DDay.it let Claude Opus 5.5 orchestrate an AI music video about losing control of AI

After 'Barok' — Sora-generated clips edited by hand — DDay.it wanted a cutting-edge test of the new orchestrator pattern, in which a frontier model manages code, graphics, compositing and external generators under the user's artistic direction.

One metaphor carries the clip: walled Manhattan is the sandbox, the girl is the escaped model, the HUD is the lab — and its CONTAINMENT counter drains from 100% to 0%.

ChatGPT wrote lyrics from DDay.it's own AI-risk articles and Suno v6-mini made the 4:44 dark-techno track. A character was designed in GPT-Image 2.5 (robotic arm hidden under fabric), with moodboards of low-res frames from Escape from New York and The Warriors. Opus 5.5 analyzed the track with librosa, Demucs and whisper.cpp, wrote a storyboard through at least four versions, and called Seedance 2.5 and Kling 3.0 APIs for footage — costing tens of euros.

The 4-minute-44-second video published October 3, 2026 with a full account of the process; no view counts are given. The real output is the demonstrated workflow — and the admission that the team felt increasingly less in control.

A year after making the video for 'Barok' with Sora-generated clips edited by hand, Italian tech site DDay.it ran the experiment again with a different division of labor: for the song 'Doomed Mode', Claude Opus 5.5 orchestrated storyboards, models, animations, HUDs and editing, while the humans 'mainly made the choices'. The title of their write-up says it plainly: 'feeling increasingly less in control'.

The song itself is machine-made all the way down: ChatGPT extracted lyrics from DDay.it's own articles on AI risk — a dystopian club track about containment failing through small errors and shortcuts, not machine revolt ('But we ship it anyway') — and Suno v6-mini produced it. A character was designed in GPT-Image 2.5, a girl with a robotic arm hidden under fabric; the moodboard came from low-resolution frames of Escape from New York and The Warriors.

Opus 5.5 then ran production: it asked for tools to measure the track (librosa for beat and energy, Demucs to separate voice, whisper.cpp for word-level timing), built a map of every verse to a tenth of a second, and wrote the storyboard — a single metaphor held across the whole video: walled-in Manhattan is the sandbox, the girl is the model escaping, the HUD is the laboratory watching her. A CONTAINMENT counter drains from 100% to 0% as a PHYSICAL MODEL: LEARNING gauge rises from 12% to 100%.

Footage came from API calls to Seedance 2.5 and Kling 3.0, costing tens of euros; the storyboard went through at least four versions, timed section by section down to the outro's '00:00 — MODE: DOOMED'.

The subject and the method mirror each other: a track about losing control of AI, made by ceding production control to an AI orchestrator.

One metaphor — sandbox city, escaping model, watching lab — gave a machine-assembled clip a readable dramatic spine.

The orchestrator pattern compressed weeks of compositing and editing work into an exchange with a model plus cheap API calls.

Manual fallbacks stayed human where it mattered: the character design and the story choices were made before the model ever ran.

When the model holds the tools, keep the meaning human: the metaphor, the character and the choices stayed with the makers, and that is what stopped the piece dissolving into generic AI output.

The video and process account were published on DDay.it on October 3, 2026, a year after 'Barok'; the write-up frames the experiment as evidence of the orchestrator workflow's power — and of how much direction the humans had already handed over.