Building LipVox: Lessons from a SaaS Founder Launching AI Video Product with Zero Revenue

The idea wasn't mine. A friend called me when the AI video hype started and asked if I wanted to build something in that space with him. Before that, I was studying chemistry and teaching myself Java, and I had worked on a few projects that never got finished, none of them shipped. So I said yes. We spent the first weeks just looking at what already existed, and most of the tools out there were bad. The avatars looked wrong. The face would move, but the head stayed frozen, the hands turned into a blur when they came into the frame, and the backgrounds were obviously generated, smeared, and empty. They looked fake, like nobody had ever been in that room. You could tell it was AI right away, and then the video was useless anyway. So we just decided to do the same thing but better. That's it, really.

Flopped, social platforms don't want to show AI videos at all. Maybe one video out of fifteen gets into recommendations, and even that one doesn't bring you a user. We have been posting for months on YouTube, Instagram, TikTok, Facebook, and LinkedIn. The Instagram account has 36 posts and 2 followers. The YouTube channel has 35 videos, and the median is 13 views. We spent about $70 on ads on top of that, and out of all of this, we have exactly one person who registered, and she never made a video. I also sent 53 cold emails from our own domain to video studios, zero opens, and zero replies, not a single one. I think the letters never even arrived. We had never set up DKIM on the domain, so they probably went straight to spam and nobody saw them at all. Worked, nothing yet. Honestly. The only thing that got any attention was a short video where the avatar asks you to guess if the person is real or AI. It got 1600 views when our median is 13. It still brought no users.

We don't have enough customers to have complaints yet. The closest thing to a complaint is my co-founders' and mine. Our first avatars looked stiff and a bit creepy. I kept generating them and rejecting them, and at first, I couldn't even say what was wrong with them. It turned out the problem was in the source photo and not in the model. A neutral face looking straight into the camera with a closed mouth gives the engine nothing to work with, so you get a frozen head with a moving mouth. Once the photo had a slight head turn, weight on one shoulder, and a small smile, the same engine produced something that looked alive. The voice had the same problem. I wrote a prompt asking for even energy from the first word to the last, and that is exactly what it gave me - a flat, bored voice. The face moves from the accents in the audio, so a flat voice also means a still face. We fixed both of them by rewriting prompts, not by changing anything in the product itself.

Everyone thinks this market is about making video cheaper. It's not. A realtor or a coach films themselves on a phone for free. Cheaper than free doesn't exist, so that argument doesn't work on them at all. The only people who pay are the ones who already pay somebody else to be on camera. Agencies, training studios, companies that book a presenter for a day. I didn't understand this for a long time. I spent about two weeks, maybe a bit more, writing to realtors, mortgage brokers, and immigration lawyers because they post a lot of video. 45 contacts on Instagram and nothing came back. The message wasn't the problem; the whole list was just wrong. The second thing people mix up is UGC. A UGC ad is a person holding your product in their kitchen and using it; an avatar is just a head that talks. They get sold as the same thing, but they are not the same at all. If you promise one and deliver the other, nobody answers you again.

The next months are mostly unglamorous work. Getting a reputation and the first real clients, and setting up partnerships with people who already have an audience that needs this. On the product side, features that take the thinking out of making an ad so you don't need to know what a hook is or how long a script should be. Better output quality as well. And we are testing other models under the hood because if a video costs us less to generate, we can drop the price for everyone.

From scratch. I had no experience in video at all and I had never shipped a product. I was studying chemistry and teaching myself Java for about a year, then picked up some other languages. I worked on a few big projects with other people, but none of them got finished, so I never saw anything go out to real users. When my friend called me about LipVox, I could write code and that was it. I didn't know how software gets sold, I didn’t know who buys it, and I had no contacts in this industry at all. Everything I know about this market now I learned in the last few months, mostly by getting it wrong first.

I was lazy about selling. I thought that if the product was good enough, people would find it themselves, so I spent months on the product and almost nothing on getting users. Nobody came. You have to sell it every day from the first day, even when there isn't much to sell yet. I also thought social platforms would push AI video the way they push everything else. They don't, and building a plan on top of that assumption cost us months. And nobody warns you how much a company costs before it earns anything. Registration, accounting, the payment system, taxes for a US LLC that made zero dollars. That money leaves every month, whether you have users or not, and it doesn't wait for you to figure things out.