Troubleshoot LipDub AI: Common Issues and Best Practices

Start with these checks

Before changing your project, confirm the following:

1. You selected the correct project type.
   Use a Multi-speaker project whenever more than one face appears at any point in the video— even when only one person speaks.
2. The face is clearly visible.
   The mouth, lips, jaw, and chin should not be covered by hair, hands, microphones, props, or the edge of the frame.
3. The source and training footage are consistent.
   Large differences in resolution, color, lighting, framing, or camera angle can increase artifacts.
4. The audio is clean and correctly timed.
   Noise, unintended dialogue, missing silence, and incorrect timecodes can all affect the result.
5. The correct generation settings were selected.
   Check the actor, audio file, language or version, Expressiveness level, lip-sync regions, and output quality before generating again.

Video upload and processing issues

My video will not upload

Confirm that:

• The file is no larger than 15 GB.
• The video is a MOV or MP4.
• It uses H.264 or a supported Apple ProRes codec.
• It uses a supported Constant Frame Rate.
• It is not interlaced or anamorphic.
• It does not contain multiple video streams.
• It is not an image sequence such as an EXR sequence.
• It uses a 1:1 pixel aspect ratio.

Phone footage is often recorded using a Variable Frame Rate. Convert it to a Constant Frame Rate before uploading.

For the complete specifications table, see Prepare Your Content.

Processing is taking longer than expected

Long videos, high-resolution footage, and higher training levels can require more processing time.

You can leave the project page and return later. LipDub will continue processing while you are away.

Contact support: If one processing, training, audio-generation, or final-generation action remains in progress for more than two hours, contact support@lipdub.ai.

The output colors do not match the source

For the most predictable result, use SDR footage in sRGB or Rec. 709.

LipDub may accept HDR, Rec. 2020, and Rec. 2100 footage, but it cannot guarantee an exact color match. Convert the footage to a supported SDR color space when color accuracy is critical.

Also avoid mixing training and source footage with substantially different color grades or color casts.

My 4K video was generated at 1080p

The selected output quality controls the maximum result resolution:

• Standard produces output up to 1080p.
• Pro supports output up to 4K.

Generate the video again using Pro when a 4K result is required.

Project type and face issues

I used a single-face workflow, but another person appears in the video

Create a Multi-speaker project instead.

Translation, Personalization, and Dialogue Replacement projects are intended for footage with one visible face. Use Multi-speaker whenever another face appears anywhere in the footage, including:

• Background people
• Brief cutaways
• Interviewers or guests
• Silent people in frame
• Faces that appear only in one scene

Multi-speaker is the only workflow that lets you control which detected faces are assigned to each actor and which people are trained.

A face was not detected

Check whether the face is:

• Large enough to see clearly
• Inside the frame
• Well lit
• Facing primarily toward the camera
• Visible for enough time
• Free from heavy occlusion

Detection may be less reliable when:

• Only part of the face is on screen
• The shot is an extreme close-up
• Hair or an object covers the mouth and jaw
• The subject is very small in the frame
• The image is dark, blurred, or heavily compressed

In a Multi-speaker project, review Discarded face detections. A correct detection may be available there and can be assigned to the appropriate actor.

The wrong face was assigned to an actor

In the Multi-speaker project:

1. Open Label Actors.
2. Select the affected actor.
3. Review all thumbnails under Face detections.
4. Select any thumbnails belonging to another person.
5. Select Remove selected.
6. Review Discarded face detections for correct thumbnails that need to be restored.
7. Retrain the actor after changing their detections.

Each actor’s training set should contain only footage of that person.

Do I need to label every visible face?

No. You only need to label and train the people whose lips you want to change.

However, the project must still be created as a Multi-speaker project whenever multiple faces appear.

LipDub says an actor’s tracks were modified

Face detections were changed and/or a new video was uploaded to the project/scene after the actor was trained.

Return to Train AI and train that actor again before generating the next result.

Actor training failed

First confirm that the actor has enough useful training data.

Training footage should show the subject:

• Actively speaking
• With their lips visible
• In lighting and framing similar to the source footage

As a general guide:

• 30 seconds of visible speaking footage is a practical minimum.
• One minute is recommended.
• Additional footage can help up to approximately five minutes.
• Footage beyond five minutes usually provides diminishing improvement.

Add supplemental speaking footage when needed, review the actor’s face detections, and start training again.

Technical training failures are normally refunded automatically. Check Credit Usage History and contact support@lipdub.ai if the expected refund does not appear.

Lip-sync quality issues

The lip-sync looks soft, unstable, or unnatural

Review the following:

• Is the full face, jaw, chin, mouth, and lip area visible?
• Is the shot an extreme close-up?
• Is hair or another object covering the mouth?
• Does the subject repeatedly leave the frame?
• Does the training footage match the source footage?
• Does the training footage show the subject actually speaking?
• Were any incorrect faces included in the actor’s training data?
• Was Turbo used for a result that requires actor-specific training?

Turbo uses LipDub’s generic base model. It is useful for previews but does not train a reusable actor model. Use Flash, Premium, or Ultra when actor-specific training and higher consistency are required.

The mouth movement is too subtle or too pronounced

Adjust Expressiveness before generating another version.

Expressiveness is a visual output preference. Choose the setting that best matches the performance and production style.

AI-generated footage has inconsistent texture or lip movement

Enable Optimize for AI generated video when generating from AI source footage.

This setting uses model tuning designed for AI-generated video. It is recommended for AI source footage, but you can create versions with the setting enabled and disabled to compare them.

Leave it off for conventional human footage unless you are testing alternatives. It can reduce visual texture quality on real human shots.

Jump cuts feel different in a translated version

Jump cuts are supported, but translated phrases do not always have the same duration or word boundaries as the source language.

A cut placed tightly between source words may therefore feel slightly different after translation. Review the pacing and timing around every jump cut before generating the final result.

Translation and voice issues

The translation is incorrect

Check the source transcript first. An incorrect source word can lead to an incorrect translation.

In the Translation editor:

1. Play the source segment.
2. Correct any transcription errors.
3. Review the translated text.
4. Edit the translation where required.
5. Regenerate or replay the translated audio.
6. Review the result in context with the surrounding segments.

Use a project context prompt when the translation requires a particular tone, audience, or subject-matter understanding.

A name, brand, or product term was changed

Use one or more of the following:

• Add the term to Custom vocabulary during project setup.
• Correct it in the source or translated transcript.
• Create a preferred entry in Translation Memory.
• Review and approve the Translation Memory suggestion for the relevant segment in the Advanced Translation editor.

Translation Memory changes are currently approved one segment at a time. They are not automatically applied to the entire project.

A translated segment sounds too fast or too slow

Look for the pacing warning above the segment.

Select Rephrase? to create wording that better fits the available duration.

Rephrasing is contextually aware and may adapt the wording for the target language and culture. Review the meaning after every rephrase.

LipDub does not overlap consecutive speech segments from the same speaker.

The end of a translation is cut off

Translated speech that extends beyond the end of the source video is cut off.

Resolve pacing warnings, shorten or rephrase the final segment, and confirm its timing before generating again.

The cloned voice does not sound close enough to the speaker

Voice cloning requires at least 10 seconds of clear source speech. Thirty seconds or more is recommended.

For better similarity:

• Use clean speech with minimal noise.
• Avoid music, overlapping voices, distortion, and echo.
• Make sure the source speaker is clearly audible.
• Try another available language style.

As a general guide:

• Style 1 often provides stronger voice similarity.
• Style 2 often provides a more native-sounding target-language accent.

Results vary by voice and language.

Voice cloning is unavailable or restricted

Confirm that the source contains enough clear speech and that you have permission to clone the voice and likeness.

LipDub may restrict recognized celebrity voices in certain cases. Contact support@lipdub.ai when a permitted use case is being blocked.

Captions contain incorrect text

Captions use the translated transcript.

Correct the translated text before generating the video again. Captions are burned into the video using white text with a black outline, and their appearance cannot currently be customized.

Audio and timing issues

Music or sound effects are missing from the result

Audio behavior differs by workflow.

Make sure the file you provide contains everything you expect to hear in workflows where the replacement audio becomes the final track.

Dialogue starts too early or too late

In a Dialogue Replacement project, open Review in Studio and reposition the clip on the Target Audio track.

In a Multi-speaker project, timing must be prepared inside each uploaded speaker file. Preserve silence from the start of the video so each line occurs at the correct timecode.

Dialogue Replacement audio is shorter than the video

When the provided Dialogue Replacement audio is shorter than the source video, the generated video is trimmed to match the audio length.

Use an audio file with the intended full duration when the complete source-video length must be retained.

Dialogue Replacement audio is longer than the video

Audio extending beyond the source video is cut off at the end.

Shorten the audio or use a longer source video.

A Multi-speaker actor stops talking before the video ends

When an actor’s uploaded audio ends, that person becomes silent and their lips are no longer animated.

Confirm that the audio file includes the required duration and all intentional silence.

If Select regions was enabled for a speaker, ensure there were no timecode input mistakes.

Two speakers talk at the same time unexpectedly

LipDub plays the speaker files according to their uploaded timecodes.

Review each file and remove unintended overlap. Each file should:

• Contain only one person’s dialogue
• Preserve silence
• Begin at the source-video timecode
• Place every line exactly where it should be heard

Audio was assigned to the wrong person

Return to Upload Audio & Generate and verify the uploaded file beside each actor.

Also confirm that the correct language or version tab is selected before generating again.

An off-screen person’s audio is still playing

This is expected. Uploaded Multi-speaker audio is included even when the actor is:

• Off-screen
• Facing away
• Not detected in the current shot
• Outside a selected lip-sync region

Edit the audio file itself when the dialogue should not be heard during that period.

Selected Regions did not limit the audio

Selected regions controls only visual lip-sync. It does not trim or mute the uploaded audio.

The actor’s complete file is included in the result, even outside the selected time ranges.

Edit the source audio when only part of the file should be heard.

A speaker set to Not at all is silent

This is expected.

Selecting Not at all disables lip-sync for that actor and makes that actor silent in the generated result.

Personalization issues

My CSV will not load

Confirm that the CSV:

• Contains a plain-text header in every column
• Includes an email column
• Contains a value in every cell
• Has no blank rows
• Has no blank values
• Uses no more than 10 personalized variables

The email value does not need to be a deliverable address, but the column must exist and contain data for every row. LipDub does not send anything to the addresses in the file.

The wrong data appears in a personalized phrase

Open Highlight Variables and confirm that each highlighted phrase is assigned to the correct CSV column.

The same column can be assigned to several parts of the transcript when a value appears more than once.

I cannot edit the transcript

Transcript editing is disabled after the first variable is created.

Correct the complete transcript before highlighting and assigning variables. When necessary, remove the existing variables, correct the transcript, and recreate the variable assignments.

A personalized name is pronounced incorrectly

Enter a phonetic spelling in the CSV value and generate a test result within a new campaign.

Review several rows before generating or distributing a large campaign.

One or more recipient videos are missing

Confirm that:

• Every CSV row contains data.
• Generate results was started.
• The affected row did not fail.
• You are viewing the correct campaign.
• The result list has finished processing.

Use Search videos to locate a specific recipient.

A Personalization link no longer works

Direct links remain active while the LipDub account and associated project remain active.

A link stops working when its project is deleted. Download important videos before deleting the project or closing the account.

Projects, results, and credit issues

I cannot find a generated video

Check:

• The correct project
• The correct Personalization campaign
• The correct language or style in a Translation project
• The correct dub in a Dialogue Replacement project
• The correct scene, source video, language tab, and result menu in a Multi-speaker project
• Additional pages in a paginated result list

A project may contain completed versions while other versions are still generating.

A project or source file has the wrong name

Projects, Multi-speaker scenes, and uploaded source files can be renamed.

Use clear names that identify the client, language, scene, version, or production stage.

I deleted a project and cannot find its files

Deleting a project removes its source media, training data, generated audio, and output videos.

Personalization links associated with the project also stop working.

Download all required files before deleting a project.

A generation or training action failed

Technical failures are normally marked as failed and refunded automatically.

1. Open Credit Usage History.
2. Confirm whether the consumed credits were returned.
3. Correct any input or configuration issue before retrying.
4. Contact support@lipdub.ai if the refund is missing or appears incorrect.

Credits are not refunded for user errors, including incorrect files, timing, actor assignments, or generation settings.

My credit usage was higher than expected

Open the account menu and select Credit usage history.

Review:

• Activity type
• Project or action description
• User email
• Credit amount
• Remaining balance
• Date

Select Export .csv when you need to review the activity outside LipDub.

Credit charges can vary by account, agreement, training model, output settings, and workflow.

Best-practice checklist

Before starting a final generation:

• Use the correct project type.
• Review the entire source video for additional faces.
• Confirm that each face is clear and unobstructed.
• Keep source and training footage visually consistent.
• Use footage of the subject actively speaking for actor training.
• Review every face detection before training a Multi-speaker actor.
• Use clean audio with no unintended speech or noise.
• Preserve exact timing and silence in Multi-speaker audio.
• Confirm every speaker-to-audio assignment.
• Review the source transcript before reviewing a translation.
• Resolve translation pacing warnings.
• Check Personalization variables against the CSV.
• Confirm the selected resolution and AI-video setting.
• Review the credit estimate.
• Watch the completed result from beginning to end before distribution.

Contact LipDub Support

Contact support@lipdub.ai when:

• One action has been running for more than two hours.
• Processing, training, or generation repeatedly fails.
• A technical failure was not automatically refunded.
• A completed result is missing.
• An expected file or link is unavailable.
• A permitted voice-cloning use case is being restricted.
• You need to discuss legacy SRT upload or limited AI-avatar testing.

Include the project name, project type, affected file or actor, and the action that is failing so the support team can locate the issue quickly.