Show HN: Get a Professional Headshot in Minutes with AI (virtualface.app)
After playing with AI Avatars (like many of us I guess around here), I started to wonder if we could instead bring real value to people by producing affordable professional head-shots using a combination of Dreambooth and ControlNet.
Obviously it's only the beginning and there are still many imperfections, but the foundational tech behind this (Dreambooth and ControlNet) are only respectively 6 months and 1.5 month old, and already delivers pretty amazing results.
I came up with this little service "Virtual Face" and I'm looking for feedback if some of you are willing to try it (you can use the HUNTER50 coupon to get 50% off, can't make it free to try yet since the running costs are still non-negligible).
Cheers, Pierre
147 comments
[ 2.8 ms ] story [ 197 ms ] threadThat’s a hard ask if you’re a photophobic person.
I can probably lower this requirement to a minimum of 5, while mentioning to the user that the more pictures the merrier for the model
(this is due to how this model learns, maybe a future model will alleviate that)
So, if you really want to reproduce this on your own without using a service like ours, I would advise to: i) train a model with Astria, ii) download the model checkpoint from Astria, iii) install Automatic1111 UI with the ControlNet extension and now you will have a local setup that is very close to what we have, the only missing piece is that we have a custom model optimize for faces.
For example, Astria has its own trick to increase the number of headshot in frame, if you download the checkpoint and run it locally you will see that it doesn't handle well negative prompts (which, IMHO, is usually a side effect observed on fine-tuned style models with dreambooth). With this trick you can get up to 70% of headshots in frame but won't reach 100% without risking a loss of prior knowledge. This is my experience so far, maybe you have another opinion based on your experience.
If I was dedicated I think there are some free solutions for training a model on your own face out there but you might have an edge in convenience. Ten dollars seems too steep for that though.
i would also consider a trial pic for free tied to email.
It reads to me like you cherrypicked the two best examples and even those aren't great, whether that's true or not.
Edit: Woof, I didn't realize you could get bigger versions of the examples. They were mostly fine as thumbnails but blown up to ~500x500 the the eyes and mouth are rough.
That's the result of automated training. If you want good results you need to properly label your dataset (15 or more images of yourself) and then use a proper upscaling solution for it. Dreambooth-based images usually need some manual fine-tuning (and sometimes inpainting) to look presentable at higher resolutions. You can use latent upscaling for adding details but it'd require some manual supervision to weed out the bad generations. Using upscalers like Lanczos is usually foolproof.
Out of curiosity, which part of the image makes you say that?
Bots can do the same: “generate a photo of me holding my driver’s license, I need it for Tinder verification”.
Unfortunately, when technology like that becomes more common, people are definitely going to be tricked, and feel tricked. It's going to be awkward going on a date and the person looks nothing like what you expected... Awkward on both ends.
At some point... The online reality might become so fake, with disinformation generated by bots, generated images and videos, fake dating profiles (that are actually also just bots)... I wonder if it's going to spur some kind of reactionary movement. A movement to disconnect from the internet completely, or at least so socialize more in real life. Maybe some bars and pubs where you have to check in your cellphone at the door. Offline cafes.
[0] The new TikTok makeup filters in action: https://www.youtube.com/watch?v=tw2euJzOk60
So now even our profile pictures aren't really us, but just some pseudo-reality version of who we think we are. And I know I know, people will argue that makeup/airbrushing/photoshop/facetune has been going on forever, but at some point I feel like we cross the line where it's no longer "reality with some touchups", but instead it's "complete fantasy made to mimic reality".
I just feel like AI is shooting us head first down this state where reality and fantasy are evermore difficult to differentiate, and I don't like the implications.
If you compare reality today to reality at any other time, reality is pretty great.
But now, seeing what AI has on the horizon, I'm honestly just like "I need to go for a fucking walk, upon which I will throw my phone into the river."
I'm also almost done with the web. I remember listening to the dialup modem with excitement, browsing the web, downloading things and just being generally enthralled with everything the web had to offer. Now it feels like a mall with 3 chain stores in it. Being mostly a web dev, I'm in a kind of weird spot mentally about what I do with 80%+ of my time (i.e. program for the web, browse the web).
I'm going to build toy computers for people like us who just like computers (something like Nand-to-Tetris, eh? Or Ben Eater's great kits.) but I'm not kidding myself that that's going to make a lot of money (not like programming used to.)
i.e. I'm not worried about AI, I'm worried about how capitalism will weild it against me.
Perhaps this is the first time I’ve been properly challenged by a new wave of technology straight from go and perhaps a sign of age. However, I still have misgivings about some of the tech of the last decade and its lasting negative impacts.
It makes me want to step away from tech and do something closer to the physical world.
- honest signal (good-looking profile pics!)
- the signal starts becoming cheaper to create (AI filters)
- users of the signal look for subtler signs of authenticity (technical cues like blur around the edges, overly-softened features, lighting is wrong, or social cues like profile information doesn't match quality of image)
- signal becomes effortless to fake, information content is wiped out entirely
- retreat to authenticity / burn everything down and start over in a new game
There's tons of value to be had in that middle area, where people assume the honest signal but it becomes possible to fake. That's what TV, for example, is all about. Take regular people, add white teeth, makeup, and flattering angles, and they become 10x as compelling as Joe from down the street (without having to be 10x as good!).
And this is all armchair philosophy, parroting stuff from ACX and Ribbonfarm. The product concept itself is fantastic (if it works!), and I'm sure tons of people will pay for it.
I think most users will want to slip under the radar, and have it not be obvious that they're using the service.
Further… I think this is much more akin to those silly “make an anime avatar of yourself” apps than anything remotely resembling a headshot taken by a photographer.
For example, there is an example of a woman with medium length brown hair and the example “headshots” are… not good. It looks like her eye color is different in every picture, the black and white example looks kind of rotoscoped, and the example in the rotation immediately after the black and white has incredibly messed up eyes. One of her irises is all wonky and non-human looking. It’s a great example of why it’s important for a human to supervise and correct SD generated things like faces.
It all started when we started to argue with my family members on which avatars looked more like me. It made me realize that we were much more sensitive than I thought about our self-image. Me and my partner would pick different pictures in a set of 10 samples ^^ as if we had two slightly different perceptions of reality.
Now, I changed my mind slightly and tend to see these models as an another type of compression of information. Almost like a new censor of data.
If you want to experiment thatwith your photos I'm available on DM: twitter.com/_dulacp
Who cares if it's fantasy? We crossed that line _long_ ago with the examples you already brought up. The only difference is that this is new.
People have made their profile pictures a "pseudo-reality version" of who they think they are since the dawn of profile pictures IMHO.
Photo and video’s honeymoon is over.
However, rationally, it's not like humanity has had picture perfect representations of themselves for very long. For most of our evolution, we relied on paintings & sculptures, which it was up to the artist (or the commissioner) to decide on how 'real' they were, and they were almost always "complete fantasy made to mimic reality".
From this lens, the use of unedited real photos of you was the strange period of time, not this AI age we seem to be headed into.
Maybe that helps put your mind a little at ease, maybe it just confuses you more (definitely the latter for me)
It's a similar concern to the updated TikTok "makeup filter" that made the rounds recently, which basically is extremely difficult to detect as a filter. I thought especially poignant was a photographer who was saying she gets these beautiful women in to do portraits, and then when they look behind the camera to view their untouched portraits, they're aghast at how "ugly" they are, because the "filtered" version of themselves has started competing with the real version in their own head.
This shit just fucks with everyone's brain long term, in an unhelpful way, in my opinion.
This is just at my desk without being worried about if I've shaved since I can tell the AI (maybe not this one) to remove my facial stubble.
For most of our evolution we relied on a little thing known as "real life". The idea of the masses routinely representing their identity through imagery is firmly an artifact of the internet age. The UK didn't even have photos on driver's licenses until 1998.
[0]https://bogdannovykov.substack.com/p/death-of-reality
Oh, you also add choosing the clothes, the hair style and your posture/facial expression to the mix.
It (the genre of image we're talking about)'s all unnatural and meant to portray people in a light somebody wants to portray them in. In fact that's most of "photography" is.
https://www.youtube.com/watch?v=-lblzHvKhRc
It does look a little plastique, but it is also (deliberately an extreme case).
Definitely going to try it in the weekend!
Allowing some flexibility in the types of costumes might be tricky though (today it's easy because I only allow for a few outfits for "business" looks).
No wonder we have designers, developers, marketers, product managers etc. Proves to show that you can have the best/most amazing product in the world BUT if you can't communicate what it does, it will always fail.
The rest of 50% remains as per the poor mobile experience.
It would be great if you could adapt the output to be more realistic.
I'm personally not a fan of beautifying our pictures.
If you are interested in trying, we can talk in DM: twitter.com/_dulacp
> All images produced are released under the CreativeML Open RAIL-M license.
How does this work, legally?
Are the rights to the likenesses of the person in these output images being licensed by someone under the CreativeML Open RAIL-M license?
If so, to whom is it licensed? And by whom?
I guess I should rephrase to something like "The rights of the images are transferred to the user who purchased the service" or hire a lawyer to have it rephrased properly.
$10 for a "well, maybe the computer will generate something good enough that you like, sorry no refunds" is a big turnoff though.
Or a refund policy could seal the deal? :)
Ironically both of you launched on the same day.
> Questions about these Terms should be sent to the Company at [insert company email or contact information].
> Yes. All your images are yours and we are deleting them 7 days after your processing. We are keeping them 7 days in case you lose the link to download them (contact us in that case).
It's not about whether you delete the files after processing. What is your infra? Where do you store the files? How do you transmit the files there? Do you use proper encryption? How secure are your servers? How secure are your (if) cloud-related keys?
See OpenAI prompts leaking. Good example of what can go wrong.
What guarantees my photos (or even visual model) won't get leaked into another user's account?
Perhaps if they were dealing with highly sensitive information then it makes sense to make those kinds of disclosures, but for novelty headshot generator app? I'm not sure it makes sense to require that level of detail or expect it.
But to your point, I hate uploading anything private or personal anywhere for this exact reason. Even if you trust the company has good intentions, you also have to trust every employee at the company with privileged access, that their cyber security is adequate, and that all of their infrastructure providers can all be trusted. And given that you basically have to assume your information could be misused regardless of the company's intent.
Such a model can be used to generate nudes of you.
Billions of people do not upload selfies as well. Billions of people accept cookies and billions do not.
These days you don't even need to suss out the negative prompts, you can use a negative text embedding (bad-hands, easynegative) to get good quality images.
Dreambooth is practically ancient now. You don't need to lug around huge converged models trained on a few images and a few tags. You can download a much smaller LORA and include it in your prompt and it just werks.
Most programmers have never made a LORA and have never used Stable Diffusion. For 99.9% of people outside the programmer class, this is absolutely an "advanced algorithm."
Just because you understand what it's doing under the hood, and just because it can be abstracted into a handful of discrete steps, doesn't mean it isn't advanced.
Where do you think we are?
>this is absolutely an "advanced algorithm."
When someone puts that on his product's page, it's assumed that he developed said algorithm, or at least had a hand in it. Instead, here the real product is the pipeline and not the "algorithm". The product would do better if it was honest about what it did.
And like I mentioned, Dreambooth is pretty "old" now. The service could probably be much cheaper if the OP moved to using LORAs, and it would give better results, because it wouldn't clash with the tokens in the underlying model, and could be used with any model.
Guide to LORA:
https://imgur.com/a/mrTteIt
https://old.reddit.com/r/StableDiffusion/comments/11vw5k3/lo...
Though, I agree with you that the real product here is the pipeline. The parent comment is also correct though in that the majority of people aren't in a place where they can train their own models yet despite it being possible for them. This product is taking advantage of the current arbitrage opportunity.
GP isn't saying "this product would never work" (which is reminiscent of the classic Dropbox comment), GP is saying "this isn't 'advanced' and you can do it yourself" which is really helpful and exactly the sort of thing I love to see on HN.
Sure it’s "easy" to do it by hand, on a few persons, tweaking the parameters. Doing it at scale at good quality is still a challenge.