Frequently Asked Questions

Q. Tell me how to use.

A. Please watch this tutorial video.

Q. How many hours of cloud usage rights should I purchase to create a deepfake?

A. Currently, 5 hours usage right ($10) is required for start making Deepfakes.
For face swaps only (by reusing the model trained by yourself), it is 1 hour ($2)
Because it is difficult to accurately predict the actual amount of time required, we calculate rough standard estimates.

Q. Do cloud usage rights expire?

A. Cloud usage rights expire after 6 months from the date of purchase.

Q. Is there a limit to the size and length of videos that can be uploaded?

A. The maximum video size that can be uploaded is 50MB and the length is 150 seconds. The format is compatible with MP4, WEBM, and MOV.
It is convenient to use a site like VideoSmaller to compress the video size.

Q. Can you handle 4K, HD video?

A. 4K, HD videos can be uploaded as long as they are within the above limits, but the resulting videos will be automatically resized to 1440px width.

Q. If the purchased cloud usage rights are insufficient, will the creation of the video stop midway through?

A. Once learning begins, creation of the video will not be only partially created due to the lack of hours. If creation fails due to system issues, the usage rights will be refunded. If the actual number of hours exceeds your usage rights, your cloud usage rights will be negative, and that time will be balanced at the time of your next purchase.

Q. Is it possible to create a Deepfakes video even if there are multiple people in the video?

A. Since the system attempts to recognize the faces everyone that appears in the video, if anyone other than the target person appears in the video, it will have a negative influence on the learning or swapping.
In the video, only the target person should appears with high precision and without blurring as much as possible.

Q. How can I improve the quality of Deepfakes videos?

1. Regarding the video you want to swap the face)
In the video, only the target person should appears with high precision, without blurring, moving and the subject is facing the front.
2. Regarding the video (or images) for training the model)
Use a video (or image) in which only the target person appears with high precision and without blurring. Also, it is better for there to be variations in the expressions and facial angles of person in the video. To improve the quality of Deepfakes, a video file should be longer than 20 sec, and the number of images should be higher than 300. You can also combine videos and images.

Although additional fees will be incurred, the quality will improve by repeating learning with the same (or different) material several times.

Q. What is Loss?

A. Loss is a numerical value representing the difference between the face given as training data and the face generated by AI. The smaller the value, the better quality of the reproduction.

Q. How can I retrain the model?

A. Proceed to the creation page from the button with 'Create a new video with this model' located on the bottom of the video detail page.
To retrain the model, please upload the same video/images as the first training (+ additional materials if you have) on all 3 steps. Do not upload deepfaked one.

Q. When a creation fails?

A. There are two main causes why Deepfakes creation fails.

1. The number of faces recognized by AI was too small.
Deepfakes creation needs to recognize faces from videos and images. But if AI could not detect enough learning data volume, creation will be interrupted.
Since blurred or ambiguous materials are excluded in the process of learning, please upload a lot of learning data with better reflection and try again.

2. Time out
If it takes more than a certain time to create, creation will be interrupted
Please change the video you are trying to swap to a shorter one, or reduce the size of the video and try again.