Automatic Music Video Maker · Open Skill 공개 스킬
Automatic Music Video Maker
Automatic Music Video Maker은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.
Music
194 17 95
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3615863/2026-08-25/7pbnk4ek.jpg
# Automatic Music Video Maker ## When to use - The user uploads a song / audio track and wants a complete music video made automatically (no manual editing). - The user wants a 16:9, ~4-minute (240-second) video built from a 6×4 storyboard (24 shots × 10s = 240s). - Default output is pure visual / cinematic storytelling (no singer lip-sync) unless the user explicitly asks for a singer. ## Inputs - Required: an uploaded song/audio file URL (mp3/wav/etc.). Ask for it first; do not proceed without it. - Optional: a visual style chosen from the Style Options below. If the user doesn't care, default to "Cinematic Film". - Optional: reference images (character / scene / props) the user uploads, to keep those elements consistent across all shots. - Fixed spec: 16:9 aspect ratio, 24 shots, 10 seconds per shot, ~240s total. ## Style options (present to the user and let them pick one) 1. **Cinematic Film** — anamorphic widescreen, film grain, natural light, shallow depth of field 2. **Anime** — hand-drawn style, vibrant colors, dramatic lighting 3. **Cyberpunk Neon** — night city, neon signs, rain reflections, high contrast 4. **Dreamy Ethereal** — soft focus, pastel palette, floating particles, surreal 5. **Retro 80s** — VHS grain, synthwave palette, chrome, sunsets 6. **Dark Fantasy** — epic landscapes, fog, dramatic skies, muted tones 7. **Minimalist Abstract** — geometric shapes, clean gradients, slow elegant motion 8. **Documentary Nature** — wildlife, landscapes, natural light, realistic ## Steps ### 1. Gather input, style, and references 1. Ask the user to upload the song (an audio file, ideally ~4 minutes). If no song is provided, stop and ask for it — the skill requires an uploaded track. 2. Present the 8 style options above and ask the user to choose one (or default to "Cinematic Film"). 3. Ask whether they want to provide reference images for characters / scenes / props. If yes, ask them to upload and label each one (e.g. "character: main girl", "scene: neon alley", "prop: red car"). Also use any reference images already uploaded this turn, and any bound as skill preset assets. Keep a labeled list; if more than 14 refs, prioritize character → scene → prop. 4. Confirm the fixed spec: 16:9, 24 shots × 10s = 240s. If the track is longer than 240s, the video will cover the first 240s. ### 2. Analyze the song 5. Call `transcribe_audio(audio_url=<song>)` to get the track's structure (timestamped segments / SRT and lyrics if any). Derive the song's duration from the last timestamp. Use the structure to plan the emotional arc: intro (shots 1–2), build (3–6), first chorus (7–10), verse (11–14), second chorus (15–18), bridge (19–20), climax (21–22), outro (23–24). If the track is instrumental or transcription returns little text, base the arc on the timecode segments and overall mood. ### 3. Write the storyboard (6×4 = 24 shots) 6. Write a shot list of exactly 24 shots: for shot i in 1..24, timecode = ((i-1)*10)s → (i*10)s. Present it as a 6×4 grid (6 rows of 4 shots). Each entry contains: scene description, camera move (e.g. slow push-in, pan, tilt, orbit, drift), mood, and visual prompt keywords. Keep one coherent narrative/visual arc across the whole board, aligned with the song's structure from step 2. If reference images were provided, feature the referenced character / scene / prop in the shots where they belong and keep their appearance consistent. Keep all prompts pure visual: no text, no watermarks, no singer lip-sync unless the user explicitly requested a singer. ### 4. Generate 24 keyframes (4K stills) 7. For each shot i (1..24), call `generate_image` with: - If reference images were provided: `model='i2i.hd'` (multi-reference, up to 14 images) and `input_urls=[the reference images relevant to this shot]` — the prompt then describes the shot featuring those referenced character / scene / prop elements. - Otherwise: `model='t2i.hd'` and no `input_urls`. - `prompt` = full 16:9 scene prompt including the chosen