Skip to main content
Use this path when you call Create Face with model_name: phoenix-4.5 and train_image_url. The image must be at a publicly accessible URL (for example a presigned S3 GET URL). You also attach a voice. See Choose a Model and Training Path if you are still deciding. For Phoenix-4 photos, see Phoenix-4 Image Requirements.
We recommend using the PAL Maker to upload an image. It runs the same checks the API uses, with real-time validation before training.
A zero-shot preview becomes usable in about a minute. The preview carries a visible watermark burned into the video. Tuning continues in the background for a few hours; the later tuned version does not carry the watermark.

​
Image requirements

Phoenix-4.5 works best as a chest-up portrait: mid-chest to just above the head, subject roughly centered with a little headroom, both shoulders in frame. Slightly wider (waist-up) or slightly closer both work. The face must stay fully visible and unobstructed. Full body is not supported on Phoenix-4.5; use Phoenix-4 for head-to-toe.
  • Formats: JPG or PNG
  • Minimum resolution: 512×512 pixels
  • Only one person visible in the image
  • Subject is an adult (18+)
  • Human-shaped subject: realistic, cartoon, anime, and Pixar-style faces are all fine. Animals, mascots, and other non-human figures are not
  • Head fully in frame, subject centered, looking straight at the camera
  • Upright, neutral pose (not leaning, lying down, or mid-gesture)
  • Face not blocked by hands, phones, masks, or heavy hair over the eyes and nose
  • Even lighting with minimal shadows across the face
  • Not a household-name public figure or an obvious copyrighted character
  • Do not send full-body, head-to-toe shots
  • Keep arms relaxed and lowered, mostly out of frame (no raised or prominent arms)
  • Glasses, jewelry, headphones, and hair in front of the shoulders are allowed as long as they do not cover the face. Shorter, crisp beards that leave the mouth and teeth visible work best - results may vary with dense or long beards.

​
Attach a voice

A photo has no audio, so nothing gets cloned the way it does on the video path. Send a voice on the same Create Face request - a stock voice or one you train yourself. See Voices for Image-Based Faces.

​
AI Image Fixer

If your uploaded image doesn’t fully meet these requirements, set auto_fix_training_image to true when calling Create Face. Tavus’s AI Image Fixer adjusts the uploaded image to fit. Some policy failures (for example under-18 or copyrighted characters) cannot be auto-fixed.
By using the image training API, you affirm that you have the rights to use the image you supply (for example likeness and publicity rights where applicable). Tavus may reject images that appear to depict unauthorized or impermissible subjects.