People use it for birthday messages, family history projects, museum and classroom storytelling, memes and greeting cards. Here is how to get a result that feels alive rather than uncanny.
Which pictures make the best AI talking photo
The face should be clearly visible, roughly facing the camera and not covered by hands, hair or glasses glare. Resolution matters less than clarity: a sharp 800-pixel scan beats a blurry 4000-pixel phone photo. Old black-and-white prints work well; small scratches are fine, but heavy damage across the mouth will show.
Wider crops that include shoulders and hands give the model room for gestures. Tight passport-style crops still talk convincingly but move less.
Choosing the voice
You can record yourself directly in the browser, upload a voice memo, or use an audio file from any text-to-speech tool. For family projects, a real recording of a relative telling a story is far more moving than synthetic narration. Keep background noise low and leave out music while recording — you can add a soundtrack afterwards in any editor.
Always make sure you have permission to use both the photo and the voice, especially when the person is still alive or the picture belongs to someone else.
Ideas people actually make
A few formats that consistently work:
- A grandparent’s wedding photo reading a short letter for an anniversary.
- A historical portrait introducing itself for a school presentation.
- A pet photo “answering” questions in a funny voice for social media.
- A painting singing a few bars for a museum or gallery post.
- A team photo where each person says one line for a farewell video.
Three steps to your first clip
It takes about a minute of your time; the render runs in the background.
- 1Upload the pictureDrop the photo into the studio. We check face visibility and size instantly.
- 2Add the voiceRecord a few sentences or upload an MP3/WAV between 4 seconds and 5 minutes.
- 3Render and sharePreview for free, then render and download a 720p MP4 without watermark.
Prepare old photos first
Scanned prints benefit from a quick clean-up before animating. Crop to the person, straighten the scan, and remove large dust spots or tears near the face with any photo editor. Gentle sharpening helps; aggressive AI upscaling can invent features that then move strangely.
If a photo is very small, scan it at a higher resolution rather than enlarging a low-resolution file. The model needs at least 300 pixels on the shortest side, and more detail around the eyes and mouth produces more natural motion.
FAQ
Does an AI talking photo work with black-and-white pictures?
Yes. Old monochrome scans animate well as long as the face is clear and the mouth isn’t damaged.
Can I use a painting or drawing?
Yes. Portraits in paint, pencil or digital illustration work when facial features are recognisable.
Can the photo sing?
Yes. Upload a song instead of speech and the face, breath and head move with the music.
How long can the video be?
From 4 seconds up to 5 minutes, matching your audio.
Will the result look real?
With a clear photo and clean audio, motion is natural and expressive. Blurry or angled faces look less convincing.
Is it okay to animate someone else’s photo?
Only with their permission or the rights holder’s. Our terms forbid using someone’s likeness to deceive.
Can I make a photo talk in another language?
Yes. Lip movement follows the audio, so record or upload speech in any language.
Can I make several photos talk in one video?
Render each photo separately and join the clips in an editor, or use a single group photo for a multi-person scene.