Subscribe
Sign in
p/pixelvibeFind high-quality, curated AI-generated stock images
Start new thread
Kat Manalac

Generative.photos - 25k diverse & customizable free stock photos generated by AI

by•

We're sharing a free collection of 25k HiRes stock photos with customizable AI-synthesized people. For this collection we focused on increasing diversity in model appearance. Reach out if you want us to create a unique model for your visuals.

Add a comment

Replies

Best
Lisha Li
Maker
Hi Everyone! I’m the founder of Rosebud AI, a synthetic media company based in San Francisco. Generative.photos is a first step in our synthetic stock photo and API offering, which will eventually allow users to edit and fully synthesize visual content with an intuitive interface. We focused on bringing forth a way to diversify stock photo content since it was a need we heard voiced by stock photo users. All the faces in our 25k photo collection are not of real people. You can use this collection to serve up dynamic ads catered to individual users, and even give your consumers the power to choose an advertising model that they can relate too. We’re excited to hear about more uses cases for this platform and collection and look forward to releasing more functionalities for control and synthesis. The processing of this collection was completely automated, so there may be corner cases, and we’re happy to hear feedback to continue to improve it. FAQs Q: How’s it different from 100,000 Faces/ThisPersonDoesNotExist? A: Generative.Photos is not only generating the headshot of a fictitious person, but placing them in the context of a high resolution stock photo (body and background). Doing so in one click at high resolution required us to develop novel methods, in addition to generating faces we can control the gender/appearance/ethnicity of. Q: When will the self serve software come out? A: We are testing it internally. Please sign up on the website to be a beta tester so you can submit your own photos to edit and customize.
Tyler Lastovich
@lisha_li1 Hi Lisha, competition is always good to see! Sadly, I can't help but be disappointed with the strong similarity of your chosen domain & wording compared to generated.photos. 😕
Lisha Li
Maker
@tylerlastovich Hi Tyler, I parked www.generative.photos back in March of this year and have been working on GAN related creative tools since September 2018, so was planning a release of these tools for a while. Definitely saw (and upvoted 100k faces) because I believe there's a lot of whitespace in addressing different use cases of synthesizing photos. And as mentioned in FAQs above, we're doing different stuff in addition to using StyleGAN.
Tyler Lastovich
@lisha_li1 Thanks for the added info, I am sure you can understand where I was coming from. Completely agree on the whitespace, I even wrote an article here on PH about just that fact.
Crypto NewsWatch
@lisha_li1 Why put a watermark on every image? That's the dumbest move ever!
Lisha Li
Maker
@crypto_newswatch We will re-enable un-watermarked downloads once we implement user accounts. As a startup, offering a free resource was a trade off of immediate profit for feedback from users. We were thrilled with the inbound of user activity as well as social media reception of the collection. Sometimes we noticed posts of Rosebud’s pictures didn’t reference the collection and sometimes ended up indirectly/accidentally referencing other startups. To the best of our knowledge, our one shot face swap and synthesis is state of the art, so brand equity and attribution is a very important goal for the launch. To make it easier for those commenting on the tech but not using the photos to reference Rosebud, we made the decision to watermark the display photos for purposes of sharing. For those who want to use the collection of photos instead, we will re-enable no-watermark downloads once user accounts are created. It’s Thanksgiving for those of us based in the US, these changes will be pushed as soon as possible. What is your use case so we have more data to prioritize development for customers?. Was free very important or just not watermarked?
Renee Shah
@lisha_li1 - thanks for building this! This is an impressive application of GANs. I love that you're targeting stock photos with bodies and backgrounds (versus just headshots from 100,000 Faces and ThisPersonDoesNotExist). I've needed sophisticated, low-cost stock photos during so many moments in my career: for blogs, decks, websites, etc. Thank you!
Lisha Li
Maker
@renee_shah Thanks Renee! We're excited to be working on broader domains for synthesis. Definitely think this tech stack will make it easier for everyone to be their own creative director.
Ram
Looks cool, great work @lisha_li1 ... It would be great to see a wordpress plugin integration, everyone would love to use this... Also on a broader perspective, I think with AI's advancement, it could take several people out of business. This tool could easily replace models and everyone related to product shoot, the demo photos look very realistic...great work
Lisha Li
Maker
@ramkumarhq Thanks for your suggestion Ram! What type of photos are most useful for you on Wordpress?
Ram
@lisha_li1 -- For Wordpress integration, It would be good to have photos for landing pages. So if you could create models with transparent background, it would be cool, as it would be easy to use on HERO column of a Wordpress website. :) or better you could create a REST API and this way it could be integrated into other apps too. Cheers, I tried the app and i must say, so far 25000 AI Photos delivers the best AI face experience, it looks flawless.
Lisha Li
Maker
@ramkumarhq Wow, so thrilled with the praise! And will keep you posted on our development re these directions (REST API and background removal).
Erik Torenberg
Congrats on the launch Lisha! Tell us more about Rosebud - what's the big vision you what can we expect to see in the future, or at least your phase two? :-)
Lisha Li
Maker
@eriktorenberg Thanks for checking us out Erik! Providing customizable face synthesis is a first step in allowing users to alter every aspect of a photo, and eventually synthesize photos in their entirety. We believe generative adversarial methods, along with other tools from deep learning, will become indispensable for creating visual content (including video), and are taking steps to realize that future! We see it as AI reducing the friction from conception to creation, to help marketers, creatives and anyone become masterful at making visuals. Before we get there, with this launch we want to test what's the best intuitive interface for users as well as which types of photos are in most demand, which is why we released a limited interface of our self serve tool on top of the 25k collection. Traction has come from folks who want to test variants of photos fast, or are mindful to diversify their targeting, because they can generate them easily. Eager to learn from feedback on more use cases.
Dmitry Kislyuk
Exciting milestone Lisha and API release plans -- lots of potential directions to take this dataset further.
Lisha Li
Maker
@dmitrykislyuk Thanks Dmitry! We're looking forward to user feedback on the platform to develop more synthesize capabilities!
Sebastien Flury
Are you going to offer exclusively the possibility to create faces or also other kind of pictures?
Dzmitry Pletnikau
Maker
@sebastienflury Faces are the first step. We are working on technology for automated editing of every visual aspect of a person in any photo.
Lisha Li
Maker
@dzmitry_pletnikau @sebastienflury Our architectures are capable of synthesizing beyond people as well (landscape, objects, animals). Faces have a dramatic impact and we've heard many voice a need for more diverse stock photos, so we tackled this first. We want to collect feedback on where the most pressing use cases are before releasing synthesis tools for other categories.
Bohdan Shloser
It are pre-render photos, right?
Dzmitry Pletnikau
Maker
@bohdan_shloser Yes, our fully automated platform pre-generated this whole collection.
Lisha Li
Maker
@bohdan_shloser @dzmitry_pletnikau However if you are interested in rendering in real time your more photos, sign up for our mailing list on the landing page. We are planning a release of that in the near future as well. Rendering takes around a minute for us and can be optimized if there is a need for faster processing.
Julian Weisser
Ridiculously impressive. But it’s @lisha_li1 so I’d expect nothing less. Watch this space!
Lisha Li
Maker
@julianweisser Thanks for the support Julian! We are so stoked to be productizing all this research!
Malia Powers
Awesome! Not just headshots anymore, great step towards generative photo diversity.
Lisha Li
Maker
@malia_powers1 Thanks Malia!
Mike Dauber
This is so exciting to see -- awesome work Lisha!! I don't do my own content as much as I really should, but I love how you've applied GANs here. And the images really are incredible. Congratulations again!!
Lisha Li
Maker
@mdauber Thanks Mike!
Stephanie Sher
It's great to see a tool that makes it easier to generate diversity in online images. A few interesting use cases also come to mind, for example ads targeting by ethnicity or allowing online shoppers to preview how different color schemes look against their skin tones before purchase. Congrats Lisha & team, excited to see this scale!!
Lisha Li
Maker
@steph_sher Thanks so much Steph! Helping users and brands serve up more relevant and diverse images is very exciting to us!
Alex Conway
WOW this is super cool, well done Lisha & Dzimitry! I think even more than the need to “diversify” stock photo content, the “optimal” stock photo that a website should show you vs. me vs. @katmanalac vs. my grandmother is different and websites could show dynamically generated stock photos. It would be interesting to test but I suspect stock photos that look like you are more likely to convert than stock photos that look very unlike you. I think combining this with some of the incredible “try-on clothing” GAN work work will also be really interesting: https://arxiv.org/pdf/1911.07926... https://arxiv.org/pdf/1908.08847... I’d love to hear a bit more about your “novel” methods to include backgrounds in the generated photo if you don’t mind sharing (totally understand if it’s your secret sauce). I’ve been holding on to www.aistockphotos.com to work on something similar but I’m giving the state-of-the-art another year or so to iron out some of the kinks (particularly around asymmetries/artefacts near the teeth and eyes). It’s a super exciting space. Congrats on your launch - I just subscribed and look forward to beta testing! :)
Lisha Li
Maker
@katmanalac @alxcnwy Thanks Alex! I agree that dynamic serving of visuals is an exciting application of this, given the tech enables such fast and easy iterations. Linking variants with how much lift is generated will be the tricky thing to solve in order to capture value. We haven't published any of our methods yet, but may consider doing it in a future machine learning conference. So will keep you posted!
Jim Kleban
Cool product! While the stock photo/ad application seems benign to me, I wonder whether there are ethical concerns about working in the area of synthetic media. For instance, the technology could be re-purposed to create convincing fakes for the purpose of misinforming around elections (in the same way there have been concerns around text synthesis and deep fake videos.) I wonder if there are any thoughts about this part of the technology at Rosebud AI?
Lisha Li
Maker
@jim_kleban Absolutely, and thanks for asking. We take the ethical use of this tech very seriously. Being at the forefront of applying this tech gives us the opportunity to educate the public about its capabilities, and steer usage towards creative and positive use cases. Building a platform for it helps us moderate its usage towards these goals. Here's our Code of Ethics linked in the footer: https://www.generative.photos/co...
Patrick Lor
Very impressive! How many photos can you generate in a day?
Lisha Li
Maker
@patricklor Thanks for checking us out Pat! Our processing time per photo with one GPU is ~ 40 second. With multiple GPUs we can bring that down much lower, which is what we did to process the collection under a day. We're trying to learn whether speed is a priority for current customers, and thus have not optimized for speed yet, but will do if it is a high priority for users. For instance, if video is a more pressing use case, or having dynamic adjustments in seconds, we will make processing even faster.
Malaika MpressMe
pretty cool
Lisha Li
Maker
@malaika_mpressme Thanks Malaika!
None none
I wanted to make something like this for adverts but I wasn't smart enough good luck. I think you have a very large market for this product.
Lisha Li
Maker
@none_none4 Thank you, definitely excited to test out the dynamic ad use case among other things.
Sukhpal Saini
The quality of the generated faces is jaw-dropping. I would definitely use this to create dynamic images for blog thumbnails, and as hero images for landing pages.
Lisha Li
Maker
@sssaini Thanks Sukhpal! Thrilled they can be used!
Ghost Kitty
Comment Deleted
Lisha Li
Maker
@jakewesorick Hi Jake, we're adding search soon so you can access photos with men in them more easily (as well as other informative tags). Will notify people who sign up once that is up (aiming for in a couple of days). And uploading your own photos is a feature that we will release too.
Lisha Li
Maker
@jakewesorick @dzmitry_pletnikau just pushed out search functionality with tags at app.generative.photos !
Ghost Kitty
Comment Deleted
Lisha Li
Maker
@jakewesorick Hi Jake, the current ones are actually from Unsplash. :)
Вадим Деревянко
Interesting idea!