GIOVISION Music Video Lip-Sync Generator · مهارة Open Skill عامة
GIOVISION Music Video Lip-Sync Generator
Turn a song and a reference photo into a finished, lip-synced music video — with the person's exact face locked in place across every shot, your camera directions actually followed, and built-in checkpoints so nothing runs away from you mid-project. What it does: Locks the character's face, features, and identity to your reference photo in every single shot — no retouching, no added or removed wrinkles/marks, no distortion, no random expressions you didn't ask for. If a shot drifts from the reference, it gets flagged and redone, not passed along. Follows your exact camera direction — angle, framing, movement. No unrequested zoom-ins or default close-ups; if you say wide shot, it stays wide. Keeps identity locked through motion — sit down, stand up, walk, turn — the face stays the same person from first frame to last. Syncs lip movement to your song, a vocal stem, or a generated voice track, with pitch/pace/emphasis locked per character so audio stays consistent across the whole sequence. Only pulls in a reference image for b-roll or cutaway shots when you specifically ask for it — no automatic insertions. Generates clips in small batches (3 at a time by default) and pauses for your review after each batch, so a bad clip gets caught and redone early instead of discovering it after generating 14 clips. Stitches the finished clips into one final cut, synced with your audio track. How it works: Provide a reference photo per character, your song or vocal track, a wardrobe/color anchor per character, and your target resolution. Give camera direction per shot if you have it (or the skill will default to a natural medium shot and flag that assumption for you to correct). It builds a shot-by-shot clip sheet, generates in checkpointed batches, verifies face and camera accuracy at every stage, and delivers a finished, stitched music video.
Music
428 20 120
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3495542/2026-09-03/ry2qmq8j.png
# Music Video Lip-Sync Generator A production skill for turning a song + character reference into a consistent, high-definition, lip-synced video sequence. It keeps one character's identity, exact facial features, and wardrobe locked across every shot, drives mouth movement from the audio track, follows the user's specific camera/shot instructions exactly, and stitches the clips into a finished cut. ## When to use - The user wants a music video / MV from a song, lyrics, or a character reference image. - The user wants a character's mouth synced to a song, vocal stem, or TTS line (lip sync). - The user wants the same character's face/body/wardrobe kept consistent across multiple scenes/shots. - The user wants to turn a song or audio into visuals, or produce cinematic HD video from a reference photo + audio track. ## Required inputs (gather before generating anything) - **Reference photo(s)** of each character — front-facing, neutral lighting, one clean "master" image per character to reuse across every shot. This master image is the single source of truth for that person's face — never regenerated, reinterpreted, or "improved." - **Audio track** — the finished song, vocal stem, or TTS line that drives the lip sync. If none is provided, generate one (see Step 2). - **Wardrobe/color anchor** — a fixed description of each character's outfit and a signature color used to visually disambiguate them shot-to-shot. - **Target resolution/aspect ratio** — e.g. 1080p / 16:9 for YouTube, 9:16 for Reels/TikTok/Shorts. - **Shot list / camera direction** — ask the user upfront whether they have specific camera angles, shot types, or actions in mind per clip (e.g. "start seated, then stand and walk," "wide shot, slow push-in," "over-the-shoulder"). If they don't have a full shot list yet, ask before defaulting to generic framing. ## FACE & IDENTITY LOCK — highest priority rule, overrides all stylistic defaults This is the single most important rule in this skill. The reference photo's face must be reproduced exactly as-is in every shot, with zero creative interpretation: - Do not alter, smooth, retouch, "enhance," age up or down, or reshape any facial feature. No added wrinkles, no removed wrinkles, no added or removed blemishes, marks, or skin texture changes of any kind. The face in the output must match the reference photo's actual skin, bone structure, and features — not a stylized or idealized version of it. - Do not introduce blur, distortion, warping, or soft-focus smoothing on the face at any point, including during motion (walking, turning, standing up, etc.). - Do not generate exaggerated, comedic, "funny," or unusual facial expressions or gestures unless the user's prompt explicitly asks for that expression. Default to natural, neutral-to-emotionally-appropriate expressions that match the song's mood — never invent a face the user didn't ask for. - When a prompt describes an action or pose change (e.g. "sitting down, then stands up and walks forward"), the reference image anchors identity through the entire motion — the face must stay the same specific person from first frame to last frame, not just at the start. - Every generation call that includes a character must pass that character's master reference image via `input_urls` (`ai2v` or `ref2v` as appropriate) — never generate a character shot from text description alone once a reference image exists for them. - If a generation result shows any drift in facial identity (different-looking person, altered features, unwanted expression, blur/distortion on the face), treat it as a failed generation: report it plainly and regenerate that shot rather than accepting or passing along a face-drifted result. ## CAMERA & PROMPT ADHERENCE — no default camera behavior - Follow the user's specified camera angle, shot type, framing, and camera movement exactly as written. Do not substitute a "default" cinematic choice (e.g. slow push-in, close-up) when the user requested something else (e.g