Pika H3 Flex Dialogue, Lip Sync and Audio: How to Prompt Speech, Music and Sound Effects
PikaH3Flex EditorialHow to make Pika H3 Flex characters speak with matching lip sync, and how to prompt music, sound effects and ambience. Dialogue rules from the official prompt guide, templates and real examples.
Pika H3 Flex generates speech, lip movement, music and sound effects together with the picture, so lip sync comes from the prompt, not from a separate dubbing step. To make a character talk, say who speaks, put the exact words in quotation marks, describe the voice, and keep the speaker on screen in a medium shot or close-up. Then list the rest of the sound (ambience, effects, music) in a short “Sound:” line. Keep each line short enough to say comfortably in the clip: one or two sentences for 5–10 seconds.
A real Pika H3 Flex dialogue clip made on this site (text to video, 768p, 10 seconds):
Prompt: “Close-up of an elderly fisherman on a wooden dock at golden hour, he looks into the lens and says "The sea always gives back what it takes." Sound: his calm voice, gulls, gentle waves.” See the watch page.
Dialogue template
[Shot], [who] [does what], then says in a [voice description] voice:
"[exact words]." [Camera]. Sound: [ambience], [effects]. Music: [score or none].
Example:
Medium close-up of a young chef in a busy restaurant kitchen. She wipes her hands,
looks at the camera and says in a warm, confident voice: "Fresh pasta, every single
morning." The camera holds steady. Sound: sizzling pans, clinking plates, low
kitchen chatter. Music: none.
Dialogue rules from the official prompt guide
Pika’s prompt guidance in the Pika-H3-Flex GitHub repository sets these conventions for speech. They apply in plain-English prompts too:
- Keep speakers consistent. The guide assigns each real voice a stable ID (S1, S2) and keeps it for the whole clip. In plain prompts: name each speaker once (“the fisherman”, “the girl”) and keep using that name.
- Put identity and delivery outside the quotes. Describe the voice before the line (“says in a quiet, raspy voice”), and inside the quotes put only the words.
- Keep dialogue in its own language. Write the scene in English, but keep the spoken words in the language they should be spoken in.
- Voice-over needs closed lips. For narration, say the line is an off-screen voice-over and that the on-screen character’s lips stay closed. Otherwise the model may animate a mouth.
- On-screen text goes in double quotes. For signs or captions: a sign reading “OPEN”.
Sound and music
The official guidance splits audio into three layers. You can mirror that in a plain prompt:
| Layer | What to write | Example |
|---|---|---|
| Sounds in the scene | Speech, singing, and music the characters can hear | “a radio plays old jazz in the corner” |
| Overall soundscape | 1–4 sentences of ambience, physical and non-verbal sounds | “rain on the window, distant traffic” |
| Score (audience-only music) | 1–3 sentences: instruments, tempo, volume changes, or none | “soft piano, slowly building” |
Tips:
- Put the most important sound first. If the clip is about a voice, start the Sound line with the voice.
- Be specific. “Crisp case click, water drops, deep bass hit” works better than “cool sound effects”.
- Say “Music: none” for dialogue scenes when you don’t want a score fighting the voice.
- Ask for silence explicitly only when the whole clip should be silent.
A comedy example with laughter and a music sting
Prompt: “Bright 90s sitcom living room, a grandpa tries to take a selfie with a flip phone and the flash goes off in his eyes, static wide shot. Sound: studio audience laughter, camera shutter, sitcom music sting.” More on its watch page.
Getting better lip sync
- Frame the face. Close-ups and medium close-ups give the mouth enough pixels; wide shots blur lip detail.
- One speaker at a time. For two people, give each a short line and say who speaks first and who answers.
- Match line length to clip length. About two short sentences fit in 10 seconds; 15 seconds gives room for a short exchange.
- Keep the camera calm while someone talks. A slow push-in or a fixed camera helps; whip pans and fast orbits don’t.
- Use a start frame for a specific face. In image to video, upload a portrait as the start frame and prompt the line. The model begins from that face.
- Draft at 480p. Check timing and delivery cheaply, then re-render the best take at 768p or higher.
Pika’s human study rated Flex well for audio sync; physics-heavy and action scenes are its weaker area. If a line sounds rushed, shorten it before changing anything else.
Use cases
- UGC-style ads: a person holds the product and says one benefit line to camera. See product video ads.
- Explainers and hooks: a character delivers a one-sentence hook for TikTok, Reels or Shorts in 9:16.
- Short scenes: a two-line exchange, then cut several clips together in an editor for longer stories.
- Music moments: a band or singer performing, with the music described as sound in the scene.
FAQ
Can Pika H3 Flex do lip sync? Yes. Speech and lip movement are generated together with the video. Put the exact line in quotes, name the speaker and frame their face.
How do I make a Pika H3 Flex character talk? Write who speaks and how, then the words in quotes: he looks into the lens and says in a calm voice “The sea always gives back what it takes.”
Can Pika H3 Flex speak languages other than English? The official guidance says to write the scene in English and keep dialogue in its intended language, so put the words in the language you want spoken.
Does Pika H3 Flex generate music? Yes. Describe the score (instruments, tempo, volume) or write “Music: none” if you only want voice and ambience.
How do I make a voice-over without the character’s mouth moving? Say it is an off-screen voice-over and that the on-screen character’s lips stay closed.
How long can a dialogue line be? Keep it to what a person can say comfortably in the clip: about one or two short sentences for 5–10 seconds.