A GPT anyone can open, and a pipeline nobody sees.
Two halves of the same practice. MAYA is a custom GPT shipped publicly on the OpenAI store, which means the work is open for anyone to judge. The other half is generative production plumbing built inside a European fashion label, which nobody will ever see and which saved considerably more time.
A media mentor for creators.
MAYA is a custom GPT built as a content-creation co-pilot for YouTube creators: ideation, scripting, thumbnail copy, packaging and channel strategy in one conversation, instead of five browser tabs and a spreadsheet.
What makes it work is not the prompt engineering, which is the part everyone focuses on. It is what went in. MAYA is built on platform behaviour observed across four monetised channels and a decade of post-production, so its advice on packaging and pacing comes from watching real retention curves rather than from summarising the same public advice everybody else has already read.
It is public, which is the important part. Anyone can open it and decide for themselves whether it is any good. That is a materially different claim from a screenshot in a deck.
Plumbing, not magic.
A European label, kept private. The brief was not to make images — it was to shorten the distance between a brief and the first usable asset, so the in-house team could try more directions in the same week.
Season-driven design directions generated with the current image, video and language models, supplied as moodboards and reference cuts the in-house designers built from.
Ad-ready cinematic footage through Runway and ComfyUI, with shot ideation handled up front so the generation stage has something specific to aim at.
Upscaling and colour finishing in the same post stack as everything else, so generated footage cuts against filmed footage without announcing itself.
The internal team trained on prompt engineering, model selection and the loops that get a clean result on the second pass instead of the tenth.
Workflows, assistants, and the training to keep them.
Two things I say to everyone.
This field runs on overstatement, so it is worth being blunt about the two things that catch people out. Neither is a reason not to do the work, but both are reasons to go in with your eyes open.
Running costs are real, and they are yours. Expect somewhere between $45 and $165 a month in platform and model usage across two or three automations, billed to you directly by the providers. Anyone quoting a build price without mentioning this is hiding it, and you will find out in month two.
A model will not fix a broken process. If a workflow does not make sense when a human does it slowly on paper, automating it produces the wrong answer faster and with more confidence. That is why the audit exists and why it is the cheapest thing on the list — sometimes the honest outcome is that you need a process change rather than a build, and I would rather charge you for half a day than for something that cannot work.