Promptfluencer.
An AI image and video generation platform built on the idea that anyone prompting enough already has the core skill of an influencer. Shipped, reached twenty users, shut down when I lost discipline on scope.
the problem
People who spend time prompting LLMs develop a real skill: shaping outputs, iterating on tone, knowing what to ask for. That skill translates directly to generating the kind of content influencers produce, except most image and video generation tools are built for designers, not prompters. The idea was to build a platform that treated prompting as the core interaction and let anyone become their own influencer without needing a studio or a camera.
the approach
Built the original version as a monolith on Next.js and React, with Clerk handling auth, Supabase as the data layer, and Cloudflare for distribution and edge delivery. Shipped a working version and brought it to beta with twenty real users. Then kept adding features. What started as a focused prompting-to-content tool became a growing surface area of generation modes, social features, and export options, none of which the twenty users specifically asked for. Debugging the combined surface area became unsustainable, and rather than limp along, I shut it down.
the outcome
Shipped to twenty users before the scope caught up with the engineering. Pulled the plug as a deliberate choice: better to kill it cleanly than keep maintaining something I could not hold the shape of. Revival is in the pipeline and will be rebuilt with a much tighter scope, broken into separate services instead of a single monolith so each piece can be deployed, scaled, and debugged independently. The next version starts with 'what would the twenty original users actually use' and builds only that.
the lessons
The most expensive thing I did on Promptfluencer was keep adding features after the product already worked. Every new feature added a new bug surface, and I was optimizing for breadth instead of reliability. The Principles page rule 'define scope properly upfront' came directly out of this experience. Also learned that killing your own product is harder than building it, and sometimes more important.
stack
- Next.js ·
- React ·
- Clerk ·
- Supabase ·
- Cloudflare