Skip to main content
Find answers to common questions about creating, training, and troubleshooting your Tavus faces.

Rights & Creation

Yes, as long as you have the rights to use that person’s likeness, voice, and training content. Uploads must follow Tavus’ Terms of Service and Acceptable Use Policy.
Yes - you may create faces using content you have the legal right and permission to use. This includes faces of public figures, celebrities, or deceased individuals. All training data and usage must comply with Tavus’ Terms of Service and Acceptable Use Policy, and assume full responsibility and liability for how the face is created and used.

Training & Errors

Use Choose a Model and Training Path. In short: Phoenix-4.5 is the newest portrait model and becomes usable quickly; Phoenix-4 is for full body or when you need no watermarked preview stage; Phoenix-3 is only for legacy edge cases (video only).
Phoenix-4.5 becomes usable early as a zero-shot preview, before tuning finishes. That preview has a visible watermark burned into the video itself. The later tuned version does not carry the watermark. Phoenix-4 does not have this preview stage. See Phoenix-4.5 Preview and Status.
On Phoenix-4.5, status completed means the preview is usable, not that tuning is done. Check finetune_status: started means the preview watermark is still on, completed means the unwatermarked version is ready, errored means tuning failed and you can keep using the preview with a tuning failed watermark. See Phoenix-4.5 Preview and Status.
Yes. Realistic, cartoon, anime, and Pixar-style humans are all supported. Animals, mascots, and other non-human figures are not. The checker looks for a human-shaped face, not photorealism.
Send "model_name": "phoenix-4.5" on Create Face.
Send "model_name": "phoenix-4" on Create Face. Pin it if you need full-body framing or specifically want Phoenix-4.
If your face fails, it’s usually due to issues with the training footage or the image used. When this happens, you’ll receive an email with an error message explaining the issue, and you can also view it in PAL Maker by hovering over the face’s thumbnail.

We recommend reviewing the training guidelines to address the issue, and if you still have questions, you can reach out to support@tavus.io for help.
When image-based training is rejected, you may see a machine-readable error code together with a short explanation. The table below lists common codes. Accessory, hair, and headphones errors apply to Phoenix-4. Crop and raised-arm style failures are the ones Phoenix-4.5 adds. Shared codes (quality, pose, under-18, celebrity, multiple faces) apply to both.For Phoenix-4.5 specifically, rejections often sound like: keep a chest-up or waist-up portrait; avoid full-body framing; avoid raised or prominent arms. See Phoenix-4.5 Image Requirements.
In most cases, your face isn’t actually stuck. Phoenix-4 training typically takes 3–4 hours, and the progress bar may stay at the same percentage while a specific step is processing. Phoenix-4.5 becomes usable when status is completed, then continues tuning in the background (finetune_status started) for a few hours. See Phoenix-4.5 Preview and Status.

We recommend allowing the full training time to complete before reaching out to support.

Credits & Billing

If your face errors out before or during training, it won’t consume any of your credits.However, if you delete the face immediately after the error occurs, our system may not have enough time to register it as a failed run. In that case, the credit might not be refunded correctly.\