How to Make a Talking Photo
Updated Sep 1, 2026 · 7 min read
What a talking photo is and where it is used
A talking photo is a short video in which a still portrait appears to speak: the lips move in sync with an audio track, the head makes small natural movements, and the eyes blink. The Talking Photo tool generates this from one image plus either an audio file or a typed script that it converts to speech.
Common uses:
- Explainer and product videos where you want a presenter without filming one.
- Personalised greetings for birthdays, holidays and customer onboarding.
- Historical and educational content: a portrait of a scientist reading their own letter.
- Brand mascots and cartoons that deliver announcements.
- Pets delivering the punchline in social posts.
- Language learning: the same face speaking a script in several languages.
Pricing is 20 credits per 10 seconds of speech, which is about $0.60 on a Pro plan. A 30-second clip is therefore 60 credits, so plan the script length before you generate.
Step 1: Pick a source photo that will animate well
The photo determines 80 % of the result. Choose or shoot one with these properties:
- Front-facing, or up to about 30 degrees of turn. Profiles cannot be lip-synced convincingly because one side of the mouth is hidden.
- Mouth closed or nearly closed, neutral expression. The model opens the mouth; it struggles to close a wide grin. A closed-mouth photo also avoids the teeth being redrawn.
- Face at least 500 px tall in the image. Small faces produce mushy mouths. If your only photo is small, upscale it 2x with the Image Upscaler first.
- Even lighting, no harsh shadows across the mouth.
- Nothing covering the lower face: no hands on chin, no microphone, no scarf.
- Sharp focus. A soft photo becomes a soft video. The Photo Enhancer can help a slightly soft source.
For pets, cartoons and paintings the same rules apply: the model needs to see a clear mouth region. A dog photographed straight on works; a dog in profile does not.
Portrait orientation (3:4 or 9:16) with the face in the upper-middle third is the most flexible framing for social platforms.
Step 2: Provide the voice: type a script or upload audio
You have two options in the Talking Photo tool.
Type a script. Enter the text and choose a voice: female warm, female bright, male calm, male energetic, or child. The built-in text-to-speech handles punctuation as pacing, so write the way you would speak: short sentences, commas for breaths, a full stop for a pause. Spell out numbers and acronyms the way they should be said ("twenty twenty-six", "A. P. I."). Roughly 25 to 30 words fill 10 seconds.
Upload audio. MP3 or WAV, up to 60 seconds. Use this for your own voice, a cloned voice from another service, a podcast excerpt or a song verse. Clean audio matters: background music and room echo confuse the phoneme detection and produce lazy lip movement. Record close to the microphone in a quiet room, trim silence from the start, and normalise the volume.
Which is better? Typed scripts give perfectly clean audio and consistent lip-sync. Uploaded audio gives authenticity and lets you match an existing brand voice. Many people draft with typed text to check the visual result, then re-run with the real recording.
Step 3: Generate and review
Click Generate. A 10-second clip takes about one to two minutes. The result is an MP4 with the audio embedded.
Review checklist:
- Sync. Play at normal speed and watch plosives (p, b, m): the lips should close on them. If they consistently lag, your audio probably had leading silence that was not trimmed.
- Teeth and tongue. These are generated. Slightly generic teeth are normal; visibly changing teeth between frames means the source mouth was open or blurred. Use a closed-mouth photo.
- Head movement. Small nods and tilts are expected. If the head wobbles unnaturally, the face was probably very close to the frame edge; re-crop with more headroom.
- Identity. Compare the last frame with the source. Some drift over 30 seconds is normal; if it is obvious at 10 seconds, the source photo was too low-resolution.
- Background. Static backgrounds stay perfectly still. A photo with a busy background is fine but the tool will not animate anything outside the head and shoulders.
If a clip is 90 % right, re-run with the same inputs; results vary a little between runs and the second attempt is often the keeper.
Avoiding the uncanny valley
Talking photos can look eerie when the motion is too perfect or too flat. A few production choices make a big difference:
- Keep clips short. Ten to twenty seconds holds attention; sixty seconds of a still photo talking becomes strange regardless of quality.
- Write conversationally. A script with natural pauses produces natural head movement. Monotone text produces a monotone face.
- Add a subtle zoom in your editor. A 3 to 5 % slow push-in over the clip hides the fact that the framing never changes.
- Cut away. Intersperse the talking photo with B-roll, product shots or captions rather than holding on the face.
- Match the voice to the face. A child voice on a corporate headshot, or an energetic male voice on a serene painting, breaks the illusion instantly.
- Do not over-enhance the source. Heavy skin smoothing before generation gives a plastic result.
For illustrated characters and mascots, lean into stylisation; audiences accept exaggerated mouth shapes on a cartoon far more readily than on a photo.
Consent, disclosure and legal use
Making a real person appear to say something they did not say is sensitive. The tool is for photos you own or have permission to use, and for speech the depicted person has consented to. Do not create talking photos of public figures, colleagues or strangers without consent, and never use one to impersonate someone or to mislead.
Several jurisdictions now require AI-generated likenesses to be labelled in advertising and political content. Adding a short "AI-generated" note in the caption or on screen costs nothing and keeps you on the right side of platform rules; TikTok, YouTube and Meta all have disclosure toggles for synthetic media.
Historical figures whose likeness is in the public domain, your own face, licensed stock models with a synthetic-media release, and illustrated characters you created are all safe ground.
Frequently asked questions
How long can the speech be?+
Up to 60 seconds per generation. Billing is 20 credits per 10 seconds, rounded up, so a 25-second clip costs 60 credits.
Can I use my own voice?+
Yes. Upload an MP3 or WAV recording and the lips will sync to it. Record in a quiet room and trim leading silence for the best result.
Does it work with pets and cartoons?+
Yes, as long as the mouth is clearly visible and roughly front-facing. Stylised characters usually look more natural than photorealistic humans.
What languages are supported for typed scripts?+
English, Chinese, Japanese, Korean, Spanish, French, German, Portuguese and more. Uploaded audio in any language will lip-sync.