How to: rotoscoping with Deep Dream Generator

Limitations: 5 megapixels. I've mostly used it in 1080p with 30fps phone cameras. Step 1: turn the video into a procession of stills. There's plenty of free tools available on the Internet that can do this. I was mostly animating at a ratio of 5:1, but it's possible to do it frame by frame if you have the people power to do that. Step 2: use the Deep Style tool to paint over each image. I was improvising, but it would be possible to figure out which image sets create the desired results and then split the images up for a group of animators to paint over. Personally, my idea was to create a flickering technique by using subtly different images. I used a still from Kiri and Lou in the Guenever set, so that's a tip of the hat to Harry and Don's current project. But, yeah, it's made the Painting Vincent effect something that's far cheaper to produce. My suggestion is that someone should do a full music video in a similar style to my famous one-take thing using that technique. I was hoping that Weeded Out would be the project that would give me a budget large enough to achieve that painted effect, but yeah, they wanted something else. I was reading about the Quay brothers in the Guardian today. Seen a few shorts, and had no idea that they'd done features. That's exciting news. It's possible that I'd use that in the Plaything to test it out, but Runway ML's video effects are quite good as well, so possibly a mix of both. Plus Motion Array, which has a bunch of really solid acid FX. Step 3: make a new GIF video of all the images. It's probably best to blend it with the original video, but that's not what I did. - The main problem is that, as you can see, the Deep Style tool is rather primitive in terms of functionality and each image has to be manually inputted. There's no way to automate this because API's are specifically banned under the same terms and conditions that allow me to ride roughshod over copyright. Maybe someone more savvy can work it out, but it's beyond my ken.

My estimate is that it was possible to generate around 10-20 images per hour, but I haven't really timed it out: I was on the dole, so it wasn't like I was counting the hours. Amanda's first talkie was 23.976fps and 2:1. That's when I realised that the phone was smoother than the DSLR and I put that down to the 30fps rate plus stabilising effects that are inbuilt for most smartphone cameras. - Budget-wise, this is way more expensive than the cartoons, which are somewhere between 10-60 points, but mostly 20-3o points of computing power. 1 still with extra smooth AI enhance and 1.5 iterations is 51 points. The Professional plan is $39/month. This is 250 points per day, recharging at a rate of 18 per hour. This equals a maximum of 682 points per day. 13.37 stills. Roughly 16 seconds per month at 24fps. The Ultra plan is $99/month. This is 750 points per day, recharging at a rate of 60 per hour. This equals a maximum of 2,190 points per day. 42.94 stills. Roughly 53 seconds per month at 24fps. frame by frame is really too expensive imo, hence why the ratio was 5:1. 10 guys working relentlessly and maximising the recharge rates would come out to around 8.8 minutes per month. It'd cost $990 in computing power, plus a shitload of wages and salaries. Bear in mind this is perhaps the only tool that people seem to think that I pioneered, but there's lots of other VFX tools that are cheaper, though they lack the level of fidelity to the initial images. However, if you're super-rich, you could probably do a deal with them to allow some kind of automated use of the tool, or just invent your own version of Deep Style. - Runway ML was responsible for the acid fantasia in Election 2023. It's fine for a broad strokes image that's generically trippy, but it can't really render faces or those fine movements on the guitar. Of course, with masking and multiple layers, it might be possible to blend the two effects together. For me, it's amazing to think about the Air for a Winter Sunset. That's the first real experiment, and it took me over a year to move from moving stills to proper videos. This series is another expression of the same idea, and this is how I got started in 2021 while I was doing the EP and focusing on Shipwrecked on Islands: https://www.amandamichelinasprogressiveparty.com/post/political-caricatures-the-story-so-far

That Winter Sunset Ad was one that wasn't outside of New Zealand, so most people only became aware of it in the Latin-Punk New Wave album. That was part of the 2020 referendum campaign that I ran with Riddell Productions, alongside the Ethel Merman song from Annie Get Your Gun and Stephen's song Know. That commemorates an acid trip where I went up Mt. Vic to meet the dude in a parking lot, then sang Nature on the way back and jammed In My Room during the trip. Plus I saw No Man's Land (for the second time) during the peak, so it was quite memorable. - I've got all the images from my early experiments, but the gifs and videos are largely missing. I'm uploading them to Dropbox so people can see those and get started with those if they're interested in using the method to produce something. Here's some of them: https://www.dropbox.com/scl/fo/1jjbxdoh7uq2nt9tdpnt3/ANGlngyakfPJnYRdiK-DnRU?rlkey=0ltl1gqb405s3vzaor4sudqrj&st=53iznms1&dl=0 My chatter about IP is super-irritating, but this I'm giving out for free: I didn't invent rotoscoping -- this is just a particular process that I haven't used in quite a long time.