AI powered filters for images & video, runs locally on OS X

#5 Product of the DaySeptember 01, 2016



You need to become a Contributor to join the discussion - Find out how.
Ben SlaneyMaker@benedictslaney · Programmer, MacDaddy
Style is currently the only artistic style transfer app which runs locally for its neural net based filters (Prisma still cannot run its best filters locally despite the offline mode announcement), and which can work with high resolution images & videos. It is only just barely possible now to do this on a laptop sized machine, it's not quite feasible on phones at this point. Running locally has advantages. You can process as much data as you want for free, with the only cost involved being the amount of time you want to leave your machine processing the data for. For reference, I had to leave that Gangnam Style video example I posted in the description for about 10 hours overnight on the latest MacBook Pro. But that is 24 FPS * 1 minute 36 seconds = 2304 frames. 4 frames a minute doesn't seem so bad when you consider that each output frame needs to be individually painted, using the input frame only as a reference. With that said, this will be getting dramatically faster in the coming months. In terms of resource usage it's on par with most intense graphical software in terms of quantity of CPU & RAM required, along the same lines as typical video & imaging rendering software. Smaller images can almost be run in realtime, but with smaller images the output is of lower quality too because the neural net extracts the features from the input image to paint the output image, and the more pixels it has to work with to do that the better. Running it on large images can drastically improve the results. On that note, RAM requirements increase almost exponentially with image size. A good size for playing around with is an image of about 1MB. Then when you find images that look good with that size you might consider scaling up for producing more final versions. Personally I use for resizing the images. People clearly have a big appetite for neural nets for enhancing selfies. It is also my hope that this technology might also provide value to graphic artists and be of use for creating stunning new forms of art when put into the right hands. For example I made the app icon using the app itself. Considering that I'm personally terrible at art I feel that the result is really good, especially considering how easy it was to get. It also seems to be the only way to produce art on a computer that does not look algorithmically generated. I have created some videos, gifs, and images which are shown on this page with the hope of sparking interest in exploring exactly what kind of content can be created using neural nets such as this. I'm a programmer, I'm lousy at art, and even I can get results with this. But what I'm really look forwards to seeing is what happens when this technology reaches the hands of artistically creative people. Feel free to post the results of your experiments here in the comments. It can help us all discover what kind of material it works best on. ——— Direct download link for those that would prefer it instead of the mac app store: & here is a big gif example:
Christian Montoya@cm0nt0y4 · Maker
@benedictslaney @bentossell I love how you used your app logo / icon as an example. Nice work!
Robert Alexander@rjalex · Polymath
Very very cool! :-) Thanks for sharing.
Sanem Avcil@sanemavcil · Augmented & Virtual Reality Influencer
Adam Schwartz@adamfschwartz · Founder, Eager
This is truly amazing. Thank you @benedictslaney.
Ben SlaneyMaker@benedictslaney · Programmer, MacDaddy
@adamfschwartz thanks :)
Noah Henscheid@hirenoah · Solopreneur, Digital Consultant
Awesome to see this on a desktop. My biggest like-to-see for the next major version is perhaps the ability to choose a scratch disk. Nevertheless, it's very sweet. Peep a montage of my profile pic here:
Ben SlaneyMaker@benedictslaney · Programmer, MacDaddy
@noahhenscheid Thank you kindly :-)