LongCat Avatar (LongCat-Video-Avatar) is Meituan's open audio-driven avatar model; Genjutsu AI runs version 1.6 online and is not affiliated with Meituan. Upload one portrait and your own 2–64 second MP3 or WAV to get a 480p or 720p MP4 as long as your audio, with your recording as its sound. A 10-second 480p video costs 27 credits (about $1.00 at the Pro plan rate); 3 free sign-up credits do not cover a clip; videos carry a watermark until you buy.
How the LongCat Avatar generator works
- 01
Choose one person or two
One person needs one portrait and one audio clip. Two people need one photo with both of them side by side and one clip for each: the left clip drives the person on the left.
- 02
Add the photo and your audio
A clear, front-facing photo (JPG, PNG or WebP up to 10 MB) and an MP3 or WAV of 2–64 seconds: a voiceover from your phone, a podcast line or a song. The price appears as soon as the audio is added.
- 03
Write or load a prompt, pick the resolution
The prompt box opens on a sample (presenter or singer for one person, two hosts or a conversation for two, each matching an example below); tap another sample or describe the expression and gestures yourself. Then 480p or 720p (720p costs about twice as much).
- 04
Generate and download the MP4
Tick the permission box and generate. LongCat Avatar usually needs 5 to 12 minutes; you can leave the page. Watch the lips on fast words, then download. If the video service fails to make the video, the credits come back.
LongCat Avatar video examples
Four LongCat Avatar 1.6 videos: a presenter, a singer with a microphone, two hosts taking turns and a face-to-face conversation. Watch the lips on each word, the pauses where the face keeps moving naturally, and whether the face and clothes stay the same from start to end. They are reference examples at 832×480; your video takes its frame from your photo and its sound from your audio.
Presenter talking to the camera
A presenter in a beige blazer explains a product in a bright studio; her lips follow every word and her hands open on each point.
单人 · 12.2 s · 带输入语音
Singing into a microphone
A young man in a grey sweater sings into a handheld microphone behind a lectern, closing his eyes on the long notes.
单人 · 7.6 s · 带输入语音
Two hosts taking turns
Two hosts in a neon TV studio: the woman on the left speaks and laughs first, then the man on the right answers while she listens.
双人 · 13.2 s · 带输入语音
A face-to-face conversation
Two men face each other against a plain wall; one talks with his hands while the other listens and nods, then they swap.
双人 · 8 s · 带输入语音
Two-person LongCat avatar conversations
Two-person mode uses LongCat Avatar 1.6 Multi: one photo of two people side by side, one audio clip per person, and a speaking order. “Left, then right” and “Right, then left” make a turn-taking conversation where the listener nods and reacts; “At the same time” suits a duet or two people laughing together.
| Order | Video length | Billed seconds | Good for |
|---|---|---|---|
| Left, then right / Right, then left | Both clips together (64 s at most) | Each clip rounded up, then added | Interviews, podcasts, sketches |
| At the same time | The longer clip | The longer clip, rounded up | Duets, cheering, reactions |
Use a photo where both faces are clear and do not overlap; a group photo with three or more people does not work, because the model drives exactly one person on each side.
Is LongCat Avatar open source? Where you can run it
Yes. LongCat Avatar (LongCat-Video-Avatar) is an open model from Meituan's LongCat team: the 1.5 and 1.6 weights are on Hugging Face under meituan-longcat, the inference code is in the LongCat-Video repository on GitHub, and the community runs it in ComfyUI through the WanVideoWrapper nodes. Running it yourself is free but needs a large NVIDIA GPU, the model download and a Python setup.
If you do not have that hardware, a hosted version is the practical route: hosting providers sell LongCat Avatar 1.6 by the second, and this page runs that same 1.6 model behind a simple upload form. What changes between the routes:
| Route | What you need | Cost | Length |
|---|---|---|---|
| Self-hosted (Hugging Face weights, GitHub code, ComfyUI) | A large NVIDIA GPU, the download, a Python or ComfyUI setup | Free software; your hardware and electricity | Long videos by continuing segment after segment |
| Pay-per-second API | A developer account and code to send the photo and audio | Billed per second of audio | Up to about 64 seconds per request |
| This page | A browser, one photo, your MP3 or WAV | 9–345 credits per video | 2–64 s per video |
LongCat Avatar 1.5 vs 1.6: which version runs here?
This page runs LongCat Avatar 1.6, the newer version: its technical report came out on 13 November 2026 (arXiv 2611.18822), after the 1.5 report (arXiv 2605.26486). Both turn a portrait and audio into a talking or singing video with natural motion in silent stretches; 1.6 is described as faster, steadier over long videos (over ten minutes when run locally) and stronger at keeping the same face.
In the 1.6 report's human preference test, 33,051 votes compared six systems (LongCat-Video-Avatar 1.5 and 1.6, OmniHuman-1.5, OmniHuman-2.0, HeyGen and Kling Avatar 2.0); 1.6 came second, after OmniHuman-2.0. Some free demos online still run 1.5, so check the version number before comparing results. There is no LongCat Avatar 2 yet; searches for it usually mean 1.6.
What photo and audio does LongCat audio-image-to-video need?
LongCat's audio-image-to-video mode needs one portrait with a clearly visible face and one speech or song clip, plus a short text prompt that only steers expression and gestures (on this page the prompt box is required and opens on a sample). A front-facing, evenly lit head-and-shoulders or half-body photo gives the cleanest lip-sync, and clean vocals without loud music give the most accurate mouth shapes.
| Input | Works best | Avoid |
|---|---|---|
| Photo | Front-facing face, mouth closed or relaxed, even light; JPG, PNG or WebP up to 10 MB, 300 px or more | Sunglasses, hands over the mouth, extreme side profiles, tiny faces |
| Audio | MP3 or WAV, 2–64 s, one clear voice per clip, music quieter than the voice | M4A or video files (convert first), several voices in one clip, heavy echo |
| Length | The video is exactly as long as the audio | More than 64 s: split the script into parts and make one video per part |
| Frame | The video keeps the photo's shape: crop to 9:16 for Reels, 16:9 for YouTube first | Expecting the model to reframe or zoom the photo |
LongCat Avatar pricing and credits
LongCat Avatar is priced by the length of your audio, rounded up to whole seconds with a 3-second minimum: a 10-second 480p video costs 27 credits (about $1.00 at the Pro plan rate), the same at 720p 54 credits, 30 seconds at 480p 81 credits, and a full minute at 720p 323 credits. The smallest one-time pack that covers a video is $5 for 24 credits; a 10-second video needs the $12 pack (220 credits), with no subscription. Every price is shown on the button before you start, and a video the service fails to make returns its credits.
| 音频时长 | 480p | 720p |
|---|---|---|
| 3 秒 | 9 (≈$0.33) | 17 (≈$0.63) |
| 5 秒 | 14 (≈$0.52) | 27 (≈$1.00) |
| 10 秒 | 27 (≈$1.00) | 54 (≈$2.01) |
| 15 秒 | 41 (≈$1.52) | 81 (≈$3.01) |
| 30 秒 | 81 (≈$3.01) | 162 (≈$6.02) |
| 45 秒 | 122 (≈$4.54) | 243 (≈$9.03) |
| 60 秒 | 162 (≈$6.02) | 323 (≈$12.01) |
| 64 秒 | 173 (≈$6.43) | 345 (≈$12.83) |
| Free | 3 sign-up credits (less than one clip) |
|---|---|
| Watermark | A moving watermark until your first purchase of any pack or plan; after that the clean file of the same video downloads |
| Two people | Taking turns bills both clips together; at the same time bills the longer clip |
| Account | Sign-in needed to generate; examples and prices need none. Works in a phone browser. |
Dollar figures use the Pro monthly rate ($29 for 780 credits). All packs and plans are on the pricing page.
What LongCat Avatar cannot do well
LongCat Avatar is strong at one or two clearly visible faces speaking or singing to clean audio. It is weaker with very fast speech, heavy accents over loud music, faces turned away from the camera and busy group photos; it does not change the background, clothes or camera angle of your photo, and it does not translate or clone a voice — the sound is always the recording you upload.
| What you see | Usual cause | Fix |
|---|---|---|
| Odd teeth or a smeared mouth | Mouth hidden or open wide in the photo, low resolution | A sharper photo with the mouth relaxed; try 720p |
| Lips behind fast words | Rapid speech, music louder than the voice | Slower delivery, the voice clearly on top of the music |
| Stiff body | A tight face crop leaves nothing to move | A half-body photo; ask for hand gestures in the prompt |
| Blurry after posting to TikTok | 480p upscaled by the app | Choose 720p and a 9:16 photo |
| The wrong person talks | Clips swapped in two-person mode | Left clip for the person on the left, right clip for the right |
| Audio refused | M4A, a video file, or over 64 s | Export as MP3 or WAV, split long scripts |
What is LongCat Avatar (Meituan LongCat-Video-Avatar)?
LongCat Avatar is the short name for LongCat-Video-Avatar, an audio-driven avatar model built by the LongCat team at Meituan on top of its LongCat-Video model. It is not a TikTok trend or a cat filter: people search “longcat ai avatar” or “meituan longcat avatar” after seeing its talking and singing demos on AI video forums. It is often compared with InfiniteTalk and Wan S2V, two other open models that make a photo speak. This site runs the published 1.6 model through a hosting provider and is not affiliated with or endorsed by Meituan.
More voice and music tools on this site: make your pet sing on the singing dog AI page, or make a song cover with the AI song cover generator.
最近更新: 2026-11-19