Mirage Avatar XThe AI Avatar Generation Model Built for Identity Preservation and Expressivity
Build expressive avatar experiences with the Avatar X API, powered by Mirage's most advanced generation model yet. Mirage Avatar X is an AI avatar generation model that sets a new standard for identity preservation and expressivity in AI video. Get started with an audio and optional visual reference, or send a script and let Avatar X generate the voice for one of ten stock avatars.Information updated as of August 12, 2026.
Start building with the Mirage Avatar X API
Two endpoints cover both directions. Send a script and Avatar X generates the voice and the performance, or send audio and it builds the performance around your track.
The Avatar X API offers access to Mirage's most advanced generation model yet, delivering industry-leading identity preservation and expressivity in AI video
The Avatar X API offers access to Mirage's most advanced generation model yet, delivering industry-leading identity preservation and expressivity in AI video
What Makes Mirage Avatar X Different?
Generate an AI Avatar Video Straight From a Script
Send a script to the text-to-video endpoint and Avatar X generates the voice, the lip sync, and the performance together in a single AI video generation.• Scripts run from 50 to 1,500 characters, up to 180 seconds of speech.• Ten stock avatars ship with the model, each in 9:16 and 16:9.• No voice track, no reference footage, and no rigging required.
Drive an AI Avatar Video With Your Own Audio
Pass an audio track to the reference-to-video endpoint and Avatar X builds the matching performance around it, so the delivery follows your recording rather than a synthesized read.• Audio from 3 to 180 seconds, in MP3, WAV, AIFF, AAC, ALAC, Opus, Vorbis, or FLAC.• Pair it with a still image, a reference video, or a stock avatar.• Your own voiceover, dialogue, or text-to-speech output all work.
Hold a Likeness Across the Whole AI Video Generation
Avatar X generates motion, audio, and expression as one performance instead of assembling them separately, which is what keeps a face recognizably itself from the first word to the last.• Microexpressions, eye contact, and head motion stay tied to the voice.• Reference video up to 120 seconds, reference image up to 25 MP.• Quality holds across long takes, not just short social cuts.
What do Mirage Avatar X generations look like?
Every clip below is the same script delivered by a different stock avatar from Mirage's catalog. The words never change, so what you are comparing is the performance: pacing, emphasis, expression, and the way each face holds its own identity.
Food and home, kitchen
"Okay, so many of you have been asking about this, and honestly, I've been excited to share it. Today, I want to walk you through the three things that matter most, and why they're simpler than they look."
Sports and hobbies, trophy room
"Okay, so many of you have been asking about this, and honestly, I've been excited to share it. Today, I want to walk you through the three things that matter most, and why they're simpler than they look."
News desk, broadcast studio
"Okay, so many of you have been asking about this, and honestly, I've been excited to share it. Today, I want to walk you through the three things that matter most, and why they're simpler than they look."
Business update, city skyline
"Okay, so many of you have been asking about this, and honestly, I've been excited to share it. Today, I want to walk you through the three things that matter most, and why they're simpler than they look."
Arts and culture, gallery
"Okay, so many of you have been asking about this, and honestly, I've been excited to share it. Today, I want to walk you through the three things that matter most, and why they're simpler than they look."
Hospitality, wine bar
"Okay, so many of you have been asking about this, and honestly, I've been excited to share it. Today, I want to walk you through the three things that matter most, and why they're simpler than they look."
How to use the Mirage Avatar X API
The client API handles the request submit protocol. It will handle the request status updates and return the result when the request is completed.
import { fal } from "@fal-ai/client";
const result = await fal.subscribe("mirage-api/avatar-x/text-to-video", {
input: {
script:
"Put in a script and Mirage Avatar X turns it into a clear, expressive video. Let's bring your next idea to life.",
avatar: "Jasmine",
},
logs: true,
onQueueUpdate: (update) => {
if (update.status === "IN_PROGRESS") {
update.logs.map((log) => log.message).forEach(console.log);
}
},
});
console.log(result.data);
console.log(result.requestId);
Two endpoints, one price
Both Mirage Avatar X endpoints bill at $0.30 per second of generated video. Pick the one that matches the input you already have, and pay only for the seconds you generate.
Who builds with Mirage Avatar X?
From marketing and localization teams to creators and agencies, people across industries are building AI avatar video with Avatar X on fal.
Spokesperson Video Without a Shoot
Produce product explainers, launch announcements, and ad variations from a script. Test new messaging by regenerating the copy instead of rebooking talent, a studio, and a crew.
Publish on the Days You Do Not Film
Keep a consistent on-screen presence across YouTube, TikTok, and Shorts. Write the script, pick the avatar, and publish on a writing schedule rather than a filming schedule.
One Performance, Every Language
Generate the same talking video in market after market by pairing a reference with localized audio. Reach new audiences without new footage or on-camera talent.
A Presenter for Every Lesson
Give courses a friendly on-screen instructor and refresh a module by regenerating the script. Ten stock avatars cover a wide range of subjects and audiences.
Guided Onboarding and In-App Help
Add a talking guide to onboarding flows, help centers, and release notes. Update the message any time by sending new copy through the same endpoint.
Volume Work Through One API
Batch hundreds of variations for clients on fal's infrastructure, billed per second of generated video with no seat licences and no minimum commitment.
Common questions about Mirage Avatar X
What is Mirage Avatar X?
Mirage Avatar X is an AI avatar generation model from Mirage, the research lab formerly known as Captions. It generates the motion, the audio, and the facial expression of an avatar as a single performance rather than assembling them in separate passes, which is what gives it its identity preservation and expressivity. On fal it is available through two endpoints, text-to-video and reference-to-video.
How do I use the Avatar X API?
Pick the endpoint that matches your input. Send a script to mirage-api/avatar-x/text-to-video and the model generates the voice and the performance for a stock avatar. Send an audio track to mirage-api/avatar-x/reference-to-video, optionally with an image or a reference video, and the model builds the performance around your audio. Both return a finished MP4.
What references does Avatar X accept?
Reference-to-video takes a driving audio track from 3 to 180 seconds, plus an optional visual reference. That visual reference can be a still image in JPG, PNG, or WebP with a short edge of at least 512 px and up to 25 megapixels, or a reference video from 1 to 120 seconds in H.264, H.265, ProRes, VP8, or VP9. Both must be 9:16 or 16:9. If a video and an image are both supplied, the video wins.
Can I use my own voice with Avatar X?
Yes. The reference-to-video endpoint follows whatever audio you pass in, so recorded voiceover, dialogue, and text-to-speech output all work. The text-to-video endpoint can also take a voice reference: supply audio_reference_url together with video_reference_url and the pair replaces the stock avatar entirely. A voice reference runs from 10 to 180 seconds, and only the first 60 seconds are used.
What stock avatars are available?
Ten stock avatars ship with the model: Ayesha, Farhan, Giulia, Jasmine, Luke, Maya, Michael, Neha, Tariq, and Valerie. Each is available in a 9:16 portrait and a 16:9 landscape variant, selected by passing the exact name, for example "Michael (16:9)". Every clip in the examples grid above is one of these avatars reading the same script.
How much does the Mirage Avatar X API cost?
Both Avatar X endpoints bill at $0.30 per second of generated video, so a 30 second video costs $9.00. There are no seats, no subscription, and no minimum commitment. You pay for the seconds you generate.
How do I get started with Mirage Avatar X on fal?
Avatar X is available now. Try it in the reference-to-video playground or the text-to-video playground with no setup required. To integrate it, grab an API key from your dashboard and call it with the Python or JavaScript SDK.
Get in Touch About Mirage Avatar X
Want to bring Avatar X into your product or workflow? Leave your details and our team will reach out.