, realistic plastic reflections and no extra accessories. 4:5.

2. Desk-display version

Better for a personal keepsake than a fake retail package.

Turn the supplied photo of a consenting adult into a small desk figurine concept, about 15 cm tall, stylised realistic proportions, matte vinyl material, stable rectangular base, natural friendly pose and a tiny desk scene behind it. Preserve the person’s distinctive hairstyle and clothing colours without making an exact duplicate of a uniform or logo. No packaging, no text, no watermark, no extra fingers. 1:1.

3. Profession without a real employer logo

Describe tools and posture, not protected marks.

Create a collectible figurine concept based on the supplied consenting adult. Dress the figure as a generic civil engineer with a hard hat, rolled plan, safety vest with no logo and a small transparent measuring tool. Matte painted resin, articulated toy details, sturdy base, neutral studio background, three-quarter view. Keep the face recognisable but not photorealistic. No employer name, trademark, serial number or readable text. 4:5.

4. Creative professional

Good for an illustrator, photographer or designer profile image.

Using the supplied reference, create a stylised collectible figure of a creative professional holding a sketchbook and pencil, with a small camera bag at the feet. Preserve the adult’s face and hair as a simplified toy likeness. Soft painted resin, warm studio light, visible base, balanced pose, no brand marks and no writing on the sketchbook. Plain background. 4:5.

5. Anime-inspired variant

Say inspired by animation rather than naming a living artist.

Transform the supplied consenting adult reference into an original anime-inspired collectible figure, large expressive eyes but recognisable hairstyle and face shape, clean sculpted hair, layered jacket, small display base and dynamic but stable pose. Use painted PVC texture, soft rim light and a simple backdrop. Do not copy an existing character, costume, logo or artist style. No text. 4:5.

6. Family-safe family set

Keep the composition simple and avoid turning children into products without guardian permission.

Create a family collectible set from the supplied reference photo, only with permission from every identifiable adult and guardian approval for any child. Preserve each person as a distinct simplified figure, standing on one shared base with comfortable space between them. Soft painted resin, coordinated neutral clothing, no logos, no names, no text, no exaggerated body changes and no extra people. 4:5.

7. Pet figurine

Name markings and anatomy before adding a fantasy accessory.

Using the uploaded pet photo with the owner’s permission, create a collectible vinyl figurine that preserves the animal’s coat pattern, ear shape and body proportions. The pet sits on a small round base beside a simple miniature food bowl. Painted vinyl, soft studio light, three-quarter view, no costume, no brand, no words and no extra limbs. 1:1.

8. Pet and owner pair

The identity-preservation line matters when two faces are close together.

Create a friendly desk-display figurine pair from the supplied owner and pet references, with permission. Keep the adult’s face and the pet’s markings distinct. The adult kneels beside the seated pet, both simplified into painted resin, one shared oval base, warm studio light and clean background. No leash text, logo, extra animal, face blending or words. 4:5.

9. Box with accessory tray

Describe the layout so the model does not hide the accessories behind the figure.

Create a retail collectible figure concept in a clear box: the main stylised figure stands on the left and four separate accessories sit in a visible tray on the right: notebook, coffee mug, headphones and small plant. Use the supplied consenting adult only for a simplified likeness. Matte resin, blank packaging, no logo or readable text, accurate accessory count, front three-quarter view. 4:5.

10. Travel-themed figure

Use generic gear instead of copying a brand’s luggage or badge.

Turn the supplied consenting adult into an original travel-themed collectible figure with a small backpack, camera, folded map with no writing and a generic water bottle. Preserve the face as a stylised likeness, keep hands plausible, use matte painted resin and a sturdy base. Warm studio lighting, clean background, no brands, flags, passport details or text. 4:5.

11. Sports pose

Ask for a stable base when the pose is energetic.

Create a collectible figurine based on the supplied consenting adult in a dynamic but physically stable running pose on a small oval base. Generic sports clothing with no team logo, painted resin, simplified but recognisable face, believable shoe contact, clean studio background and three-quarter camera angle. No trophy text, brand marks, numbers or extra limbs. 4:5.

12. Brand-safe mascot

Build a fictional mascot rather than imitating a known character.

Design an original friendly mascot figurine for a fictional neighbourhood bakery: a cheerful rounded animal baker holding a blank tray, simple apron with no letters, warm painted ceramic texture, stable base, soft studio light and clear silhouette. It must not resemble a known cartoon character or use a real logo. No text. 1:1.

13. Before-and-after concept

Keep the two states in separate panels so the transformation is understandable.

Create a two-panel product-concept image. Left: the supplied consenting adult as a normal portrait reference, unchanged. Right: an original stylised desk figurine based on that person, with simplified proportions, painted resin, base and soft studio lighting. Label neither panel; use no text or logos. Do not alter the person in the left panel or claim the right panel is a manufactured product. 16:9.

14. Transparent prototype

Useful for an internal concept board.

Create a clean concept render of a stylised collectible figurine based on the supplied consenting adult, shown on a transparent-looking acrylic base against a pale neutral background. Preserve the person’s distinctive hair and clothing palette, matte resin with a few glossy accents, front three-quarter view, soft shadows. No packaging, words, logos or invented accessories. 4:5.

15. Limited-colour ceramic

Material language helps more than the word cute.

Using the supplied reference with permission, create a small ceramic figurine with a limited palette of cobalt, cream and terracotta. Keep the face as a simplified likeness, use hand-painted brush variation, a flat oval base and a calm seated pose. Product photography on a neutral tabletop, no text, no maker mark, no logo and no photoreal skin. 1:1.

16. Multiple outfit options

Ask for separate miniatures, not one body wearing impossible layers.

Create a clean four-view concept sheet of the same stylised collectible figure based on the supplied consenting adult: casual outfit, work outfit, festival outfit and rain outfit. Keep face, hair and body proportions consistent. Each view has its own small base, equal spacing and blank background. No labels, text, logos, weapons or trademarked costume elements. 16:9.

17. Packaging edit

Use an edit when the figure is good but the package is wrong.

Edit only the packaging around the supplied figurine concept. Keep the figure, pose, face, colours and accessories unchanged. Replace the package with a simple clear box, blank cream card insert, visible accessory tray and no brand, name, barcode, logo or readable writing. Preserve lighting and camera angle. Do not add objects.

18. Manufacturing reality check

Ask for a concept board rather than pretending the model has validated production.

Create a concept-board render for a possible collectible figurine based on the supplied consenting adult. Show front, side and back views of the same simplified figure, with a separate callout-like blank area for notes but no text. Keep anatomy, base and accessories consistent. Use matte resin, neutral studio light and no claim of manufacture, safety testing, licensing or retail availability. 16:9.

Review checklist before sharing

Start with the literal subject. Does the person still look like the permitted reference? Are all people and pets present, and are their identities separate? Check hands, eyes, ears, jewellery, repeated objects and the number of accessories. A polished style can make a structural mistake easier to miss, so zoom in before posting.

Next check the visual brief. For an instant print, look for a border, direct-flash character, restrained grain and a blank caption area. For a figurine, check the base, material, scale cues, stable pose and accessory count. If the prompt asked for a blank sign or package, inspect every surface for accidental letters or logos.

Finally check context and rights. Do not present an invented scene as a real memory, historical record, testimonial or product photograph. Do not imply that a real company endorsed a generated figure. Avoid uploading sensitive identity documents, private addresses or other personal information simply to chase a trend. Keep an original copy of the reference and the prompt, and use a normal editor for captions and disclosures.

A practical iteration loop

Use a three-pass loop instead of asking for a perfect image in one enormous prompt. Pass one is composition: get the crop, number of subjects, pose and major objects right. Ignore small style defects while the structure is wrong. Pass two is preservation: tell the model exactly what to keep from the reference and exactly one thing to change. For example, keep the face and clothing, change only the background; or keep the figurine and accessories, change only the package insert. Pass three is finish: tune grain, reflections, colour temperature, empty space or border width. Saving each prompt makes the change history inspectable.

When a result fails, describe the failure as an observable property. “The face changed” is more useful than “make it better.” “There are five accessories but I asked for four” is more useful than “clean this up.” “The lower caption has invented letters” tells you to request a blank area and add type later. This language also makes prompts portable across tools because it describes the output rather than relying on a vendor-specific control that may move or disappear.

For a reference image, put preservation rules first and the creative transformation second. Name people by position when there are multiple faces. Name a pet’s markings, not just its breed. For products, state whether the supplied shape, colour and label must remain accurate. For a fictional subject, say that it is fictional so the model does not accidentally imply a real person or event. If the model keeps changing a detail, simplify the scene and rebuild from a cleaner base image.

For public-facing work, create an approval copy before the final export. Ask a colleague or the subject to check likeness, context and caption. If the image could be mistaken for a photograph, add a short disclosure in the post or surrounding copy. If the result is a figurine concept, call it a concept render. If it is a childhood-style scene, call it an imagined recreation. Clear labels protect viewers from a false impression without taking away the playful nature of the work.

Prompt anatomy you can reuse

Subject: identify the person, pet, object or fictional character and state what reference is allowed to control. Composition: set the crop, camera angle, position and negative space. Material or photographic language: specify resin, ceramic, vinyl, paper, direct flash, film grain or studio light. Constraints: list what must not change, including faces, markings, logos, text, accessory count and background objects. Output: finish with the aspect ratio and any export intention. This order gives the model a readable hierarchy and gives you a checklist for review.

Do not pile on synonyms to create intensity. “Ultra-realistic, hyper-real, cinematic, photoreal, 8K, masterpiece” does not replace a lighting direction, a material, a crop or a constraint. One precise sentence about a blank package is stronger than five style adjectives. Likewise, a named lens can suggest depth, but it cannot guarantee a physically accurate camera simulation. Treat these words as visual cues and inspect the image rather than promising a technical property the model did not verify.

Keep private working notes separate from the shareable prompt. The prompt may need a person’s first name or a production reference while the published example should use placeholders such as [person], [pet] or [product]. Remove addresses, document numbers, private messages and unneeded metadata from uploads. If you are preparing a client or campaign asset, preserve the client’s approval and the final source file outside the image model as part of your normal content process.

What these tools cannot guarantee

Image generation is non-deterministic. The same wording can produce different composition, text, hands or facial details. A model may also extend an edit beyond a selected region. Treat a good-looking result as a draft to inspect, not as proof that the requested details are correct. Product packaging, uniforms, logos, children’s likenesses, historical scenes and real-person transformations need extra human review.

Availability and controls can change. OpenAI’s current help documentation says ChatGPT Images is available on web, iOS and Android and that image creation and editing are available across tiers, while some features such as Images with thinking have narrower availability. Google’s current documentation identifies Nano Banana 2 as Gemini 3.1 Flash Image, with the stable API ID gemini-3.1-flash-image. Check the linked documentation before promising a user a specific button, model, price or resolution.

Related guides and sources

For broader image workflows, see ChatGPT image prompts, Nano Banana 2 prompts, the Nano Banana prompt library and our transparent image-tool comparison protocol. Those pages cover editing, reference images, products, portraits and fair evaluation without claiming fabricated benchmark results.

Frequently asked questions

Can I turn any person into a figurine?

Use an identifiable person’s photo or likeness only with permission. Be especially careful with children, private individuals, public figures and images that imply endorsement. A playful concept is not permission to sell a likeness, use a logo or suggest that a real product exists.

Why does the packaging contain nonsense text?

Rendered lettering is still unreliable. Ask for blank packaging and add a real name or label later in a design tool. This also prevents accidental logos, fake safety claims and invented product information.

How do I make the figure look like a collectible?

Specify a material, scale, base, camera angle, lighting and accessory layout. ‘Make me a toy’ leaves too many decisions open. Tell the model what must stay consistent and what it is allowed to stylise.

Is Gemini or ChatGPT better for this prompt?

Both can be used for image creation and editing, but the current interfaces and model names differ. ChatGPT Images supports uploads, selection edits and aspect-ratio controls. Google’s Nano Banana 2 is the current Gemini 3.1 Flash Image model. Choose based on access and review the result rather than assuming a universal winner.