What prompt makes a good AI character?
Paste this into Claude, ChatGPT, Gemini or Whizi and fill the brackets. It asks for the traits a story can use and one line you can hand to an image model.

Create a character for [a fantasy novel, a one shot tabletop game or a brand mascot]. Tone: [grim, warm, comic]. Role: [villain, mentor, shopkeeper]. Give me a sheet with: name, age, look in 40 words, one speech habit, one fear, one goal, one secret, one thing they would never do, and a two sentence backstory. Then write one line of what they would say if a stranger asked for help. Keep every field under 25 words.
The last two items matter most. A speech habit and a fear make a character behave, and a character who only has adjectives reads like a stock photo. The "never do" line stops the model from bending them to fit a plot later.
How do you turn the sheet into a character image?
Give the image model the look and the style, and leave out the personality, because a picture can't show a secret. Take the "look in 40 words" field from your sheet and wrap it in this.
Portrait of [look line from the sheet]. Style: [oil painting, flat vector, anime, 3D render]. Framing: waist up, facing the camera, plain [color] background, soft light from the left. No text, no logos.
Plain framing is on purpose. A neutral background and a straight pose give you a clean reference. You can move the character into a scene later with the editing step in the how to steps. On Whizi the image models are in the composer, and an uploaded picture can be attached to a message for the model to look at.
What do three real characters look like in practice?
Each example gives the input and what a good answer must contain. They are written for this page, not the output of a tested model, so run them yourself.
- A tabletop villain. Prompt: "Make a villain for a one shot game for four level 3 players. A bridge toll keeper who is secretly guarding a flooded vault. Use the sheet format." A good answer has a fear that players can use in a scene, such as drowning, and a speech habit that can be played aloud in a minute, such as answering a question with a price.
- A novel protagonist. Prompt: "Make a protagonist for a quiet literary novel set in 1970s Lisbon: a night shift pharmacist, 41, newly widowed. Give the sheet, then three contradictions that would create scenes." A good answer has contradictions that pull against each other, such as careful with strangers but reckless with her own savings, and nothing that reads like a résumé.
- A brand mascot. Prompt: "Make a mascot for a children's dental clinic. Give the sheet and then a one sentence visual description with a simple shape, two colors and no more than three features, so it works as a vector logo." A good answer is simple enough to draw in one stroke. A character with a scarf, glasses, a hat and a belt will not survive as a small icon.
Is there a free AI character generator?
Yes, several, and they split by what they make. Image tools like Firefly, Canva and Kapwing give a picture, and a chat model gives the sheet. The table lists what each tool says on its own page as of November 2026, and where a page was silent the cell says so.
| Tool | Free use | Sign up | Paid price | Best at |
|---|---|---|---|---|
| Perchance | Title says free and unlimited; the page shows no limit text | Title says no sign up | Not stated | Quick portraits from a description, with a style and a shape picker |
| tchepai site | Not stated | Required | Not stated | 3D model from a reference photo, 1 to 3 minutes |
| Adobe Firefly | Limited free daily generations | Free Adobe account | Standard US$9.99 a month for 2,000 credits (up to 400 images), Pro regularly US$19.99 for 4,000 | Concept art in set styles |
| Canva | Described as free, cap not stated | Sign in or sign up if prompted | Not stated on this page | Characters that go straight into a design |
| Kapwing | Page says free to start with full commercial rights | Sign in links on the page | Not stated | Characters reused in video |
| ChatGPT | Has a Free plan; limits not stated here | Account | Plus $20 a month (OpenAI help center) | The written half |
| Claude, Gemini | Write the sheet; plan prices not read for this page | Account | Not stated | The written half |
| Whizi | 2 pictures a month and 10 messages a day, no card | Account needed | Pro $29.99 a month | Sheet and portrait in one chat |
Whizi loses on three counts. Perchance says no sign up and Whizi needs an account. Firefly and Canva come with editors for fixing a picture afterward. And the tchepai site is the only one here that says it makes a 3D character for game engines.
The tchepai page says uploads of JPG or PNG up to 10MB turn a reference picture into a 3D character, and it lists FBX and OBJ export. Adobe's Firefly page says "Free users receive free daily generations to create custom characters from text prompts across a range of styles and themes." Canva's page says to "Sign in to your Canva account or sign up for free, if prompted." Neither gives a number, so test with your own first prompt before you plan a project around it.
The Perchance page is a single form: a description box, an Art Style menu that opens on Painted Anime, a Shape menu that opens on Portrait, and a generate button. Its page says it uses a text to image plugin. That makes it quick for one portrait and weak for anything that needs a sheet, a backstory or the same face twice.
This page is about making a character. If you meant chatting with characters inside the Character.AI app, that is a different product, covered in Character.AI subscription plans.
Which prompts work for characters, first try?
These 14 are grouped by job. Fill the brackets and run them in any chat model. The first four build the written character.
- Full sheet: use the prompt at the top of this page.
- Contradictions: "Here is my character: [paste sheet]. Give three contradictions in their behavior that would each produce a scene, and one line of dialogue for each."
- Backstory with one wound: "Write a 120 word backstory for [name], [age], [job]. Include one event that explains their fear of [fear]. Don't make it a tragedy unless I ask."
- Voice sample: "Write five lines of dialogue from [name] to a shopkeeper who won't give them credit. Keep their speech habit: [habit]. Don't explain it."
The next four are for games and groups.
- Tabletop NPC: "Make an NPC for a [setting] game: role [innkeeper], one thing they want, one thing they'll lie about, one secret the party can find by asking the right question."
- Party of four: "Make four player characters that need each other: each has one skill, one flaw and one reason to trust the next. Give a table."
- Antagonist with a fair point: "Make a villain for [story] whose goal is reasonable and whose method isn't. Write the one argument they'd win in a debate."
- Relationship map: "Here are my three characters: [paste]. Write one sentence for each pair on what they owe or hide from each other."
Then the image prompts, which pair with the sheet.
- Portrait: use the portrait prompt above with your look line.
- Three styles: "Turn this look into three image prompts in three styles: [style A], [style B] and [style C]. Same face details in each: [look line]."
- Expression sheet: attach your reference and write "Keep this face. Show it in a row as happy, afraid and angry, same lighting, plain background."
- Outfit change: attach your reference and write "Keep the face and hair. Change the clothes to [outfit]. Same framing."
Last, two prompts for keeping everything straight.
- Consistency card: "Compress this sheet into a card of 100 words that I can paste at the start of every chat: [paste]."
- Continuity check: "Here is the card and the scene I wrote: [paste both]. List every line where the character acts against the card."
How do you keep the face the same in every image?
Edit from a saved picture and never re describe the face from scratch. An image model that gets only words draws a new person each time, because the words allow thousands of faces.
Pick the one portrait you like and save it. For every new scene, attach it and say what to change, as in prompts 11 and 12. On Whizi, Nano Banana is the editing model, and one edit can carry up to 4 source images, so you can attach a face and an outfit together. Flux Schnell and Stable Diffusion 3.5 have no instruction edit mode, so a request with attachments runs on a model that can edit.
For the written side, keep the card from prompt 13 in a project and the model rereads it every time. How to keep characters consistent across chapters shows the same habit for a long story.
What is an AI character card, and how do you write one?
A character card is the short sheet you paste at the start of a chat so a model plays or writes the same person every time. People searching for AI character cards usually want one of two things: a card for a story, or a card for a roleplay chat. Both are the same sheet, with one extra block for roleplay.
For a roleplay chat, add this to the sheet from the first prompt: "Write the opening message this character would send to [the user's role, such as a new recruit], in their voice, in under 60 words. Then write two example replies to a rude question. Keep them in character and never mention being an AI."
Test the card by asking the character something it wasn't told, such as what they ate yesterday. A good card gives an answer that fits the fear and the speech habit. A weak card gives a generic one, and that tells you which field to rewrite. Best AI chatbot for roleplay compares the apps that run these cards.
How do you describe a character to an image model?
Use concrete nouns for the face, the clothes and the light, and skip words a picture can't show. "Brave" draws nothing. "A scar through the left eyebrow and a worn canvas coat" draws something.
Start with the shape of the face, then hair, then one object that tells the story, such as a ferry token on a cord. Add the style last, in a few words. Put the framing in every prompt, because the same face looks different from the side. If the result drifts, shorten the prompt instead of adding to it. Each extra feature gives the model another thing to get wrong.
Can you use an AI character commercially, and how long does it take?
Terms differ by tool, so read them before you print a mascot on a box. Kapwing's page says free to start with full commercial rights. Adobe and Canva do not state commercial terms on the character pages read for this article, and the tchepai page has a commercial use question in its FAQ that the page text does not answer clearly.
Speed also differs. The tchepai page gives 1 to 3 minutes for a 3D character built from a reference picture. The 2D tools return pictures in the time it takes to type, so run a few styles before you commit to one.
Use the prompts above for the part no picture tool offers, the written character, and treat each picture tool as the drawing half.
What mistakes ruin an AI character?
Four mistakes cause most of the weak results, and each has a quick fix.
- Describing only looks. The fix: ask for a fear, a goal and a speech habit before you ask for a face.
- Piling on features. The fix: three visual features at most, or the image model drops half of them.
- Re describing the face in every prompt. The fix: attach the saved picture and edit it.
- Leaving the "never do" line out. The fix: add it to the card, so a plot twist can't quietly turn the character into someone else.
Where does Whizi fit as an AI character generator?
Whizi is one app instead of a pile of single purpose tools: the sheet, the portrait and the edits happen in one chat. The creative mode in the composer switches the picker to image models, and the catalog offers Whizi Image for making any image, Nano Banana for editing your photos and Flux for sharp photoreal art.
The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, plus 2 pictures a month, with no card. Pro carries Claude, GPT and Gemini in one app, includes 100 pictures a month, and its first week is $1.99. Side by side Compare, on Pro, runs one sheet prompt through two models so you can keep the better voice. Pin your character card in a project, and see Whizi pricing for the plans.
A dedicated tool is quicker if you only want one random character with no account.
What can an AI character generator not do?
It can't promise a distinct character, a consistent face or a character you own. Models lean toward the common choice, so a sheet with no contradictions comes back as the average fantasy innkeeper.
Ownership is the open question. The US Copyright Office published Part 2 of its report on AI and copyright on 29 January 2025, and it addresses whether AI outputs can be copyrighted. Read the Office's AI page before you sell a character, and check each tool's terms for commercial use. Don't upload a real person's photo to make a character without their say so.
- Ask for a fear, a goal and a speech habit before the look
- Keep the look to three visual features
- Save one portrait and edit from it for every new scene
- Keep a 100 word character card and paste it into every chat
- Check each tool's terms before you use a character commercially
Frequently asked questions
Is there an AI character generator with no sign up?
Perchance's page title says free, no sign up and unlimited, as of November 2026, but the page shows no limit text. Adobe Firefly needs a free Adobe account, and the tchepai site requires registration. Test the first character before you rely on any of them.
Can I make an AI character from a photo?
Yes, with limits. The tchepai site says it turns a JPG or PNG up to 10MB into a 3D character. In chat, you can attach a photo to a message for the model to look at. Only use photos of people who agreed.
What is the difference between an AI character generator and Character.AI?
A generator makes a character sheet or a picture you keep, while Character.AI is an app where you chat with characters. This page covers making one, and Character.AI subscription plans covers the app.
Can I use an AI generated character commercially?
It depends on the tool and on copyright law. Kapwing's page says free to start with full commercial rights, and the US Copyright Office addresses the copyrightability of generative AI outputs in Part 2 of its report. Read each tool's terms before you sell.