Fully local music video: 52 AI clips via LTX 2.5 and ComfyUI in 4 hours
HMICsource · reddit · 2026-11-01
A Redditor built a rap music video about his GPU entirely with local pipelines.
- 13 stills (Gemini), 4 LTX 2.5 image-to-video prompts each, batched through ComfyUI's API on a 5090 — about 25s per 4s clip at 832x480, 52 clips total, with a queue script that skips already-rendered clips so renders can resume.
- Two woven tracks: a story line (trench-coat man in a neon city) and a literal line where lyrics happen on screen ("cut the tether" = astronaut cutting a cable).
- Every cut snapped to the kick drum — the grid was one beat off, landing cuts on the snare, and fixing that was the biggest upgrade; hard cuts at vocal entries via word timestamps plus energy checks.
- Beat from ACE-Step 1.5, vocals self-performed; 4 hours end to end, mostly pipeline building.
More from Multimodal
- Z Image Turbo on an 8GB VRAM Laptop: 40-60s per Image, a Beginner Is Impressed — Tonka-Jahari-Pizza · 2026-11-01
- "Found Footage" From a Hiker's Camera in 1997 — Actually AI-Generated — ati29 · 2026-11-01
- Beyond the Text Bottleneck: Why Native Speech-to-Speech Changes Voice AI — afigueiredo · 2026-11-01
- Slopus 0.4.0: open-source local AI studio adds music gen, MOSS, Krea 2, Fibo 2.0 — BittiAI · 2026-11-01
- Apple's karaoke model ships natively on macOS with on-the-fly voice dampening — banteg · 2026-11-01
- Halloween Cast List Includes 'Content Blocked' in All-AI Luma Short Film — mrjonfinger · 2026-11-01