Show HN: I built an AI-powered professional headshot generator
quickheadshots.comHello HN,
I've created QuickHeadshots (https://www.quickheadshots.com), an AI-powered service that generates high-quality professional headshots from a few selfies.
As a developer, I found existing solutions lacking in consistency and quality, so I decided to build my own. The app uses fine-tuned stable diffusion models to create realistic, professional-looking headshots suitable for LinkedIn profiles, company websites, or any other professional use. This is the result of a lot of experimentation around this to get to a decent quality. The goal has been quality and faithfulness to uploaded images.
I'd love to get your feedback on the user experience, technical suggestions or any other aspects of the project. What do you think about the current state of AI-generated images for professional use? Are there any features you'd like to see added?
I'm also happy to answer any questions. If you're interested in trying it out, drop a line at drop a line to support-at-quickheadshots.com for a discount code.
Looking forward to your thoughts and suggestions!
Be ready for the PAYMENT page AFTER you have uploaded your photos.
I don't get this. So I upload photos of myself to the service to get... more photos of myself...?
What is the use-case where I would have 10+ good photos of myself to upload but needed 80 more photos?
(Also at first I thought this would be about some sort of ai cheating in fps games, but english isn't my first language)
So the 10+ photos that you upload could be from anywhere - eg: casual selfies or photos in your phone, could be in any situation, dress, background. The generated photos you get are headshots as if they were professionally shot.
There are so many of these fine-tuned SD models floating around. Can you share your learnings around the training?
I've been wanting to try out creating illustrations after using a good training set
A high quality training set is super important, clean sharp images with good lighting are ideal. I mention this to the customers in the app, but this is out of my control. If you are choosing your own data, then this may not be as much of a problem. Same goes with images being representative of them. Eg: not too much make up, not using old photos, not using sunglasses etc. As for training, regularization or class images help a lot. This shows the dreambooth model other images of people that are not your subject and helps it differentiate. Number of training epochs is something to experiment with. That number will change with the underlying base model you use, with the number of training images used, with learning rate etc. Both SD1.5 and SDXL models are good, have their pros and cons, but that's a longer discussion.
Thanks mate, will try out. Been waiting to put my GPU credits to use. Time to beat the inertia! :)
So this is "lorum ipsum" for headshots?
I would say the level of realism and similarity to the users' photos stands out in Quick Headshots vs the competition. But obviously I am biased :)