AI Avatar Video From Your Photos: Omni, Photo and Twin

Avatars in Kolbo.AI turn 1 to 8 photos of a person, plus a voice, into a reusable avatar that speaks any script you type. It is built for creators, founders, teachers and marketers who need talking videos without a shoot. You pick one of three take modes: Omni (most lifelike, up to 60 seconds), Photo (up to 3 minutes) or Twin (your real video, new words).

This guide walks through building the avatar, giving it a voice, choosing the right mode, picking a lipsync model and directing the performance. At the end you will also see how to do all of it from an AI assistant through the Kolbo MCP server.

Open Avatars

Launch offer (dated 2026-11-12): until November 30, 2026, Omni is 50% off and Photo and Twin are 20% off. A Photo video with a lipsync model you pick yourself is billed at that model's own price and is not part of the offer.

What an Avatar Is in Kolbo.AI

An avatar is a saved person, not a one-off video. It holds the photos, a voice, optional "looks" (outfits), optional places and, if you add one, a short talking video of the person. Once it exists you can make as many videos with it as you like, in different places, camera styles and languages, without uploading anything again.

Avatars live in Video Tools, next to Lipsync. Videos are saved inside your current project, so open a project first.

If all you need is one picture that speaks exactly as it is, the AI talking photo guide covers that simpler path. Avatars is the tool for a recurring presenter.

The Three Take Modes Compared

Every avatar video is made in one of three modes. Here is what each one does and what it needs.

Every mode supports 9:16, 16:9 and 1:1, can speak a script or your own recording, and can return the video with the background removed (up to 1080p and 60 seconds, for an extra per-second charge).

When to use Omni

Choose Omni when the video has to look like a real recording. The person moves, glances and gestures in the place you picked, and you choose how the camera films them. It is the right mode for ads, UGC-style hooks, short social clips and anything where realism matters more than length. Long scripts are filmed as several takes that continue from the last kept frame, so joins stay smooth.

When to use Photo

Choose Photo for longer talks: explainers, lessons, product walkthroughs and updates up to 3 minutes. Kolbo.AI first makes a still of your avatar in the chosen place and shows it to you. You approve that first frame, and only then are credits used for the video. You can also skip the place and make your own photo talk exactly as it is.

When to use Twin

Choose Twin when you already have footage of yourself and want it to say something new. Twin keeps your real room, clothes and gestures and changes only the speech. To unlock it, add about a minute of the person talking to the camera, 720p or better.

How to Make an AI Avatar Video, Step by Step

1. Open Avatars. Go to Video Tools and choose Avatars, or use the button above. Open a project first if you have not already.
2. Create the avatar. Click New avatar and add 1 to 8 clear, recent photos of one face. More angles give a closer likeness. Avoid sunglasses, a hand over the face, group shots or blurry images.
3. Give it a voice. Record or upload 8 to 60 seconds of speech in a quiet room, pick a voice from the library, or choose one of your own saved voices. A recording is cloned with Cartesia or ElevenLabs, your choice.
4. Add a talking video (optional). About a minute of the person talking to the lens unlocks Twin. You can skip this step.
5. Name it and confirm consent. Confirm the avatar is you, or that you have the person's permission, then create it.
6. Pick the mode, place and camera style. Choose Omni, Photo or Twin. For Omni and Photo, pick a place from more than 60 ready rooms and locations, describe a scene in words, or use your own image. In Omni, pick the camera style.
7. Write the script. Type what the avatar should say, in any language, or upload your own recording (up to 3 minutes). Stuck on wording? Ask Kobi, the Kolbo.AI assistant, to draft it.
8. Listen first (optional). Preview how the script will sound in the avatar's voice before you make the video.
9. Add directing notes and generate. Write a short note such as "excited, laugh on the last line", check the price shown before you start, and press Generate.

Building an Avatar From a Visual DNA

If you already use Visual DNA for a character, you do not need to start again. Choose From a Visual DNA when you create the avatar. Kolbo.AI copies the character's images, reference sheet, voice and description into the new avatar, so the same person appears in your image work, your video scenes and your avatar videos.

A location Visual DNA can be used as a place too, which keeps a recurring set, office or studio consistent from one video to the next.

Voice: Clone, Library or Your Own Recording

The voice decides whether the avatar feels like the person. You have three routes:

• Clone a voice. Read the short prompt on screen, or upload a clean sample of 8 to 60 seconds. Pick Cartesia or ElevenLabs as the cloning engine. ElevenLabs adds emotion to your scripts automatically. The voice cloning guide covers how to record a good sample.
• Pick a library voice. Listen, filter by language and choose one. Useful for brand characters or when you do not want to clone anyone.
• Use your own recording for one video. Upload or record the exact audio. No voice is needed on the avatar, and the price follows the real length of the recording.

Typed scripts can be written in any language and play in the avatar's voice.

Choosing a Lipsync Model

Omni handles its own lip movement. Photo and Twin let you go further:

• Photo keeps a default engine, or you can pick any image lipsync model from the catalog, for example Avatar IV or OmniHuman 2.0. A picked model is billed at its own price, the same as in the Lipsync tool.
• Twin keeps its default video lipsync model, or you can pick another video lipsync model from the list.

Not sure which to use? Keep the default for your first video, then run the same script on a second model and keep the take you like better.

Directing the Performance

Directing notes are a short line of up to 400 characters that tells the avatar how to deliver the script. They work in Omni and Photo, and they take priority over the default acting.

Good notes describe energy and one or two moments:

• "Calm and warm, small smile at the end"
• "Excited, laugh on the last line"
• "Serious, look down at the start"

In Photo, Kolbo.AI also reads the still itself, including what the hands hold, the framing and the mood of the words, so the movement fits the picture. Turn on Holds mic when you want a small handheld microphone in frame, like a UGC creator.

Omni checks every take: the lipsync is compared with the speech that was sent, and a take that misses or jumps to a different place is filmed again.

Camera Styles and Places in Omni

Omni offers four camera styles in the app:

• Talking head: chest up, eyes on the lens, fits any place.
• Podcast: a side angle, talking to an unseen host, on podcast and TV sets.
• Selfie: front camera, like a video message to a friend, at home, outdoors, in cafés and cars.
• Walk and talk: walks toward the camera on a street, in a park, on a beach or a rooftop.

Places include home rooms, studios, offices, a stage, a news desk, a clinic, shops, cafés and outdoor spots. Picking a style moves you to a place it suits, and you can still change it.

Using Avatars From Claude or ChatGPT With MCP

Everything above is also available to AI assistants through the Kolbo MCP server. The tools include create_avatar (photos, voice, optional talking video, or a visual_dna_id), list_avatar_lipsync_models, quote_avatar_video, create_avatar_video and animate_photo. Since @kolbo/mcp 1.133.0, create_avatar_video accepts a direction parameter for directing notes on Omni and Photo takes.

A typical request: "Make an avatar from these three photos, clone the voice from this file, then a 30 second Omni video in the podcast style." The assistant quotes the price and runs it.

Use It Responsibly

Only make avatars of yourself, of people who gave you permission, or of characters you created. Kolbo.AI asks you to confirm consent when you create an avatar and when you add a voice or a video. Do not use anyone's face or voice to say something they did not say.

FAQ

How do I make an AI avatar video from a photo?

Open Avatars in Video Tools, add 1 to 8 photos of the person, give the avatar a voice, pick a mode and place, type the script and press Generate. The whole setup takes a few minutes.

What is the difference between Omni, Photo and Twin?

Omni acts naturally in any place and camera style, up to 60 seconds, and is the most lifelike. Photo makes a still talk, up to 3 minutes, and suits longer explainers. Twin uses your own talking video and changes only the words.

Can I clone my own voice for the avatar?

Yes. Record or upload 8 to 60 seconds of clean speech and choose Cartesia or ElevenLabs to clone it. You can also pick a library voice or upload a recording for a single video.

Can the avatar speak other languages?

Yes. Type the script in any language and it plays in the avatar's voice. You can also upload your own recording in any language.

How much does an AI avatar video cost?

Each mode is priced per second of speech, and the price depends on the mode and resolution. The exact credits are shown before you generate. See pricing facts for how Kolbo.AI prices video.

Can I use an existing Visual DNA as an avatar?

Yes. Choose From a Visual DNA when you create the avatar, and its images, sheet, voice and description are copied in.

Can I get the avatar video with a transparent background?

Yes. Turn on Remove background before you generate, or apply it to a finished video. It works up to 1080p and 60 seconds and costs extra per second. Read more in the background removal guide.

Ready to build your first avatar? Start at app.kolbo.ai.