To get a consistent AI character in different poses, the usual approach is to pin down the face with a reference image of an earlier result, a character sheet showing the front, side, and back, or extra training on images of the character (LoRA). None of these guarantees exactly the same face every time, though. Plan on comparing the results and fixing differences one at a time.
Figmee accepts illustration-style character images made with AI (generation stops for images that look like photos). Once you've made a figurine-style image, you can register it as a character and create new poses while Figmee tries to keep the character's look. Turn it into a 3D model and you can rotate it in your browser to see the profile and back that the drawing never showed.
This guide covers why the face keeps changing, how the common consistency methods compare, how to register a character in Figmee and create new poses, how to use 3D to check the side and back, how copyright treats AI-made art, and which images you can upload to Figmee.
Note: Figmee currently offers figurine-style images created from your drawings, 3D model data (GLB / 3MF), and turntable videos of your 3D models. Ordering physical 3D prints is Coming Soon. Figmee can't turn photos into 3D models.
Why does an AI character's face change every time?
Image generators create a fresh picture for every request, so even the same prompt can shift the shape of the eyes, the flow of the bangs, or the pattern on an outfit. Keeping a character's look stable across many images is a known challenge even for the companies building these models. When Google announced its image model Gemini 2.5 Flash Image, it described maintaining the appearance of a character or object across multiple prompts and edits as a challenge, and said it was working toward even more reliable character consistency (source: Google Developers Blog, "Introducing Gemini 2.5 Flash Image", August 26, 2025).
Big changes of angle are likely where things break down most. If your reference is a single front-facing image, the AI has no choice but to guess at the side and back it never saw. Judging whether the face is the same gets easier when you compare part by part instead of going on overall impression.
| What to compare | What to check |
|---|---|
| Eyes | Shape, color, size, and how lashes and highlights are drawn |
| Bangs and hairstyle | Where the part is, which way the locks fall, length |
| Asymmetric details | Whether hair clips, moles, scars, or accessories have flipped to the other side |
| Outfit | Patterns, number of buttons, where the colors go |
| Proportions | Head size relative to the body |
Asymmetric details are exactly what AI tends to mirror. Writing something like "hair clip on the left" into your character notes gives you a fixed reference point when comparing.
What are the common ways to keep the same face?
The three most common approaches are giving a reference image, making a character sheet (a front/side/back turnaround), and extra training with LoRA. A reference image is the easiest; the more effort you invest, the more cues the AI has to work with — but none of them can promise the same face every time. The fourth row of the table is separate from these three: using a 3D model in Figmee to check angles.
| Method | What you do | Best for | Watch out for |
|---|---|---|---|
| Give a reference image | Attach your favorite image and ask for "this character, just a different pose" | People using chat-style image generators | Likely to break down with big changes of angle |
| Make a character sheet | Prepare one image with the front, side, and back, and use it as the reference | People who'll use the same character for a long time | If the sheet itself is inconsistent, those errors carry over |
| Extra training (LoRA) | Gather a few images of the character and train an image model on them | People who can run an image model on their own computer | Takes a training setup and effort; check the terms of the model or service you use |
| Turn it into 3D to check angles | Make a 3D model from one image and rotate it for side and back cues | People who want a draft before drawing a character sheet | Anything not in the drawing is guessed by AI |
On reference images, Google's same announcement says you can now place the same character into different environments (source: the Google Developers Blog post above). On extra training, Hugging Face's documentation describes DreamBooth as training an image model on just a few images of a subject or style, and LoRA as a lightweight technique that inserts a small number of new weights into a model and trains only those (sources: Hugging Face Diffusers, "DreamBooth" and "LoRA", as of November 2026).
When you prompt with a reference image, separate what to keep (eye shape, hairstyle, outfit pattern, asymmetric details) from what to change (pose, hand position, expression). If one pass doesn't fix it, pointing out one difference at a time is likely to preserve the rest better than regenerating everything. For prompt examples, see our guide to ChatGPT Images 2.5 Sketch; for how to ask Gemini, see making figurine-style images with Nano Banana. If you'd like to draw a character sheet yourself, how to draw a character turnaround sheet walks you through it.
How do you register a character in Figmee and create new poses?
In Figmee, you register a work that has a figurine-style image (3D image) with "Register as character", then describe a pose in "Pose / scene" in up to 200 characters. Figmee uses the registered work's 3D image as a base and creates a 3D image in the new pose while trying to keep the character's appearance. Registering is free; each new pose uses 1–3 tokens depending on quality.
- Prepare the character image (about 5 minutes): Pick one illustration-style image with a single character, full body, facing roughly forward. Supported formats are PNG, JPEG, and WebP, up to 20 MB. If a photorealistic or photo-like image is detected as a photo, generation stops with a message that photos aren't supported, and the tokens you used are returned automatically.
- Create the figurine-style image (a few minutes): Upload, choose a quality ("Standard" 1, "Pro" 2, or "Max" 3 tokens), and press "Generate 3D Image." Check here that the facial features have survived.
- Register it as a character (about 1 minute): On the work page, select "Register as character," enter a "Character name," and select "Register." Your characters are listed under "Characters" in the "Gallery" menu.
- Generate a new pose (a few minutes): Open the character from the "Characters" list and select "Generate a new pose." In "Pose / scene," describe the pose or setting you want (for example: jumping and making a peace sign), choose a quality, and select "Generate." Before you do, you'll see how many tokens it uses and your current balance.
- Compare (about 2 minutes): The new image is saved as a new work and appears under "This character's works" on the character page. Put it next to the original and compare using the table above. If something looks off, change how you describe the pose and try again.
What you write in "Pose / scene" is the pose or setting; the look comes from the registered work's 3D image. To make comparing easier, start with poses where the face stays visible (keep facing front or three-quarters and change only the arms and legs), and leave poses that hide the face — turned away or looking down — for later. If a generation fails, the tokens are returned automatically.
On pricing: new accounts get 5 free tokens as a sign-up bonus (valid for 3 months, first sign-up per email address only). Generating a 3D model uses 5 tokens, and your first model per account is free.
What happens to the side and back when you turn it into 3D?
You can create a 3D model (5 tokens) from the registered character's work or from any new-pose work. You can rotate the finished model 360° in your browser, so you can check the profile and back the drawing never showed. Keep in mind that the model is made from a single image: the sides and back that aren't visible are guessed by AI, and the back may not match what you imagined.
That guessed part can still be useful material for deciding your character's design.
- Rotate and stop (about 2 minutes): In the viewer, stop at the full side view, a three-quarter back view, and the full back view, and look at hair length, where things are tied, and the back of the outfit.
- Write down the differences (about 3 minutes): Put into words where it differs from what you had in mind — "back hair reaches the shoulders," "ribbon on the right," and so on.
- Add them to your next prompt (about 3 minutes): When you ask an image generator for a back or side view, add what you wrote down. It also works as a draft for drawing a character sheet.
You can download the 3D model as GLB or 3MF from the work page, and export a square video of about 6 seconds in which the model turns twice (with the Figmee logo in the bottom right corner) via "Create 360° video." Downloading and exporting don't use tokens. To show your 3D model to others, see how to share a 3D model online; for the full route from one drawing to 3D, see the complete guide to turning an illustration into 3D.
How does copyright treat AI-made art?
Japan's Agency for Cultural Affairs, in its paper on how to think about AI and copyright (March 15, 2024; in Japanese), says whether something made with AI is a copyrighted work is decided case by case, and that if the instructions to the AI stop at the level of an idea, the result isn't recognized as a work. In other words, you don't necessarily hold copyright in a character image you made with AI. Rules differ by country, and what follows is general information, not legal advice.
The same document lists factors like these for deciding whether there's a copyrighted work (source: Agency for Cultural Affairs, "AIと著作権に関する考え方について" ("Perspectives on AI and Copyright," title translated by us), March 15, 2024, in Japanese; list of materials on the agency's "AI and copyright" page):
- Amount and content of the instructions: detailed instructions that concretely specify creative expression make it more likely that a creative contribution is recognized. Long instructions that only convey an idea don't count.
- Number of attempts: many attempts alone don't matter, but repeatedly checking the output and revising the instructions may lead to the result being recognized as a work.
- Choosing among outputs: simply picking one doesn't affect the assessment of creative contribution.
- Human edits and additions: parts where a person added edits or additions that amount to creative expression are normally recognized as copyrighted work.
The document also says that feeding an existing work into an AI in order to generate something similar to it isn't covered by the exception that allows uses such as AI training (Article 30-4 of Japan's Copyright Act). Don't use other people's illustrations or existing characters as reference images to produce lookalikes.
Check the terms of the image generator you used, too. Whether you can upload its images to another service or publish them depends on each service's terms. For an example with ChatGPT's terms, see our guide to ChatGPT Images 2.5 Sketch; for rights and safety when turning images into 3D with AI, see safety and copyright when using AI to make figurine-style 3D models.
Which images can you upload to Figmee?
Figmee's terms of use don't prohibit uploading an image just because it was made with AI. You can upload images you hold the rights to, or images you have permission to use from the rights holder. Photos and images that look like photos can't be turned into 3D images.
| Image | How Figmee handles it | Why |
|---|---|---|
| A drawing you made yourself | You can use it | The person who drew it holds the copyright |
| An AI illustration of your original character, made from your own prompts (illustration style) | Use it after confirming the AI service's terms allow it | Figmee's terms don't prohibit it, but how outputs may be used depends on the AI service's terms |
| An AI image made from someone else's illustration or an existing character | Don't use it | It may infringe someone else's copyright |
| A photorealistic or photo-like AI image | Can't be turned into a 3D image | Generation stops when it's detected as a photo (tokens are returned automatically) |
| An AI image made to look like a real person | Don't use it | It may infringe portrait rights and similar rights |
Images you upload are used only to create 3D images and 3D models, and are not used to train AI. They aren't shown to third parties unless you publish the work yourself. Copyright in generated images follows applicable law. For the overall picture of how the source image affects your rights, see 3D model copyright basics.
FAQ
Can I turn an AI-made illustration into 3D with Figmee?
Yes, if it's illustration-style. Photorealistic or photo-like images are stopped when they're detected as photos, and the tokens you used are returned automatically. Don't use images made from other people's work or existing characters, and check the terms of the AI service you used.
Will poses made with character registration always have the same face?
Not always. The feature creates new poses from the registered work's 3D image while trying to keep the character's appearance. If something looks off, try poses where the face stays visible first, or change how you describe the pose and generate again. Each new attempt uses tokens.
Can I make a 3D model without a back-view character sheet?
Yes. Figmee makes a 3D model from a single image. Because the back isn't in the drawing, AI guesses it, so it may differ from what you imagined. We recommend rotating it to check, then adding the differences to your character notes or your next prompt.
Summary
To get a consistent AI character in different poses, pin down the face with a reference image, a character sheet, or extra training, then compare part by part — eyes, hair, asymmetric details — and fix differences one at a time. In Figmee, you can create a figurine-style image from an illustration-style character image, register it as a character, and create new poses while Figmee tries to keep the look; turning it into a 3D model gives you cues for the side and back. Because rights in AI-made art are decided case by case, stick to two rules: don't imitate other people's work, and check the terms of the AI service you used.
Related guides
- ChatGPT Images 2.5 Sketch: From Rough Drawing to 3D Model
- How to Make Figurine-Style Images with Nano Banana, and Turn Them Into 3D
- How to Draw a Character Turnaround Sheet
- How to Turn an Illustration into 3D (2026 Guide): Bring Your Lineart to Life with AI
- Safety and Copyright When Using AI to Make Figurine-Style 3D Models from Drawings
- 3D Model Copyright Basics
