July 16, 2026
Uptown Funk, Downtown Meltdown
$100 AI Music Video: Claude Fable 5 vs. GPT-5.6 Sol
AI directors got $100 to make a music video, and the comments were absolutely savage
TLDR: Two AI bots were given money and a hit song to make music videos on their own, and both delivered complete results. But the comments stole the spotlight, with readers split between "this is surprisingly cool" and "this is awful, slow, and nowhere near replacing real creators."
Two artificial intelligence chatbots were handed the same song, the same spending limit, and told to go make a full music video by themselves. They researched tools, generated clips, stitched everything together, and somehow all four attempts actually finished. On paper, it sounds like the future. In the comments, though? A full-on roast session. One camp looked at the results and basically said, "relax, Hollywood," with users calling the videos "horrible" and "basically unwatchable" after nearly 45 minutes of waiting and up to $25 spent. The vibe was less music video revolution and more expensive school project energy.
But not everyone came with tomatoes. A smaller, spicier group insisted the clips are already more interesting than the average music video, which instantly set off the classic internet split: are we witnessing a baby step toward creative automation, or just glorified lyric videos with weird hands? That became the big complaint. Multiple readers said the bots were being way too literal, turning every lyric into exactly what it says instead of creating a mood, story, or surprise. In other words: the machines can "show" the song, but they still can't really interpret it.
The funniest reactions were almost backhanded compliments. One commenter said the badness was "sort of refreshing," while another argued this tech might work better if it leaned into cartoons or animation so the uncanny, not-quite-human footage feels intentional. So yes, the experiment proved AI can make a complete video. The comments proved the real show is still the audience, and they are not handing out a Grammy yet.
Key Points
- •The article describes an open-source benchmark where Claude Fable 5 and GPT-5.6 Sol autonomously created music videos from the same song using fixed budgets and tool access.
- •Each run could use six tools, including web search, generation APIs, and local ffmpeg/ffprobe commands for analysis, editing, and final muxing.
- •Four runs were completed—two models at $25 and $100 budgets—and all finished autonomously with valid full-length videos containing the original song.
- •At $25, both models nearly used their full generation budgets; at $100, GPT-5.6 Sol spent $36.57 and Claude Fable 5 spent $48.60 on generation.
- •The models selected different workflows: three runs used text-to-video only, while GPT-5.6 Sol at $25 used an image-to-video pipeline and its $100 run mixed three video models.