DAZ Animation tests

I'm not a big fan of looping 5 second animations, but since everyone else is doing them I figured I'd see if it was workflow I could work within. I started with creating a 5 second animation and rendering it out at 15 FPS. I also used a very low iteration count (so the number of times it gets to try before the image is considered 'done'. My normal renders use 3000 iterations; this animation clip uses 120).

https://vimeo.com/415834861

That looks really.. cheap and nasty. I could of course make a longer animation, and run it for longer, with a higher iteration count, but it would end up using my PC as a space heater. Time to bring out the AI.


https://vimeo.com/415831779

The very first step was to run the whole thing through a de-noiser to remove the fire-flies left over from my low iteration count. After processing it, we end up with these two clips; you can see the 'noise' in the image on the left.

https://vimeo.com/415832019

Our second problem is the low framerate; at 15 FPS it's just too jerky to be fun jerking off to. So second step is to run the whole thing through another neural net, this one called DAIN. Where the intel network specializes in removing noise and fireflies, DAIN focuses on interpolating frames. So I gave it 15 FPS, and asked it to make a best guess on what was missing to bring the end result up to 30 FPS. Here they are back to back. It's not quite apples to apples; my dinky graphics card couldn't handle the video at full resolution, so it ended up having to drop it to 1/2 resolution (1152x768)  to fit it in VRAM.

https://vimeo.com/415839217

Finally, time to try to move the terrible looping. Part of that is my own fault; I thought it would be cool to give it a camera-pan. Sadly the point when it reaches the end of the camera loop, and starts returning, is *really* obvious.

After effects to the rescue. I took the clip into AE and looped it 3x at 30FPS. I then added a bit of a camera movement in the middle rather, and a bit of time-warp to try to blur the point of discontinuity (with limited success). Finally I added a bit of motion blur to the final output to smooth out movements.

The final result suffers from the classical programmer problem of trying to fix a bug by process; if I'd taken my medicine at the beginning and just set it to render overnight I wouldn't have spent all this time playing with tools to patch up my low-res input. But it did teach me a lot about the tools available to optimize daz animations, and as my first character animation I'm not unhappy with it. Hope you enjoyed it; going into the next build as an Act IV+ event, triggering off the Grassmarket at night.