Photos
The best photos to use for an AI rap video
A practical checklist for adult portraits, shared photos and pet references: useful crops, file sizes and the limits of generated identity.
Choose the studio mode that matches your references: two separate adult portraits, one photo of two adults, one solo adult portrait or one pet photo. Good references give the generator a less ambiguous subject to work from. They do not guarantee exact identity, expression, mouth movement, or clothing in the result.
Start with the face, not the background
A forward-facing or gently angled face is a useful starting point. Look for even light across both cheeks and enough detail to see the eyes and mouth. Avoid strong backlighting, deep shadows across the face, and motion blur. If you have a choice between a sharp phone photo and a heavily edited professional portrait, choose the one that represents the person most clearly.
The two people do not need to stand in the same place or have matching backgrounds. The studio combines two portrait references with your selected performance stage. For a shared photo of two adults, select the shared-photo mode; the two-portrait mode still needs two separate files.
Make a crop that leaves room around the face
Choose a head-and-shoulders view with the whole face visible. Leave some space above the head and around the sides instead of cutting through the forehead, chin, or ears. Avoid extreme close-ups and distant full-body photos where the face occupies only a few pixels.
Remove screenshot borders and unrelated people before selecting the file. Keep the person's natural proportions when resizing: stretching a portrait to fit a rectangle changes the reference. A slightly wider crop is usually a better choice than squeezing the face into a narrow strip.
Check these common problems
- Sunglasses, masks, or hands across the mouth: choose another photo with the important facial features visible.
- Beauty filters or exaggerated expressions: use a natural expression and a lighter edit so the reference is easier to interpret.
- Several people in the file: use one adult per file in two-portrait mode, or a clear photo of two adults in shared-photo mode. Crop unrelated people out of the reference.
- Tiny or blurred faces: return to the original photo; enlarging a small thumbnail does not restore missing detail.
- Unclear permission: ask the person before using their likeness, including for a surprise birthday video.
Solo mode uses one adult portrait. Pet mode uses one pet reference. None of these modes permits photos of minors. Pet performances and identity preservation are generated interpretations and need review in the result. Do not use a public figure's image to imply an endorsement or copy someone's identity without permission.
Check the file requirements
Each portrait must be JPG, PNG, or WebP and no larger than 10 MB. Both image dimensions must fall between 301 and 5,999 pixels, inclusive. The width divided by the height must be greater than 0.4 and less than 2.5. A normal portrait often fits; an extremely tall screenshot or wide panorama may not.
If a file is rejected, export a supported image from your photo editor, check its size and dimensions, and select it again. Renaming a file extension does not convert the image. Keep the originals somewhere you control so you can make a different crop later.
Use the preview for preparation
After adding the photos required by your mode, try the free stage layout in the studio. Check that you selected the right people and the frame you want. This preview does not animate faces or show final lip sync. Choose a 5-, 10- or 15-second request and check the displayed quote. A 15-second Standard 480p request uses 10 credits; HD 720p uses 22. Shorter options have their own quote. Rendering must be available before you can submit.
Inspect the finished MP4 before publishing. If a result looks wrong, compare the references first and use the troubleshooting guide. A sharper photo may help a later attempt, but another render uses credits and cannot guarantee a particular improvement.