From one photo to a character you can reuse anywhere

If you've ever tried to build a series, an AI influencer, a comic, or anything that needs the same character in more than one image, you've hit the wall: the face changes every single time you generate. One great result, then the next prompt hands you a stranger. Getting one character to stay the same across dozens of images, and into video, is the difference between a one-off and an actual body of work.

Until recently, the fix was heavy. You collected 20 or more photos of the character from different angles, trained a custom model with a Low-Rank Adaptation (LoRA), and babysat the whole pipeline. Now a single clear photo does it. You upload one image, and that face becomes a reusable character you can drop into any prompt, with no training and no dataset.

This is the hands-on walkthrough: how to turn one photo into a locked character, whether the face is your own or one you invent, how to prompt it into new scenes, and how to push the consistency as far as it will go. If you want the deeper why behind face drift and how identity locking actually works, that's in our guide to AI character consistency.


What makes a good reference photo

Your single photo sets the ceiling for everything that follows. The model reads the character's identity from that one image, so whatever is wrong with it or missing from it shows up in every generation afterward. It's worth 60 seconds to pick the right one.

A strong reference photo is:

  • Front-on or 3/4 angle. The model needs a clear read on the face to lock its structure. Steep angles and pure profiles give it less to work with, and the identity comes out shaky.

  • Evenly lit. Harsh shadows hide half the face, and the model fills those gaps with guesses. Flat, even light shows the real features.

  • A neutral or natural expression. An extreme expression gets baked into the character and fights you later. A relaxed face is easier to re-pose into anything.

  • Sharp and high resolution. Soft or low-res in means soft, mushy faces out. Detail carries.

  • On a clean background. A busy background can bleed into the lock. A plain one keeps the focus on the face.

What to avoid: heavy filters (they alter the real features before the model even sees them), sunglasses or hats or a hand across the face (the model can't lock what it can't see), group shots (it won't know which face you mean), and tiny or blurry crops. A decent phone selfie in good light beats a dramatic studio shot with deep shadows. Clarity wins over production value here.

One rule before you upload: use your own face, or a face you have explicit permission to use. More on that below.


How to create your character from one photo (step by step)

The whole thing runs on Mango 3, Mage's flagship image model, unlimited on Pro ($30/month) and up. No training, no dataset. Here's the loop end to end.

Step 1: Open Characters and click Create New.

Step 2: Upload your one clear portrait, name it, and create it. The name you give becomes the tag you'll type in prompts, so keep it short and memorable. The character saves to your library, so you can come back to edit or publish it later.

Step 3: Call the character into any prompt with @charactername. From the Prompt Bar, write your scene and drop the tag where the character goes. Prompt @maya standing on a rooftop at golden hour and you get Maya, in a place that never existed, wearing the same face as your photo.

Step 4: Change everything around her. Because the identity is locked to the character and not the prompt, you can swap the pose, outfit, lighting, and setting freely and the face holds. Describe the scene as specifically as you'd describe a photo you were trying to shoot, the more direction you give, the closer the result:

@maya in a red trench coat, walking through rain, cinematic 

@maya laughing at a cafe table, soft morning light 

@maya in a sci-fi cockpit, neon reflections on her face

Step 5: Refine and curate. Each generation varies a little, so if one comes back slightly off, regenerate, tighten the prompt, and keep the ones that land. You're curating a set, not expecting one-shot perfection.

Lock the face once, reuse it forever. That's the loop.

One thing to keep straight: creating a character takes one photo, full stop. That's a different feature from Mango's multi-image reference input, where you can feed up to 10 images to blend into a single composition. Don't confuse the two.


How consistent will it actually be?

Honest answer: very consistent, but not photocopied. Across ordinary scenes, outfits, and lighting, the face reads clearly as the same person. Push it into hard territory, tiny faces in wide shots, extreme angles, wild lighting, and you'll start to see small drift. That's normal, and it's controllable. Here's how to keep it tight:

  • Start from a better reference. Most "close but not quite" results trace straight back to a weak source photo. The checklist above is the highest-leverage fix.

  • Keep the face a decent size in frame. When a face fills less than about a fifth of the frame, there isn't enough room for detail and identity slips. For a wide establishing shot, generate a strong portrait first and build the wide scene around it.

  • If the face reads like a "cousin," close but subtly wrong, your prompt is usually too loose or your reference too soft. Tighten both.

If you want the underlying reason faces drift at all, the character-consistency guide breaks down what's happening under the hood.


Two ways to use it: an AI version of you, or an original character

The workflow is identical. What changes is where the photo comes from and what you're optimizing for.


Your own face. Upload a selfie and you've got an AI version of yourself: the base for an AI influencer, a set of styled portraits, or content you'd rather not shoot on camera every day. Here the whole game is fidelity. It has to actually look like you, not a near-miss. Start from your sharpest, most natural photo and it will.

This is where consent matters. Mage bans non-consensual deepfakes and enforces it hard, with multiple layers of detection and permanent bans across accounts, devices, and IP addresses, and it removed photorealistic celebrity models under the Take It Down Act. So the line is simple: your own face, or one you have explicit permission to use. Anything else is off the table.

An original character. Don't want to use a real face? Invent one. Use Character Creator to build a face from preset attributes, or Character Builder to start from a muse and refine the features with presets or your own prompts. Generate until you land on a face you like, then save it as a character. Here the game is reproducibility. Creators lose great faces constantly because they never saved a clean reference, then can't get that exact person back. Lock it as a character and it's yours to reuse forever, with no likeness questions attached.

Just want to drop a face onto a single existing image? That's a face swap, a quick one-off, not a reusable character.


Prompt examples for a consistent character

Once the character is locked, consistency comes down to one habit: keep the tag identical every time, and describe everything else. The tag carries the face. The rest of the prompt is just the scene around it.

  • New setting: @maya reading on a train, afternoon light

  • Saved outfit: @maya wearing @leatherjacket at a concert (pair the character with a saved outfit reference)

  • Saved pose: @maya doing @yogapose on a beach at sunrise

  • Two characters, no blending: @maya and @julian arguing across a kitchen table

  • New medium, same identity: @maya as a Renaissance oil painting, dramatic side lighting

The mental model is save once, reuse infinitely. The character is an asset you build a single time and pull off the shelf for every project after.


Take your character further

The same locked character carries well past still images. Bring it into motion with Character Reference (Image)-to-Video on Cherry (Pro Plus, $60/month) and the face holds across the clip. Give it a voice with Audio Generation. Put several locked characters in one scene with Multi-Characters.

Video is its own challenge, since faces wobble in motion in ways they don't in stills. If you're headed that way, the character-consistency guide goes deeper on holding a face steady across clips.


Tips for the most consistent results

  • Lock the character first. Get the face right before you build scenes around it, not the other way around.

  • Change one variable at a time. Swap the outfit, then the pose, then the setting, so you can always tell what threw a result.

  • Keep a character sheet. A short note (name, key features, signature look, reference file) makes a recurring character easy to manage across a long project.

  • Fix plastic skin at the end. If a face comes back waxy or over-smoothed, run it through the Skin Enhancer app to bring back real texture.

  • You can sell what you make. Paid plans include a commercial license, so the characters you build are yours to use commercially.


Frequently Asked Questions


Is one photo really enough, or do I need more? 

One clear photo is enough to create a character on Mage. A sharper, front-facing reference gives a stronger lock, so the quality of that single photo matters far more than the quantity.


Do I still have to train a LoRA? 

No. Training a custom model still gives the highest fidelity for very high-volume work, but for almost everyone, locking a character from one photo gets you there with no dataset, no GPU, and no training time.


Why does my AI character look like a different person each time? 

Because text alone can't pin down a specific face, and without a locked identity every generation starts fresh. Creating a character fixes that, since the tag carries the same face into every prompt. Our character-consistency guide explains the full reason.


How do I make an AI version of myself that actually looks like me? 

Start from your best photo: front-facing, well lit, sharp, no filters. Create a character from it, then generate with the tag. If results look off, the reference photo is almost always the thing to improve first.


Can I create an original character instead of using a real face? 

Yes. Build one with Character Creator or Character Builder, save it as a character, and you have a fully original, reusable face with no likeness concerns.


Do I need a powerful computer or any coding? 

No. It all runs in the browser, with no graphics processing unit (GPU) and no setup required.


Is it okay to use my own face? What about someone else's? 

Your own face is fine. Someone else's needs their explicit permission. Mage bans non-consensual deepfakes and enforces it across accounts, devices, and IP addresses.


Can I put my character in a video? 

Yes. The same character carries into video through Character Reference (Image)-to-Video on Cherry and the other character-compatible video models.