Solo dev launch: AI video prompts without prompt engineering
Hey everyone đź‘‹
I just launched Movyo, a small AI video tool I built solo.
The idea came from my own frustration: every AI video tool assumes you know how to write great prompts, and the advice is always "learn prompt engineering". But prompt styles, formats, and rules change constantly with new models, and I kept getting inconsistent results.
So instead of trying to get better at prompts, I built a layer that handles that for you. Movyo takes very simple ideas and turns them into detailed, model-optimized prompts using structured templates and curated styles per model.
I started building this shortly after I realised the normal job life wasn't going to fit me anymore, and I shortly quit to work full time on this, and other similar projects. Now I made an actual launch and I'm getting my first few users.
I’d love feedback from other builders here:
– Does the problem resonate?
– Does this feel meaningfully different from typical “prompt enhancers”?
– Anything you’d do differently if you were building it?
Happy to share numbers, mistakes, or lessons as I go.
Nice work, shipping is hard.
Curious: how are you handling spam / bot abuse on this yet?
Manually for now. Didn't have any user that spammed accounts and abused the free credits or something yet.
Only thing spammy I got were some emails post-product-hunt asking me to enter partnerships and other stuff. Some seemed genuine but didn't help me so I responded back with a simple deny, others were most probably fake since gmail addresses didn't match what they said they were representing, so I just ignored
That tracks early on, manual + credits usually buys you a lot of time.
Do you think the first real pressure point will be automated account creation, or more subtle stuff like people scripting against specific prompts / niches once something takes off?
Also curious: if abuse does show up later, is your instinct to add friction immediately, or to keep pushing detection as far as possible before users ever see a gate?
The problem resonates hard. "Learn prompt engineering" is terrible advice when every model update changes the rules. Building a layer that abstracts the formatting away from the user is the right call.
To your second question, I think the key differentiator from typical prompt enhancers is whether you decompose the input into typed components or just rewrite the whole blob. Most enhancers take a sentence and expand it into a longer sentence. That helps a little. What helps a lot more is separating the intent into distinct pieces: what style, what motion, what framing, what constraints. Then each piece gets formatted the way the model actually wants it.
I've been building something similar for text LLMs called flompt (flompt.dev). It splits prompts into 12 typed blocks (role, objective, constraints, output format, examples, etc.) on a visual canvas, then compiles them into XML that Claude parses well. Open source: https://github.com/Nyrok/flompt
Same core thesis as yours: users should think about what they want, not how to format it for the model.
Great article — concise, well-researched, and full of practical insights. I especially appreciated the clear examples and actionable tips; they made complex ideas easy to apply. Thanks for sharing this valuable resource!
Highly recommended — I'll be revisiting this for reference
J'aimerai bien l'utiliser pour lancer mon application !
This comment was deleted 6 months ago