AI Characters

Create a consistent character once and use them everywhere: in images, in videos, in pipelines and from Claude. A character can be a real brand ambassador (from photos of them) or an original AI persona, such as a 101-year-old health influencer, generated from scratch.

Each character has:

• 1 to 5 reference images of the same person. A character sheet works best: several angles, a full-body shot and a close-up on a plain white background, each as a separate image.
• A description of their appearance and persona, added to your prompts to keep them consistent.
• Style notes (optional), for example "Realistic skin texture, natural light, not airbrushed".
• A voice (optional), for talking videos and voiceovers. See Voices.

The Characters page

Open Characters in the menu to see all your characters, with their images, voice and how much they've been used. Click a character to open their profile, where everything about them lives in one place:

Section	What you'll find
Quick actions	Generate an image or video with the character, run a character template, or lip-sync one of their videos
Profile	Description, style, voice and reference images. View images large, reorder them, or add and remove them (a character keeps at least one)
Voice	Their voice settings and a playable history of every voice preview
Gallery	Every image and video made with the character, filtered by kind (images, videos, talking clips, lip-sync, voiceovers, captions), with players, downloads and a link to each job
Audio	The spoken tracks from their lip-sync and voiceover clips, with the script
Scripts	Every line they've spoken. Reuse one to lip-sync or run a template again
Runs	Pipeline runs that used the character

Download all saves the character's reference images, finished media, speech and voice previews as one zip (up to 150 files and 400 MB, newest first). A manifest.json inside lists everything, including anything that was skipped.

Voice previews and speech tracks also appear under Content Library → Audio.

Create a character

1. Go to Characters (or the Character Library on the Generate page)
2. Click + New Character
3. Enter a Character name, for example "Amos - 101-year-old health influencer"
4. Add a Description and optional Style notes
5. Add 1 to 5 images using From Library (images you've already generated), Upload or Use URL
6. Optionally choose a Voice
7. Click Save Character

Create an AI character from scratch

1. Generate one strong head-and-shoulders portrait on a white background, with realistic skin. This is the identity anchor.
2. Make the other views (three-quarter, side profile, full body) as separate images edited from the anchor, so every view has the same face. One composite sheet with many views on it holds far less detail per view.
3. Use GPT Image 2.5 Sunburst when details must be exact, such as colours, button counts or which side the hair is tucked.
4. Check each image closely (in Claude, ask it to zoom in with view_image), then save the best as the character.

Create an AI influencer walks through this with a real example.

Click Edit to change a character later. Replacing the images, description or voice applies to everything you generate afterwards.

Generate with a character

Select a saved character, write a prompt describing the scene, and generate. How the character is used depends on the model:

Model	How the character is used
Grok Image 2.0 / Grok Image	The character's images are added as reference images (5 in total with any others you add), and their description and style notes are added to your prompt
GPT Image 2.5 Sunburst / Flare	The same as Grok: the character's images are reference images (5 in total) and their description is added. Sunburst holds exact details, such as clothing colours, most reliably
Luma Photon 1 / Photon Flash	The character's photos become Luma's character reference (up to 4)
Grok Video 1.5, text to video	Up to 3 of the character's images guide the video (reference to video, up to 720p). Add a Spoken script to have them speak to camera
Grok Video, image to video	Your image stays the first frame and the character's description is added to the prompt. Tip: animate an image you generated of the character

Character videos from text are capped at 720p. For 1080p b-roll, generate an image of the character first, then animate it.

Voices

A character's voice is used for talking videos and voiceovers:

Voice type	Used for	Notes
xAI preset (eve, leo, rex, ara, gork, aurora)	Talking videos and voiceovers	The character speaks on camera with matching lip movement
OpenAI voice with a Voice direction	Voiceovers and lip-sync	Choose a base voice (for example cedar, ash, marin) and describe how it should sound, for example "Slow, warm, slightly gravelly; gentle humour; pauses between sentences; rural American accent"

Click Preview to hear the voice on a short line before you save. Previews use no allowance, and each preview is kept in the character's voice history so you can compare them later.

For a character with an OpenAI voice, also set Voice in talking videos (xAI preset): the preset Grok uses for Voiced video clips, which can't speak OpenAI voices.

Talking videos in a designed voice

Grok's talking videos (Voiced video) can only speak xAI presets. For a character with an OpenAI voice, use Lip-sync instead: it makes the character speak your line in their designed voice, with matching mouth movement. The Character talking clip (designed voice) template does this in one run. If you use Voiced video with an OpenAI-voiced character, it falls back to Voice in talking videos (xAI preset).

Characters in pipelines

Add a Character (saved person) node to a pipeline and connect it to the refs input of Generate image (up to 5 images) or Voiced video (up to 3). The node is free and creates no job. Voice steps use the character's voice unless you set one on the step.

Three templates on the Pipelines page start from a character:

• Character talking clip (character → voiced video)
• Character scene (character → image → video)
• Character talking clip (designed voice): character, image, video, lip-sync and captions in one run

Add an Add captions (burn in text) step to burn the script into the video as subtitles, styled for 9:16, either spread evenly or timed to the speech. See the Pipeline editor for details. To add captions or timed text cards to a clip you've already made, use Overlays & Captions; the new video stays linked to the character.

Batch campaigns with characters

Character + template batch

1. Select a character
2. In the template gallery, check multiple templates (multi-select mode)
3. Click Generate Campaign
4. Each template generates a separate creative, all with your character

Multi-platform batch

1. Select a character and write a prompt
2. Check "Generate for all platforms"
3. Click Generate. This creates 3 images:
   • 9:16 for TikTok and Instagram Reels
   • 16:9 for YouTube and websites
   • 1:1 for Instagram feed and Facebook

Style reference

You can also lock a visual style alongside the character:

1. Upload a reference image that defines the look you want (for example a polished magazine ad, a colour palette or an artistic style)
2. Set the weight (0.0 to 1.0) to control how strongly the style is applied
3. Generate: your character appears in the scene with that visual style

Photo tips

Do	Don't
Clear, well-lit face shots	Sunglasses or face coverings
Different angles (front, 3/4 profile)	Heavy shadows across the face
At least one full-body shot	Group photos with other people
Neutral expressions	Extreme expressions or motion blur
3 to 5 images	Just one image, or two different-looking faces

Use cases

Use	Character
AI influencer	An original persona posting advice, stories or product content
Real estate	An agent appearing across all property listings
E-commerce	The same model wearing or using different products
Fitness	A trainer demonstrating different exercises
Food and beverage	A chef or barista across menu items
Personal brand	A creator across a content series

Good to know

• Characters belong to your workspace, so every team member can use them.
• Images and voices can be changed at any time. Deleting a character is only possible in the app (on their profile), not through Claude.
• Speech from lip-sync and voiceover clips is kept so you can play and download it later. Clips made before 28 September 2026 don't have a separate speech track.
• Generating with a character uses your plan allowance like any other generation. Creating a character and previewing voices don't.