Feature

Talking Photo: Lip-Sync Video From One Portrait

Give Rigel a portrait and a script, and it generates a talking video with realistic lip-sync — the person in the photo appears to speak your words, with expression that tracks the delivery.

Updated · Rigel by Kata Technology

Talking-head video is the most reused format on social, and it normally needs a camera, a setup and a retake budget. A talking photo needs one portrait and a sentence you want said.

How it works

  1. Upload a portrait. Forward-facing, sharp, with the mouth and jaw clearly visible.
  2. Add your script. Type what should be said in the same step where you describe the scene.
  3. Generate. Rigel produces a 5–30 second clip with lip-sync, facial expression and head movement that track the delivery.

Describing the scene and adding a script before generating

The output saves to your camera roll in vertical or square format, so it goes straight to Reels, Shorts, TikTok or Stories with nothing in between.

Writing a script that lands

  • Write for the ear, not the page. Short clauses. Contractions. One idea per sentence.
  • Front-load the hook. Social viewers decide in the first second and a half; the strongest line should be the first one.
  • Stay inside the clip length. Roughly 15–70 words fits a 5–30 second generation at natural pace. Cramming more makes delivery rushed and lip-sync less forgiving.
  • Add delivery direction in the scene prompt. "Calm, low-energy, slight smile, talking to camera" changes the performance, not just the words.

What makes a good source portrait

The single biggest quality lever is the photo. Face reasonably large in the frame, even light on the face, mouth unobstructed, head roughly facing the camera. Profile shots, sunglasses, hands near the mouth, hard shadow across the jaw and low-resolution crops are the recurring causes of weak mouth shapes.

If your best source is a Live Photo rather than a flat still, live photo to video covers that path. The general still-to-clip workflow lives on the photo to video page.

What to use it for

  • Creator content — a week of talking-head posts from one photo session, without re-shooting.
  • Narration for characters — give a fictional character a voice across an episodic series.
  • Product and ad variants — test five different hooks with the same presenter and see which one holds attention.
  • Explainers — a short spoken answer attached to a face, which consistently outperforms text on social.

For longer pieces, generate several talking scenes and stitch them together as described in long AI video.

Honest limits

Lip-sync in this generation of models is good, not perfect. Fast speech, heavy accents, extreme angles and very short vowels are where it shows. Fine detail around teeth can soften under close crops. The practical workaround is the same as in real production: pick the better take, shorten the line, and shoot — in this case, source — a cleaner frame.

And the rule that isn't technical: only upload faces you have the right to use. A talking photo of someone who never agreed to say those words is the misuse case this feature has, and no prompt makes that acceptable.

Getting started

Rigel is free to download on iOS 16.6+ and Android, with starter credits included, so your first talking photo costs nothing. Credit packs and a subscription are available in-app; every generation shows its credit cost before you confirm.

Frequently asked questions

What do I need to make a photo talk?
One portrait where the face is clear and forward-facing, plus the words you want spoken. You add the script in the same step where you describe the scene, then generate.
How long can the script be?
Keep it to what fits in a 5–30 second clip — roughly 15 to 70 words at natural speaking pace. Longer messages work better split into several scenes.
Does it work in languages other than English?
Lip-sync language coverage is being expanded over time. English is the safest choice today; if your language produces weak mouth shapes, shorter sentences and slower pacing usually help.
Can I use a photo of someone else?
Only with their permission. Making someone appear to say things they never said is the obvious misuse of this feature, and it is on you to have consent for the face you upload.
Why does the mouth look wrong sometimes?
Usually the source photo. Profile angles, partly covered mouths, heavy shadow on the lower face, sunglasses and very small faces all reduce the information the model has to work with.