Video Audio
This Skill is responsible for sound design and cross-shot voice consistency in video production, but does not create a separate project-level audio/ directory.
Long-term reusable sound assets are stored under video/materials/, for example:
video/materials/voices/CHR01_voice.mdvideo/materials/voices/CHR01_reference.wav
Shot-specific dialogue, breaths, sound effects, or temporary sounds remain in the corresponding video/shots/<shot-id>/; full-film music, narration, mixing, and final sound engineering go into video/edit/.
For the complete sound layers and the boundaries between natively generated sounds and post-production sounds, see references/SOUND-LAYERS.md. When the video requires background music, brand/character motifs, animatic temp music, or AI-generated scoring, read references/MUSIC.md: temp and final music are clearly separated; music serves the story/dialogue/product and does not break character pacing just to hit beats. When a narrator, character inner voiceover, brand narration, or explanatory voice-over exists, read references/VOICEOVER.md: narration text belongs to video-script, voice identity/generation belongs to this Skill, and the final full-film narration goes into video/edit/.
1. First, Distinguish Three Types of Sounds
Character Long-Term Sounds
Requires stable age perception, gender presentation, vocal range, timbre, resonance, default speaking pace, volume, rhythm, and pronunciation/pause habits across shots.
Shot-Specific Sounds
Dialogue, breaths, sighs, shouts, footsteps, object collisions, and environmental changes that belong to a single shot; they are not copied to the public sound definition.
Film-Wide Sounds
Background music, full-film narration, overall sound design, and final mixing are managed by video-editing only during post-production.
2. Character Voice Design
When a character has dialogue across multiple shots, first create a voice definition, then generate or record specific lines. The character voice definition must clearly describe what the current project needs: age perception, basic vocal range, timbre, resonance, default tempo, default volume, speaking rhythm, pronunciation characteristics, variations under different emotions, and prohibited voice drift.
See the detailed template in references/VOICE-DESIGN.md.
3. Separate Voice Identity from Emotion
The same character can change reverb, environmental spatial feel, slight EQ, current emotion, and current speaking intensity across scenes and worlds, but the basic age perception, fundamental timbre, vocal range, pronunciation habits, and core rhythm should not change randomly.
If the generation tool provides speaker ID, voice reference, voice seed, speaker embedding, or equivalent stability mechanisms, prefer to reuse the same identity mechanism; the specific fields are determined by the current tool/model's official rules, and this Skill does not hard-code them.
4. Dialogue and The Visual
Dialogue length, visible lip-sync, multi-character speech, and cross-shot voice planning belong to video-script; if the current task requires this part, load that Skill as needed, or if missing, hand it back to akira-video to fill in, rather than copying script rules within this Skill. This Skill is only responsible for voice identity and the current voice performance.
Dialogue text continues to be owned by video-script; this Skill is only responsible for voice performance. When a video model that requires synchronized speech is used, compile dialogue, tone, pauses, and sound events into the generation prompt, but do not add lines that don't originally exist just to support the model's native sound capabilities.
5. Intentional Silence
If longer passages without dialogue are narrated by suspense, danger, music, visual climax, or action itself, the director's purpose for the silence should be clearly stated in the SHOT.md. If a character is actively operating for a long time without dialogue, reaction sounds, ambient sound, or music to carry the narrative, hand it back to video-shot to check if it's a dead section.
6. When External Voice Generation is Needed
Pass the character voice definition, current lines, reference audio, and return naming to video-generation, which organizes them according to the one-time generation package contract; this Skill does not create its own separate directory or temporary file rules.
7. Return for Review and Post-Production
- Voice identity drift, lip-sync, tone, pitch, and timing are reviewed by
video-review; - Shot-specific audio remains in the shot;
- Full-film music, mixing, subtitles, and final audio levels are handled by
video-editing.
Completion Criteria
Character voices can be stably recognized across shots; the responsibilities of shot sounds and full-film sounds are clear; no audio has been duplicated just for directory classification, and no tool-specific speaker parameters have been mistakenly written as general rules.