I remember when DeepDream first came out, and WaveNet not long after. I was immediately convinced this stuff was going to play a huge role in media production.

I'm a big hobbyist filmmaker. I told all of my friends who actually work in film (IATSE, SAG, etc.) and they were so skeptical. I tried to get them to make an experimental film using DeepDream.

This was about the same time Intel was dabbling in 360 degree filmmaking and just prior to Epic Games / Disney working on "The Volume".

I bought a bunch of Kinects and built a really low-fidelity real time version of what Intel was working on. The sensors are VGA resolution, so it's not at all cinematic.

When Stable Diffusion came out, I hooked up Blender to image-to-image and fed it frames of previz animations to convert to style transferred anime. Then IP Adapter. Then Animate Diff.

My friends were angry with me at this point. AI was the devil. But I kept at it.

I built an animation system for the web three years ago. Nonlinear timeline editing, camera controls, object transformations. It was really crude and a lot of work to produce simple outputs: https://storyteller.ai/

It was ridiculously hard to use. I typically film live action for the 48 Hour Film Project (twice annual film "hackathon" that I've done since I was a teenager). I used Mocap suits and 3D animation, and this is the result of 48 hours of no sleep:

https://vimeo.com/955680517/05d9fb0c4f

We won two awards for this. The audience booed us.

The image-to-video models came out right after this and immediately sunk this approach. Luma Dream Machine was so easy and looked so much better. Starting frames are just like a director and DP blocking out a scene and then calling action - it solved for the half of the problem I had ignored, which was precisely controlling for look/feel (though this abandons temporal control).

There was a lot of slop, but I admired the work some hard-working people were creating. Those "movie trailers" people were criticizing were easily 10 hours of work with the difficulty of the tech back then.

I found use in model aggregation services like OpenArt and FreePik. ComfyUI is too far removed for me - I appreciate people who can do node magic, but it's not my thing.

I've been working on ArtCraft ( https://github.com/storytold/artcraft ), which is a more artist-centered version for blocking out and precisely articulating scenes.

My friends and I have been making a lot of AI films, and it's almost replaced our photons-on-glass filmmaking output. (We've done some rotoscoped AI + live action work.)

https://www.youtube.com/watch?v=Tii9uF0nAx4 (live action rotoscoped film)

https://www.youtube.com/watch?v=v_2We_QQfPg (EbSynth sketch about The Predator)

https://www.youtube.com/watch?v=tAAiiKteM-U (Robot Chicken inspired Superman parody)

https://www.youtube.com/watch?v=oqoCWdOwr2U (JoJo grinch parody)

We're going to do a feature length film at some point, but we're still building up the toolbox.

If you're skeptical about artists using AI, you should check out Corridor Crew. They're well respected in our field, they have been for over a decade, and they love AI:

https://en.wikipedia.org/wiki/Corridor_Digital

https://www.youtube.com/watch?v=DSRrSO7QhXY

https://www.youtube.com/watch?v=GVT3WUa-48Y

https://www.youtube.com/watch?v=iq5JaG53dho

They're big ComfyUI fans. I just can't get used to it.

Real filmmakers and artists are using this tech now. If you hate AI, please know that we see this more as an exoskeleton than as a replacement. It enables us to reach the look and feel of a $100+ million dollar Pixar, Star Wars, or Marvel film without the budgets we could never have without insane luck or nepotism.

If anything, this elevates us to a place where we will one day be competing with Disney. They should fear us instead of the other way around.