LongCat-Video-Avatar 1.6 · photo + your audio

LongCat Avatar — Turn a Photo and Your Recording Into a Talking Video

LongCat Avatar, Meituan’s open LongCat-Video-Avatar 1.6 model, makes the person in one photo speak or sing your own recording; TryH3 lets you use it online with one MP3 or WAV.

One or two people · MP3 or WAV, 3–64 s · 480p or 720p

Make a LongCat Avatar video

Photo

People in the photo
Portrait

JPG, PNG or WebP up to 10 MB. The face and mouth should be visible.

Recording

Voice or song

MP3 or WAV, 3–64 seconds, up to 20 MB. One clear voice works best; iPhone voice memos (.m4a) need converting to MP3 first.

Prompt (required)

Describe expression, gestures and camera, at least 10 characters. The server also asks the model to keep the face, hair, clothes and background from the photo.

Sample prompts (tap to load, then edit)

Model
Resolution
Length
Frame

From 11 credits (3 s at 480p). The price follows your recording’s length. · Free accounts get a moving watermark until any purchase.

LongCat Avatar 1.6 examples

Two single-person and two two-person clips with their own sound. Tap a sample prompt to see the matching clip.

Presenter talking to the camera

A presenter in a beige blazer speaks in a bright studio with open-hand gestures; 12 seconds, one person, with her voice.

Your videos ↗

LongCat Avatar on TryH3 is an independent tool, not a Meituan product: it sends your photo and recording to a hosted LongCat-Video-Avatar 1.6 API and returns an MP4 whose sound is your recording. Made for an existing voiceover, podcast clip or song, it renders 3–64 seconds at 480p or 720p with one person or two, from 11 credits for 3 seconds at 480p. Free accounts get a moving watermark until any purchase.

How to make a LongCat Avatar video

Making a LongCat Avatar video takes one photo, your recording and a short prompt. Choose one person or two, upload a sharp front-facing photo, add an MP3 or WAV of 3–64 seconds (one per person for two people), pick who speaks first, load or write a prompt and choose 480p or 720p. The price is shown from the recording’s length before you pay, and the model uses your recording as the video’s sound instead of writing new audio.

  1. 01

    Choose one person or two

    Two people means one photo with one person on the left and one on the right, plus a recording for each.

  2. 02

    Upload the photo

    A clear, front-facing photo with the mouth visible. The frame of the video follows the photo.

  3. 03

    Add your recording

    MP3 or WAV, 3–64 seconds. For two people, choose whether they take turns or talk at the same time.

  4. 04

    Write the prompt and generate

    Load a sample prompt or write your own (expression, gestures, camera), pick 480p or 720p, confirm permission and generate.

Talking and singing video examples

These four clips cover the two ways the tool works: one person driven by one recording (a presenter explaining a product and a young man singing into a microphone) and two people in one photo driven by two recordings (two TV hosts taking turns and two men chatting face to face). Each sample prompt in the tool is written for one of these clips.

Presenter talking to the camera A presenter in a beige blazer speaks in a bright studio with open-hand gestures; 12 seconds, one person, with her voice.
Singing into a microphone A young man in a grey sweater sings into a handheld microphone behind a lectern; 8 seconds, one person, with his voice.
Two hosts taking turns Two hosts in a neon TV studio: the woman on the left speaks first, then the man on the right answers; 13 seconds, two people.
A face-to-face conversation Two men face each other against a plain wall; one talks with his hands while the other listens, then they swap; 8 seconds, two people.

LongCat Avatar 1.5 vs 1.6: which version runs here?

LongCat Avatar on TryH3 runs version 1.6, which Meituan’s LongCat team released with a new technical report on 13 November 2026. Version 1.5 came first and is still the one most ComfyUI workflows and quantized downloads use. The 1.6 model card says it holds the same identity across videos longer than ten minutes and runs faster, and its report ranks it second of six systems in a 33,051-vote preference test, behind OmniHuman-2.0.

LongCat Avatar 1.5 vs 1.6: which version runs here?
VersionTechnical reportWhat the model card or report says
LongCat-Video-Avatar 1.5arXiv 2605.26486Audio-image-to-video, audio-text-to-video and video continuation; the version in most ComfyUI and GGUF guides.
LongCat-Video-Avatar 1.6arXiv 2611.18822 (13 Nov 2026)Weights under the MIT license (trademarks and patents excluded); identity held over 10+ minute videos; faster inference; second of six in the human preference test.

Is there a LongCat Avatar API?

LongCat Avatar is available two ways for developers: as open weights you run on your own GPU, and as paid hosted APIs that bill per second of audio, usually with a 3-second minimum and a 64-second cap per request. TryH3 does not sell API access; this page is a browser tool that charges credits per video, which suits a few clips a week. For hundreds of automated clips a month, a hosted API or your own hardware costs less per minute.

What recording and photo work best for LongCat Avatar?

LongCat Avatar works best with one clean voice and a sharp, front-facing photo: an MP3 or WAV of 3–64 seconds with little music or room noise, and a photo where the face is unobstructed and the mouth is visible. For two people, use one photo with one person on the left and one on the right, and a separate recording for each. These are preparation tips; we have not run paid test generations, so results vary by photo and recording.

What recording and photo work best for LongCat Avatar?
InputWorks bestAvoid
RecordingOne speaker or singer, close microphone, 3–64 sLoud backing music, echo, several voices in one file
PhotoFront or three-quarter face, even light, head and shoulders or half bodySunglasses, hands over the mouth, tiny faces in a wide shot
Two peopleBoth faces clear, one on each side of the frameOverlapping people, a third face in the photo
Longer scriptSplit at sentence breaks into parts of 64 s or less, same photo for each part, join them in an editorOne file over 64 s (it is refused here, not cut)

LongCat Avatar pricing and credits

LongCat Avatar on TryH3 is charged per second of your recording, rounded up, with a 3-second minimum: 11 credits for 3 seconds at 480p, 37 credits for 10 seconds at 480p and 73 at 720p, up to 462 credits for 64 seconds at 720p. Starting from zero, the cheapest way in is the $5 Trial pack (24 credits), which covers one 480p clip of up to 6 seconds. Two people talking at the same time cost the longer recording’s seconds; taking turns costs both recordings added together.

LongCat Avatar pricing and credits
Billed seconds480p credits720p credits
3 s1122
5 s1937
10 s3773
15 s55109
30 s109217
45 s163325
64 s231462
Account and packs
Account and packsWhat you get
Free account5 sign-up credits. Videos carry a moving watermark until any purchase.
$5 Trial pack24 credits: one 480p video of up to 6 seconds; any purchase also unlocks the clean original of earlier videos.
$9.90 pack180 credits: for example one 30-second 480p video (109 credits) or a 15-second 720p video (109 credits).
Failed videosIf the model fails, rejects the request or times out, the credits are returned automatically.
See all credit packs and plans ↗

What the model cannot do well

LongCat-Video-Avatar 1.6 animates the person already in your photo to match your audio, so it cannot write the speech for you, change the outfit or background, or add a third person. Each video is capped at 64 seconds. The prompt asks the model for accurate lip sync and the photo’s look, but results vary: fast rap, heavy background music, side profiles and covered mouths are the usual trouble spots.

  • No text-to-speech: bring a recording (you can record one on your phone).
  • 64 seconds per video; longer scripts need several videos.
  • One or two people only; the photo’s scene and clothes stay as they are.
  • 480p or 720p; no 1080p option for this model.
  • Results vary by photo and recording.

Where the LongCat-Video-Avatar model comes from

LongCat Avatar is the short name for LongCat-Video-Avatar, an audio-driven talking-avatar model from Meituan’s LongCat team built on its LongCat-Video base model. It turns a portrait plus speech or singing (optionally with a text prompt) into a lip-synced video, and also supports text-plus-audio generation, video continuation and two-person dialogue. The 1.6 weights on Hugging Face are under the MIT license; the project page’s ethics note says its own demo content is for academic use. InfiniteTalk from MeiGen is the model it is most often compared with.

LongCat-Video on GitHub ↗

More questions

Is LongCat Avatar free on TryH3?

No. LongCat Avatar on TryH3 is not free: the shortest video (3 seconds at 480p) costs 11 credits, more than the 5 sign-up credits on a new account. The $5 Trial pack covers a 480p video of up to 6 seconds. Free accounts get a moving watermark until any purchase.

Can I run the model myself with ComfyUI or Hugging Face?

Yes, if you have a large GPU. The weights are on Hugging Face (meituan-longcat/LongCat-Video-Avatar-1.6), the code is in the LongCat-Video GitHub repository, and ComfyUI support comes through community nodes such as kijai’s WanVideoWrapper. Running it locally costs nothing beyond hardware and setup time; this page is for people who would rather upload a photo in a browser.

Can I use these videos commercially?

The 1.6 model weights are published under the MIT license, which excludes Meituan’s trademarks and patents, and what you make here follows our Terms of Service. You still need permission from the person in the photo and from whoever owns the voice or song, and the project page’s academic-use note means you should read the model card yourself before client work.

Does it work on a phone?

Yes. The tool runs in Safari or Chrome on a phone and there is no app to install. Phone voice memos are often .m4a files, so export or convert the recording to MP3 first, then upload it with a photo from your camera roll.

Terms of Service ↗