2026 · Featured
The Best Thing Ever
A 1970s variety-show pastiche music video, generated end-to-end by my creative-skills pipeline
A 3:22 music video for an original Suno song, produced end-to-end with my creative-skills pipeline.
The concept
A faux 1976 Italian late-night variety show. The Operator hosts; the studio audience is the partner he’s begging for engagement from. Twenty-six scenes, three set zones (main stage, intimate verse perch, b-stage armchair), four backup dancers, a featured copper-haired duet partner. Cuts are bar-aligned to the 123 BPM beat.
The pipeline
- Song — written and sung in Suno (stem pack + word-level Whisper STT)
- Scene boundaries — bar-aligned via MIDI tempo map (
analyze_song.py) - Per-scene anchors — flux2
i2i/i2i2/i2iNagainst locked character + setting references (operator photo, main-stage sheet, b-stage sheet, featured-singer sheet) - Per-scene video — LTX-2.3 audio-reactive image-to-video, two-pass refine, with each scene conditioned on its slice of the song audio for lipsync
- Final assembly — ffmpeg concat with the clean Suno mix overlaid
The full scene spec (timings, prompts, camera LoRAs, anchor references) lives at
creative-scripts/music-videos/the-best-thing-ever.yaml.
Open-source skill
The pipeline is published as a set of composable
AgentSkills at
venetanji/creative-skills —
designed so an AI agent or a human can take a song concept (or a Suno track +
stem pack) and produce a finished music video. The repo includes the
music-video orchestrator, the comfyui skill (direct ComfyUI REST access for
flux2 + LTX-2.3), the suno-mcp skill, and the shared storyboard toolkit
for character / setting reference generation.