Make a Photo Talk With AI: Lip Sync Walkthrough, Free to Learn

You have a single photo and a short voice recording, and you want the face in that photo to say the words. Three ingredients decide whether it looks real: a clear, straight-on portrait with the lips together, a clean recording of the line saved as an MP3, and a lip-sync model that animates the face in time with that audio. On Spicy Studio, the model is Spicy Lip Sync, found under Lip Sync.

The recipe at a glance: - Photo: facing the lens, lips closed and uncovered, vertical 9:16 - Audio: one voice, nothing playing underneath, trimmed tight, MP3, 60 seconds or less - Quality: Standard (480p) while testing, Pro (720p) for the take you post - Prompt: optional; the playbook's default is “The camera remains still at all times.”

What it costs: Standard is 10 credits a second and Pro 18, charged per whole second with a 5-second minimum, so the shortest clip costs 50 credits. A new account's 25 free credits don't reach that, but they do pay for making the portrait itself.

The one rule: only animate your own face, a fictional character, or a real person who has said yes. Never use a talking photo to put words in someone's mouth without their consent, or to fool anyone.

The lip-sync workflow in the Spicy Studio playbook covers the same steps with before-and-after examples, and it's free with a Spicy Studio account.

A talking photo makes someone appear to say something. That's why the bar for consent sits higher here than for an ordinary AI image.

Faces you can use: your own; a fictional character you made, for example in the Image Lab; or a real adult who has clearly agreed to this particular use. A picture found online, a celebrity, an ex, a coworker or a customer is off limits without their permission.

Voices you can use: your own recording, a voice actor who agreed to it, or a voiceover you've licensed. Don't lift lines from podcasts, films, songs or other creators and put them in another face's mouth.

Spicy Studio's rules: the Terms prohibit realistic images of identifiable real people made without their consent, and any use for deception, fraud or malicious impersonation. You're also responsible for the rights and consents behind every file you upload.

Platform rules: as of October 2026, TikTok's community guidelines don't allow using a private person's likeness without consent, and they count AI that makes it look like someone said something they didn't as content that must be labeled. More on labels in the last section.

Lip sync works with the pixels it's given. It won't invent a mouth it can't see or rescue poor lighting, so run through this list before you spend a credit.

Facing the lens. Eyes toward the camera, head straight or only slightly turned. In profile, the model has half a mouth to animate.

Lips together, face relaxed. A neutral expression gives the sync a clean starting point. A wide grin makes the first syllables fight the pose.

Nothing near the mouth. Keep hands, microphones, straws and hair clear of the lips, and avoid a shadow falling across them.

Soft, even light. Soft daylight falling on the face from the front beats one hard lamp or a bright window behind the head.

Framed for the feed. Vertical 9:16, from the chest or waist up. If your prompt will ask for gestures, keep the hands in the shot; otherwise the model has to make them up.

A sharp file. In focus and not crushed by compression. Lip Sync takes JPEG, PNG, GIF or WebP images up to 50 MB.

Making the photo with AI. For a recurring character, follow the first-frame workflow: one face reference you reuse every time, a real front-facing photo you took (or have permission to use) as the base, and a one-line face-replace prompt in GPT Image 2, GPT Image 2.5, Nano Banana 2 or Nano Banana 2.1 at 9:16, asking for the mouth closed. Then sharpen the result with a Nano Banana 2 enhance and the playbook's “Enhance the image quality” prompt, which leaves the person unchanged. A GPT Image 2 frame at 1K medium quality is 10 credits and the enhance another 10.

The audio sets everything: the mouth shapes, the timing and, if you prompt for it, the energy of the body. A clean track improves the result more than any prompt can.

Record it yourself if you can. A phone voice memo in a quiet room with soft furnishings is enough. Keep the phone close, speak at an everyday pace, and save or convert the recording as an MP3, the only format Lip Sync accepts. Your own voice also settles who owns the audio.

Is the line inside one of your videos? An MP4-to-MP3 converter pulls the soundtrack out.

One voice, nothing underneath. Music, a TV or other people talking under the line can drag the mouth out of step. Leave music out of the file and add it in your editor afterward, beneath the finished clip.

Cut the dead air. Trim the silence before the first word and after the last. Shorter clips sync more accurately, according to the playbook, and because billing rounds up to the next whole second, silence costs credits too.

Over 60 seconds? Spicy Lip Sync accepts up to a minute of audio per clip. Split a longer script at natural pauses into parts of 60 seconds or less and run every part with the same photo, then stitch the clips together. Punching in slightly at each join disguises the cut, the same move the playbook uses to cover small glitches.

1. Open the tool. Go to Lip Sync and sign in; generating needs a free account and credits.

2. Add the face image and the MP3. The charge is based on the audio's length, rounded up to whole seconds.

3. Pick the quality. Standard renders 480p at 10 credits a second; Pro renders 720p at 18. Test on Standard and switch to Pro only for the take you'll publish.

4. Prompt, or leave it blank. For a calm, even read, blank works, or use the default camera sentence. For an emotional read, describe the body language (next section). In the lip-sync workflow, the Use in Lip Sync button opens the tool with the prompt already filled in.

5. Review, then caption. Watch the first and last second closely; that's where a rough start or a clipped ending shows. Then add captions: the playbook calls them essential for talking clips, since most people watch without sound.

A voice climbing with excitement over a face that barely moves looks wrong, and so does a whisper paired with big gestures. The optional prompt is where you match the body to what people will hear.

The playbook's emotional-delivery example shows the pattern: it names the feeling (excitement that builds), adds two physical actions (raised eyebrows, a gesture with one hand) and ends on the camera sentence. You can open it straight in the tool from the lip-sync workflow.

Four more to adapt (swap she for he or they):

Recipe or how-to tip: “She explains each step in a warm, steady voice, nodding on the key words and tapping the counter once with two fingers. The camera remains still at all times.”

Reply to a comment: “She reads the question with one eyebrow raised, then answers with a knowing half smile and a little shrug. The camera remains still at all times.”

Big announcement: “She starts calm, then leans in as her voice speeds up, ending with a quick laugh and both palms opening toward the lens. The camera remains still at all times.”

Late-night voice note: “She speaks slowly and quietly, as if trying not to wake anyone, blinking gently with her head tilted a little to one side. The camera remains still at all times.”

What keeps it believable: describe only what the audio supports, ask for gestures only when the hands are in the photo, stop at two or three actions, and end on the camera sentence, the way all the playbook's examples do.

Plenty of tools advertise a free talking photo with no sign-up and no limits. Here's exactly where the free part ends on Spicy Studio.

Free: this guide; the How to use playbook, with a free account; and 25 credits at sign-up, with no card needed.

What 25 credits buy: the portrait. A GPT Image 2 frame at 1K medium is 10 credits and a Nano Banana 2 enhance is 10, so a finished photo costs 20.

What they don't buy: the sync. Lip Sync charges every whole second of audio, rounded up, with a 5-second minimum, so even a 3-second line costs 50 credits on Standard and 90 on Pro.

Sample Standard costs (10 credits a second): - 9 seconds: 90 credits - 15 seconds: 150 credits - 30 seconds: 300 credits - 60 seconds: 600 credits, a full month of the €10 Basic plan On Pro (18 a second), 10 seconds is 180 credits and a full minute 1,080.

The cheapest way in: the €2 Spark top-up adds 90 credits that never expire. With the 25 free ones you have 115, enough for a finished photo (20) plus a 9-second Standard clip (90), with 5 left over. A 10-second clip and its photo would need 120. Plans and top-ups are listed on the pricing page.

No sign-up, unlimited? Reading needs neither. Generating needs a free account, and you pay for each second, so there's no unlimited tier.

Spicy Studio offers three ways to give a still image a voice. Pick based on what's already in hand.

An exact recording plus one photo: Lip Sync. The face repeats your recording exactly, for up to a minute per clip. Right for voiceovers, storytime and anything scripted to the syllable.

A script but no recording: Seedance 2.0 or 2.5. With audio on, Seedance creates the voices, lip movement and background sound in one pass from the lines in your prompt. Seedance 2.0 tops out at 15 seconds per clip and 2.5 at 30, and a 10-second Seedance 2.0 clip at 480p costs 70 credits. The sketch workflow (Spicy Pro) uses it for street interviews, POVs and podcast-style clips, and our Seedance Prompt Guide has a free template.

A movement to copy: Kling 3.0 Motion Control. Found in Character Swap, it drives your character with the motion and timing of a clip you filmed or have permission to use, and it can keep that clip's sound if the sound is yours. The face has to stay visible: the playbook's rule is that when it turns away or gets covered, lip sync is the better route. It costs 20 credits a second on Standard, with a 3-second minimum. Steps: the motion-control workflow and our Kling 3.0 Motion Control guide.

Can Seedance lip sync your own audio? Only loosely. In the Video Lab, both versions take reference audio (WAV or MP3), 15 seconds' worth on 2.0 and 30 on 2.5, as long as a reference image or video is attached too. There are two catches. Frames and references don't mix: if the photo goes in as the first frame, the reference audio isn't used, so put it in the reference-image slot instead. And ByteDance's Seedance 2.0 prompt guide treats reference audio as a guide to a voice's timbre, lists a generated voice that strays from the reference as a known issue, and suggests describing the voice in the prompt. So write the lines into the prompt as well, and expect a new performance in a similar voice rather than your file matched frame by frame. When every word has to match, use Lip Sync.

No speech at all? That calls for plain image-to-video instead; see our guide to turning an image into a video.

The first words look smeared. The photo probably shows an open mouth or teeth. Regenerate it with the lips together and the face relaxed.

The mouth lags or jumps. Listen for music, echo or a second voice under the line, and for long gaps. Clean the track, trim it, and break long scripts into shorter clips.

The face warps when the head moves. Start from a straighter, front-facing photo, and keep hair and hands away from the mouth.

The body doesn't match the voice. Add a body-language prompt with two or three actions that fit the energy of the audio.

Hands melt or appear from nowhere. Frame from the waist up so the hands are in the photo, or drop gestures from the prompt.

The camera drifts. End the prompt with the camera sentence.

It's accurate but flat. That's an editing job. The playbook's edit checklist: a caption on frame one, an emoji or two where the words land, a meme sound under the reaction, a fast punch-in on the punchline and no heavy effects.

These work because the voice carries the clip and the face only has to deliver it.

The 15-second tip. One useful thing said straight to camera: a cooking shortcut, a spreadsheet trick, a phrase in a language you're learning.

Reply to a comment. Put the viewer's question on screen and answer it in your own voice, using the comment prompt above.

Storytime in episodes. One story split into clips of up to a minute from the same photo, each ending on a small cliffhanger.

A character's voice note. A recurring AI persona speaking her own scripted lines to followers. How to Make an AI Influencer for Free covers building the character.

A small-business FAQ. The owner records the answers (opening hours, how to book, what's included) over their own photo. If an AI presenter speaks instead, say so on screen and never present it as a customer.

For hooks and pacing, see AI Influencer Reels: 6 Viral Formats. New to AI video? The Free AI Video Course for Beginners works through every playbook workflow.

Talking photos land exactly where platforms look hardest: realistic people saying things on camera.

Label it. TikTok requires a label on AI-generated content showing realistic-looking people or scenes. Its AIGC label or a clear caption, sticker or watermark all count, and unlabeled content may be removed, restricted or labeled by TikTok. YouTube asks creators to disclose realistic content that makes a real person appear to say or do something they didn't, using the disclosure setting in YouTube Studio when you upload.

Never pass it off as a real recording. Don't present a talking photo as genuine footage of the person shown, and don't make a real person appear to endorse a product or take a position they haven't.

Ads with AI presenters. An AI-generated person can present a product but can't be a customer: the FTC's rule on fake reviews covers reviews by people who don't exist, AI-generated ones included. Some places go further: New York requires advertisers to disclose synthetic performers.

Running a persona account? Is AI OFM Legal? covers platform rules for AI personas, and Do AI Videos Get Less Views? covers how labels affect reach.

When your photo and MP3 are ready, open the lip-sync workflow and press Use in Lip Sync.