Can an AI character keep the same face in every picture?

Yes, if the look is stored on the character rather than re-described for each picture. Talaforge locks a visual identity to the character when it is created and carries it into every later scene, so the face survives changes of era, lighting and rendering style.

Do I have to upload a reference photo?

No. The character's own portrait is made when the character is made, and that portrait is the reference from then on. There is nothing to upload and no settings screen to open. And if you have a particular face in mind, you are the one who describes it — a character you make is yours to write and re-edit at any time.

Do I need an app, or does this work inside Telegram?

It works entirely inside Telegram, in a normal chat with @talaforge_bot. There is no separate app, no account to create and no browser tab to keep open.

Can the character send voice messages as well as pictures?

Yes. Voice and images are both native to the conversation — the character replies with them when the moment calls for it, and you can turn either off in the chat settings if you would rather it did not.

Can I turn a character into its own Telegram bot?

Yes. A character can be attached to a bot of your own with its own @handle, so other people talk to it directly. You keep the token; we run the character behind it.

One character, every scene: how a face stays the same when nothing else does

The picture at the top of this page and the one on the character's card were made hours apart, in different styles, by different requests, in the middle of an ordinary conversation. Nobody uploaded a reference. Nobody opened a settings screen. That is the part worth explaining, because it is the part that usually does not work.

The problem with AI pictures of a person

Ask most image tools for the same character twice and you get two people who happen to match the same adjectives. Brown wavy hair, blue eyes, grey sweater — and a different face. The description survives; the person does not.

For a single illustration that is fine. For a conversation it is fatal, because the whole point of a companion is that they are somebody in particular. A face that re-rolls every message is a stranger wearing the same outfit, and you feel it immediately even if you cannot name what went wrong.

What is actually locked

When a character is created here, their appearance is fixed to the character rather than re-derived from the words each time. Every later picture starts from that anchor and changes only what the scene changes — the room, the clothes, the light, the era, the mood, who else is in the frame.

That is why the frames below can disagree about almost everything and still agree about him. Look at the cafe one: the grey ribbed sweater and the cosmic teardrop from the original card came along, unasked. Then look at the drawing room — different century, different palette, hard side-light, and it is still the same face doing the listening.

The card — Where he starts: the character's own portrait, made when the character was made. The grey ribbed sweater and the small cosmic teardrop are part of him, not part of a prompt. — **The card**Where he starts: the character's own portrait, made when the character was made. The grey ribbed sweater and the small cosmic teardrop are part of him, not part of a prompt.

The cafe — Hours into a conversation. Different framing, another person in the shot, coffee on the table — and the sweater and the teardrop came with him, unasked. — **The cafe**Hours into a conversation. Different framing, another person in the shot, coffee on the table — and the sweater and the teardrop came with him, unasked.

The drawing room — A different century entirely: three-piece suit, hard side-light, a scene with weight in it. Everything changed except who is sitting in the chair. — **The drawing room**A different century entirely: three-piece suit, hard side-light, a scene with weight in it. Everything changed except who is sitting in the chair.

Daylight — Rendered photographically rather than drawn. The style of the picture is not the identity of the person — this is still unmistakably him. — **Daylight**Rendered photographically rather than drawn. The style of the picture is not the identity of the person — this is still unmistakably him.

The apartment — Full-length, flat anime rendering, an ordinary room. A third style, a third distance, the same face. — **The apartment**Full-length, flat anime rendering, an ordinary room. A third style, a third distance, the same face.

Close — Near enough to check: the eyes, the mouth, the way the hair falls. This is the frame to compare against the card at the top. — **Close**Near enough to check: the eyes, the mouth, the way the hair falls. This is the frame to compare against the card at the top.

Swipe, or tap a frame. Every one of these came out of the same conversation with the same character.

Nothing was switched on

This is the part people expect to be a feature you enable. It is not. Pictures can arrive on their own when the scene calls for one — the character decides a moment is worth showing and shows it — and the voice button sits under replies from the first message if you would rather hear the line than read it.

You can steer it, of course: ask for a picture directly and you get one, set images to on-request-only if you would rather they never surprise you, or turn them off entirely. But the default is that the conversation has pictures in it, the way a conversation with somebody who is somewhere has pictures in it.

It goes the other way too

Send him a photo and he looks at it. The view from your window, the shirt you were unsure about, a drawing you made — he answers in character about the thing that is actually in the frame, not about a caption you had to write for him.

Which closes the loop: he shows you where he is, you show him where you are, and the conversation stops being a text box.

The honest details

Pictures cost energy, topped up with Telegram Stars inside the chat, and you can see the balance at any time. A render that fails or is blocked is not charged — you never pay for a picture you did not get. There is a daily ceiling, so a runaway afternoon cannot empty the balance.

You choose which image engine draws, and the result lands in the conversation rather than in a gallery somewhere else. Mature content sits behind an age check inside the bot; this page and the rest of this site stay as you see them.

Try it with your own

One sentence is enough to start a character — a role, a mood, a name — and the rest is drafted for you and stays editable forever. The first picture usually arrives before you have finished deciding who they are.

And if the character turns out to be somebody other people should meet, they can become a separate Telegram bot with their own @name, in a couple of minutes and without code.