Choose a good photo for your custom AI avatar

Start with face visibility

Choose an image with one face looking roughly toward the camera. The mouth, jaw and eyes should be visible. Hair across the lips, a hand near the chin or a large foreground object can make the animated result harder to judge and less useful.

Look at the portrait at the size people will see it on a phone. If the face is too small to read, crop or choose a closer image before generating. Avoid solving this by zooming into a tiny low-quality face; that cannot recover missing detail.

Frame for the final video

A vertical introduction benefits from a portrait that already fits a vertical frame. Leave some headroom and space below the chin. A shoulders-up view often gives the face enough presence while keeping the image simple. For a wider service explainer, compose the presenter with space where a title can sit without covering the face.

The Knowledge Base examples show their original source portraits alongside the finished videos. Compare them directly. The portrait controls the starting composition; it does not guarantee exact movement, perfect identity preservation or an identical frame in every generated result.

Use readable light and a quiet background

Even lighting makes the features easier to see. Heavy shadows across the mouth, extreme contrast or bright light directly behind the subject can distract from the speaker. Natural-looking light does not need to mean a plain white studio; a softly lit room can support an approachable presentation.

Keep the background purposeful. A simple office suits a service explanation, while a lived-in desk setting may suit a product concept. Avoid detailed signs and small text that could become distracting when animated. Check any visible product or brand marks before using the result publicly.

Compare your candidate images

Use this checklist before spending credits. These are selection guidelines, not a controlled benchmark claiming a measured success rate.

• Prefer: one clear face, visible mouth, moderate expression, suitable framing and enough source detail.
• Reconsider: profile angles, extreme expressions, strong blur, tiny faces, hidden lips or multiple people.
• For consistency: keep the same approved portrait when testing a new script, so you can compare the voice and animation more easily.
• For a new composition: create or select a genuinely suitable source image instead of stretching the existing one.

Create a portrait when you need a fictional presenter

The Knowledge Base presenters were generated specifically for these demonstrations using Mozify’s image tool. A useful prompt describes an adult subject, framing, expression, clothing, light and setting. Keep the prompt focused enough that you can evaluate whether the output meets the brief.

Choose one output, inspect the face and save the approved image to your workspace. Then use it in Talking Head with a finished narration. If the first portrait is unsuitable, revise the relevant part of the prompt instead of changing the whole visual direction. The example page publishes the actual prompt and portrait so you can see the relationship between brief and result.