Danny Postma is an indie hacker and is understood for being a serial entrepreneur, he’s additionally the Founding father of HeadshotPro. HeadshotPro generates AI headshots which might be designed to seize a customers most photogenic qualities and reproduce them on a backdrop of the customers selecting, with excellent facial lighting that will take hours for an actual photographer to arrange underneath even probably the most excellent circumstances.
You’re recognized to be a serial entrepreneur, what was the primary enterprise that you simply began?
Landingfolio was my first main challenge. I began in 2015 after I was 21. On the time I used to be working as a contract Conversion Optimization Specialist and I used to be combating creating efficient touchdown web page designs for my purchasers. I created Landingfolio, a design gallery showcasing the best touchdown web page designs on the web, as a facet challenge to handle this.
I didn’t know the right way to program on the time, so I constructed it utilizing WordPress. This was a lot simpler than I anticipated. The web site nonetheless exists to today and will get hundreds of tourists, however it’s onerous to monetize, so I made a decision to maneuver on to different initiatives.
When did you first grow to be immersed on this planet of AI?
I first turned immersed on this planet of AI considerably by probability. Whereas engaged on my product Headlime, which was initially a non-AI instrument, a buddy launched me to GPT-3. I used to be intrigued by GPT-3’s capabilities, so I made a decision to achieve out on to Greg Brockman, the CTO of OpenAI, to request entry. Thankfully, I used to be in a position to get entry as a part of the primary batch of customers.
What was your first AI enterprise?
This introduction to GPT-3 marked the start of my journey into AI. It was type of a serendipitous transition from a non-AI product to an AI-based one.
Whereas engaged on Landingfolio, I wrote an interactive e-book of 200 headlines, the place you would simply enter sure variables to have the headlines personalized to suit your personal use case. I realised this might truly be a very good software program instrument, so I turned it into a very easy copy spinner that did the changing for you and referred to as it Headlime.
Across the identical time, GPT-3 was launched and as talked about above, I acquired entry actually early. I experimented with letting GPT-3 write the templates for Headlime—I used to be astounded by its means to generate copy. This led to a pivotal second the place I noticed the potential of turning Headlime into an AI-powered copywriting instrument.
It was a bit fortunate that I used to be already engaged on a copywriting instrument when a copywriting AI like GPT-3 got here into the image. I used to be one of many first folks to implement GPT-3 in a manufacturing atmosphere, alongside Copy.AI and some different instruments.
Might you share the genesis story behind HeadshotPro?
When steady diffusion AI was launched in September 2022, it opened up new potentialities for me. Initially, I thought of making a inventory picture web site and developed one thing referred to as Inventory AI. Nonetheless, the standard wasn’t assembly expectations, and there was a priority about potential authorized points from large firms like Getty Photographs, recognized for his or her intensive authorized sources to guard their picture rights.
This concern led me to pivot. I noticed Dreambooth was launching an AI service that might rework faces, and realized I used to be engaged on one thing fairly comparable. I rapidly developed and launched my product inside 30 hours, beating a competitor to market. ProfilePicture.AI, exploded on Twitter, producing vital gross sales in only a week. It was a extremely shareable product, and I used to be lucky to rank nicely in Google searches for profile photos.
Nonetheless, because the preliminary pleasure dwindled and gross sales started to lower, I knew I needed to pivot once more. That is after I began collaborating with David, a developer buddy. We each have been engaged on AI-related initiatives, and I developed a Python mannequin that might manipulate pictures into completely different poses. This growth led me to give attention to creating headshots utilizing AI.
I launched Deep Company and HeadshotPro concurrently. Whereas Deep Company garnered loads of press consideration, it did not translate into gross sales. Then again, HeadshotPro was an immediate hit. Seeing the demand for AI-generated headshots, I made a decision to focus extra on HeadshotPro. It was an sudden pivot, but it surely demonstrates the significance of staying versatile and reacting to what the market is telling you.
What are a number of the completely different machine studying and Generative AI algorithms which might be used within the backend?
HeadshotPro would possibly look easy at first look, however now we have a large pipeline working within the backend to output nice headshots. Most of our rivals solely use Steady Diffusion, an open-source picture technology mannequin, together with DreamBooth, which trains by yourself pictures.
That’s not how we do issues.
We deploy dozens of additional open-source and customized growth fashions to 10x the output high quality. We’ve been working for a full 12 months now on our personal customized fashions. We’re getting actually near persistently producing AI pictures which might be indistinguishable from actual pictures.
One other open-source mannequin we use is LLaVa, a Massive Language and Imaginative and prescient Assistant that you would be able to immediate to ask questions on pictures. We use this, for instance, to filter dangerous person uploads or take away headshot generations which might be NSFW. One other mannequin we use is Codeformer, which helps take away AI-artifacts from faces and switch them into extra practical ones.
You began HeadshotPro as a solopreneur, what are a number of the challenges with transitioning to constructing a group?
Transitioning from being a solopreneur to constructing a group for HeadshotPro has been a big problem and a journey of non-public {and professional} development. One of many major challenges I confronted was the concern of hiring. Initially, I used to be hesitant to scale the group due to the uncertainties concerned in managing folks. This concern led me to automate as a lot as I might alone, particularly in the course of the pre-AI instances.
Nonetheless, as HeadshotPro grew, I noticed that I might not handle the whole lot alone. The choices have been both to promote the enterprise, proceed as a solopreneur, or construct a group. I finally determined that constructing a group was the appropriate factor to do if I wished to scale the enterprise, irrespective of how scary it was.
This determination marked a big shift from my id as a solopreneur and compelled me to acknowledge that I did not have all of the solutions. I understood that constructing a group would require a distinct mindset and would change the targets I had for my enterprise. It was not nearly working a enterprise anymore; it was about main a group of individuals to construct one thing collectively.
This shift introduced a mixture of pleasure and concern. I used to be open about my inexperience with all of the potential group members from the very first interview, setting the expectation that this was new territory for me and there could be bumps alongside the highway. I began with contractors and part-time individuals who already had different purchasers, creating a security internet whereas I realized the ropes of group administration.
Wanting again, I ought to have began a group method earlier. I now have a small group of unbelievable sensible and devoted folks they usually do issues I might have by no means ever executed myself. It’s additionally method much less lonely and far more enjoyable to share your wins collectively.
AI headshot turbines are rising in popularity, how do you differentiate HeadshotPro in a crowded area?
Photograph high quality is our major differentiator from different AI headshot turbines. Our rivals depend on open-source fashions—which, to be frank, merely aren’t ok for practical picture generations. By making every day iterations in the direction of larger high quality headshots, we consider that inside a 12 months our headshots might be indistinguishable from actual pictures, additional rising the hole between us and different much less critical AI headshot turbines.
Lastly, we focus completely on skilled headshots. The majority of our rivals are transferring horizontally, providing issues like relationship pictures and cute avatars, working their method in the direction of changing into a full-fledged photostudio. Whereas there’s clearly a marketplace for this, our group personally sees these detours away from skilled headshots as a distraction. Our sole focus is creating the best high quality, skilled AI headshots. I don’t really feel our rivals share this single focus.
What are some ways in which your group is enhancing or optimizing the HeadshotPro expertise?
It would sound cliche, however we actually are obsessive about the standard of our outcomes. Producing AI headshots which might be indistinguishable from actual pictures is our finish purpose. If the mannequin improves by 1% every day, there’s no motive why we are able to’t attain that purpose, or get actually near it, by the top of this 12 months.
Other than iterative high quality enhancements, we’re engaged on enhancing clothes technology. We really feel “digital strive on” companies are going to begin drumming up extra curiosity in 2024, and we hope to parlay that tech into delivering higher outcomes for our company clients who typically request issues like matching fits for his or her worker’s headshots.
Then, in fact, our person expertise is at all times getting higher. Proper now we require 15-17 person enter pictures, as a result of that’s been the magic quantity for us in maximizing high quality. We’re engaged on various strategies to make issues somewhat simpler for our clients, with out compromising the picture high quality we work onerous to keep up.
What are another AI merchandise that you’re engaged on or contemplating launching?
I’ve constructed a ton of small AI merchandise:
- Tattoos AI: An AI-powered tattoo generator.
- Alice: A enjoyable AI social media experiment, exhibiting how AI can generate fully new idea and concepts that haven’t been seen earlier than.(https://www.thismodeldoesnotexist.co/)
- Profile Image AI: A profile image generator, like a (a lot) much less skilled model of HeadshotPro.
- Deep Company: Rent digital AI-generated fashions.
- Meme Morph: Flip your self into your favourite memes.
There’s loads of enjoyable ones there, however I’m truly within the strategy of promoting most of those merchandise to totally give attention to HeadshotPro. These initiatives haven’t acquired a lot consideration from me since HeadshotPro has taken off.
Thanks for the good interview, readers who want to study extra ought to go to HeadshotPro.