PROMPT GUIDE
How to Keep the Same Character in AI Art (Face, Hair and Outfit)
"I want another picture of the same girl, but her face and clothes keep changing" is one of the most common problems in AI art. Character consistency comes in three levels: fixing her with words, fixing her with a reference image, and fixing her with training (a LoRA). This guide explains how each idea works in Stable Diffusion, NovelAI, Midjourney and ChatGPT, with steps and common mistakes.
Last updated: approx. 5 min read
The short answer
The easiest way to keep the same AI character is to describe her hairstyle, hair color, eye color and outfit with exactly the same words every time. If the face still drifts, give the original picture as a reference in services that support it, or use a LoRA trained on that character in Stable Diffusion.
How do you keep the same character in AI art?
Write one "character sheet" sentence and paste it in unchanged every time. If even one word changes, the AI tends to draw a different person.
The AI creates every image from scratch, so it does not remember the face from the last one. To make a picture look like the same character, you need to give it the details that define her look (hair length and color, bangs, eye color, accessories and her usual outfit) in the same order and with the same words every time. Once that is fixed, you can change the pose or place and still get pictures that clearly show the same girl.




How do you write a character sheet sentence?
Make a short sentence (or tag list) that lists her features in order of "least willing to change". Add the scene and pose words after it.
Steps to make a character sheet sentence
- Keep your favorite picture as the "reference image".
- From it, write down her hair length, hairstyle, bangs, hair color, eye color, skin tone and build.
- Add the things she always wears (ribbon, glasses, hairpin) and her usual outfit.
- Order the words as face and hair, then accessories, then clothes, and save them in a note.
- For each new picture, paste the saved sentence at the start, then write the scene, pose and expression.
- Put the new picture next to the reference image and add or strengthen only the words for parts that look different.
1girl, solo, long black hair, straight hair, blunt bangs, red eyes, red hair ribbon, white blouse, navy cardigan, plaid skirt, [scene], [pose], [expression]A young woman with long straight black hair, blunt bangs, red eyes and a red ribbon in her hair, wearing a white blouse, a navy cardigan and a plaid skirt. [Write the scene and pose here]Rinka's tips
- Copy and paste the character sentence every time. Retyping it by hand slowly changes the wording
- Too many features get mixed up. Narrowing them to 5–8 standout details is usually more stable
- When you want a new outfit, swap only the clothing part and keep the face and hair words
How do you fix a character in each service?
In every service the foundation is a fixed character sentence. On top of that, add whatever each service offers, such as reference images or add-on models.
Stable Diffusion
Besides the tag-based character sentence, write down the seed of a picture you like; that makes it easy to create variations where only the expression or outfit color changes. If you are drawing the same character dozens of times, a LoRA trained on that character is the most stable option. There are also extensions that borrow the feel of a face from a reference image, but how well they work depends on the model and settings.
NovelAI
NovelAI also accepts tags, so the basic approach is the same as Stable Diffusion: keep the character sentence fixed. Depending on the version, there may be a separate field for each character or a feature that uses a reference image. Names and usage change between versions, so check the official documentation.
Midjourney
Write the same character sentence every time and combine it with the feature that takes a picture you like as a reference. The name of the character reference parameter can differ between versions, so check the official docs for the version you use.
ChatGPT and Gemini
Within the same conversation, a request like "the same girl, now reading a book in the library" uses the previous image as a clue. In a new conversation, upload the reference image again and write "the same character as in this image" to get closer.
What if the face changes even though it is the same character?
Most causes are one of three: the character sentence differs slightly, the scene words are too strong, or the face is too small in the frame. Use the table to find the cause from the symptom.
How do you make variations with only the expression or outfit changed?
Keep the character sentence and the seed (or reference image) as they are, and swap only the words for the part you want to change, one at a time. Changing two or more things at once makes the face more likely to change too.
- Expression variations: change only the expression word (smile → surprised, etc.)
- Outfit color variations: change only the clothing color word (navy cardigan → beige cardigan)
- Different places: change only the background words and keep the composition
- To fix just one part, use a feature that redraws part of the image (such as inpaint)
To make standing art, turnaround sheets and expression sheets: Full-Body Standing Art, Turnaround and Character Sheet Prompts (Plus Expression Variants) →
How to write a LoRA in Stable Diffusion and set its strength: How to Use LoRA (Syntax, Strength and Trigger Words) →
How to redraw only the face or outfit with img2img and inpaint: How to Use img2img and Inpaint (Fixing Only Part of an Image) →
How to ask ChatGPT for images and request fixes: How to Prompt ChatGPT for Images (Asking and Editing) →
FAQ
Frequently Asked Questions
What is the easiest way to keep drawing the same character with AI?
Make a "character sentence" listing hairstyle, hair color, eye color, accessories and outfit, and paste it at the start every time. Changing even a few words tends to create a different person, so paste it instead of retyping.
Does using the same seed give the same character?
With exactly the same prompt you get almost the same picture. But changing the words also changes the image a lot, so the seed is best for small variations such as expression or color.
Do I need to train a LoRA to keep a character consistent?
A fixed character sentence and a reference image already get you quite close. A LoRA helps when you draw the same character dozens of times or want to match fine facial details.
I started a new ChatGPT conversation and got a different character.
A new conversation does not remember the previous image. Upload the reference image again, ask for "the same character as in this image", and include the character sentence as well.
Can I make the same face from a photo of a real person?
Using someone else's photo can raise portrait-rights and privacy issues. Character consistency is safest with an original character you created yourself.
RINKA WORLD
Play in Rinka World
MORE