AI Lip Sync — make a photo talk or sing
Use AI Lip Sync to make a portrait speak or sing with your own audio. Audition a 4–15-second excerpt, choose 768P or 2K and a frame shape, then preview and download an MP4. Sign in with at least 20 usable credits; every request costs 20 credits, including for gift and check-in balances.
Portrait
One clear face · JPG, PNG or WebP · up to 30 MB · at least 256 × 256
Voice or song
Choose 4–15 seconds with clear speech or singing.
Loading…
Explore portrait performances for spoken messages, podcasts, lessons, and singing.
Podcast host
Classroom lecturer
Singing performance
AI Lip Sync: from a portrait and audio to a performance clip
01
Choose a spoken line or a sung phrase
Start with the sound you want the person in the picture to perform. A spoken introduction, a short message, and a sung refrain ask for different expressions, even when they use the same portrait. Listen to the recording first and decide whether a viewer should focus on the words, the melody, or the character's reaction. For a singing photo, pick a passage where the vocal line is easy to follow. For speech, choose a complete thought with an audible beginning and ending. You are supplying the audio, not asking the page to invent a script or compose a song.
A talking portrait can introduce a project, deliver a greeting, or explain a short idea. A singing portrait can perform a recognizable line from your own song. Start with one complete phrase and a portrait whose expression suits it. The finished MP4 uses the audio excerpt you selected, so listen to that excerpt while planning the performance.
02
Give the face enough room in the portrait
Choose one clearly visible face with the mouth unobstructed. A front-facing or gently angled portrait is a practical starting point; heavy shadows, hands over the lips, and a very small face make it harder to judge the result. Leave space around the head so the composition still works in the output shape. Avoid treating a higher resolution as a repair for a blurry source. Sharp facial detail in the original photograph matters more than simply enlarging the file. If several people appear, prepare a portrait centered on the person who should perform rather than expecting a speaker selector.
The upload accepts JPG, PNG, and WebP images up to 30 MB. The current server also requires both sides to be at least 256 pixels and rejects images over 32 million pixels. You can use an available character portrait instead of uploading a new file; access to saved characters follows the account controls. Check that the selected picture is the intended one before submitting. The page fits the portrait within the chosen frame, so a different aspect ratio may leave borders. A tight crop prepared for one shape will not necessarily look equally balanced in another.
03
Prepare audio that you can hear and understand
Upload MP3, WAV, M4A, AAC, OGG, or FLAC audio, or choose music available in your account's cloud library. Local audio uploads can be up to 30 MB; a valid file extension alone is not enough if the file cannot be decoded. Play it in the clip editor and check that it is the recording you intended. If the browser cannot preview a particular encoding, try a playable version before spending credits. A clear voice gives you a more useful basis for evaluating the mouth movement than a passage dominated by crowd noise or a long instrumental section.
Choose a passage with one clear lead voice. If the accompaniment masks the words, first separate the vocal with our Vocal Remover, then upload that vocal file here. Keep the opening consonant, the end of the phrase, and natural breathing space. Audition the cleaned excerpt before selecting the portrait’s performance direction.
Separate a clear vocal with Vocal Remover04
Audition a complete 4–15-second excerpt
A generation uses an integer duration between 4 and 15 seconds. The selected start can fall between whole seconds, which helps you begin just before the first word rather than chopping off its opening consonant. Use the waveform and playback to locate a phrase, then listen to the whole selection. Check both ends: a clip that begins well may still cut off the final word or a sustained sung note. You can upload a longer recording and choose an excerpt, but the page does not turn the entire song into one long lip syncing performance.
The start and duration determine the audio sent for this request and the excerpt used in the final video. They are not separate settings for billing and playback. If the selected range runs beyond the recording, choose an earlier start or a shorter duration. For a first test, a short self-contained sentence often gives clearer feedback than fifteen seconds covering several unrelated ideas. Keep a note of the excerpt when comparing attempts, so you can tell whether a changed portrait or instruction helped rather than accidentally comparing different words.
05
Pick a frame and quality for the destination
Choose 768P or 2K and one of five frame shapes: 9:16, 16:9, 1:1, 4:3, or 3:4. The initial settings are 768P and vertical 9:16. A vertical frame suits a portrait-led clip; a wider frame gives surrounding details more space. Review the image preview and select the composition that works for your destination.
Each generation costs 20 credits for either 768P or 2K and any supported 4–15-second clip. Quality and duration affect the video settings, while the request price stays the same. Before sending your portrait and audio, the workbench checks your available credits and shows an immediate message if the balance is too low. It refreshes the balance before upload, and the server checks it again when deducting credits. Keep the same excerpt when comparing portraits or performance directions, and plan how many new requests your remaining balance can cover.
06
Describe the performance, then review it with the sound on
The optional performance field accepts up to 1,200 characters. Describe a calm expression, gentle head movement, or a steady camera using a few clear priorities. Match the visual delivery to the recording: a soft spoken line and an energetic refrain call for different emphasis. Compare attempts using the same audio excerpt.
Watch the result with the sound on. Check the first word, fast consonants, longer vowels, and the ending, then look at the eyes, teeth, and face outline. If you want another version, change the portrait, excerpt, or performance direction one at a time. Download the MP4 once you have reviewed the full performance.
07
Check access, save the MP4, and use authorized material
Sign in and check that your usable balance is at least 20 credits. Registration gifts, check-in rewards, subscription and purchased credits all count; no paid plan is required. Administrators pay the same 20 credits. You can prepare the portrait and audition local audio first, then review the settings before sending. Save the finished MP4 with its source files. Confirmed request rejection or generation failure returns the deducted credits, and an identical request is not charged twice.
Use photos and audio you own or have permission to publish. For a recognizable person, confirm permission for the intended performance; for a song, confirm the relevant usage rights. Keep the source files and permissions with the finished clip. Follow the destination platform’s guidance for labeling generated performances.
To prepare an excerpt before choosing the performance, trim your recording with the audio cutter.
For a new musical idea to use as source material, explore the AI music generator.
Questions about making a lip sync video
How do I make a photo talk or sing?
Upload a portrait or choose an available character, add audio, and audition a 4–15-second excerpt. Set quality, frame shape, and optional performance instructions, then review the estimate and sign in to generate.
Which image and audio files can I upload?
Images can be JPG, PNG, or WebP; audio can be MP3, WAV, M4A, AAC, OGG, or FLAC. Each upload can be up to 30 MB and must pass media validation. Check that the audio plays before submitting.
Can I upload a full song and choose a section?
Yes, if the file is within the upload limit and decodes correctly. Use the clip editor to select a 4–15-second excerpt. One request does not generate a full-song video.
How do I choose the quality and video ratio?
Select 768P or 2K and one of the five frame shapes. Check the portrait composition and the updated estimate; higher quality does not guarantee better lip synchronization.
Do I need to sign in, and how are credits calculated?
Sign in with at least 20 usable credits. Every generation costs 20 credits for both 768P and 2K, regardless of the supported clip length. Gift and check-in credits can be used, and administrators pay the same amount. Confirmed failures return the credits.
What should I check if the lips do not match well?
Listen for a clear voice, inspect the selected phrase boundaries, and use an unobstructed portrait. Try simpler performance instructions and compare one change at a time. Perfect word-by-word synchronization is not guaranteed.