I get a lot of questions, and I get into a lot of arguments, about hybrid workflow.

So rather than answer every post, I'll just use this project to illustrate the behind the scenes reality of artists that mix conventional and AI workflows. It's commonly misunderstood, and it doesn't help that there are about 1000 variations, each a bit different. I can't explain them all, you know, time constraints, but what I can do is walk through what goes into a "small hybrid project". In example, this artwork is perhaps 2% of the man hours that go into something like Dragon Coin, and some very tiny fraction of a percent of what's gone into Labyrinth so far.

----------------
How this work came to be.

I had been doing openGL shader programming for 20 or 30 hours, and I got bored and needed to take a break.

So I was playing guitar and came up with a catchy bassline on the low strings of an Ibanez Gem 7v.

I recorded that riff, and provided it to an AI, requesting that it be transferred to a blues Bass/80s pop sound.

I worked with detailed prompts over the course of about 30 versions of an instrumental track based on the riff.

After working for several hours, I was listening intently to one of those variations, and about 2 minutes into the song, I found it, the exact version of the groove that I had imagined. Maybe there were around 200 different takes on that same riff amongst all the versions.

I took that perfect sample loop, and fed it back into a music AI I had trained myself on about 30 of my own music tracks that I had played with guitars and keyboards in the past, without AI. The training data for this model is made exclusively from recordings of me playing different kinds of music. No moral statement here, no activist ideological agenda. I just wanted it to sound like me across all the instruments.

I created about another 30 or 40 versions, all instrumental, trying out all different kinds of ensembles, mixes, genre variations for the master mix. Soon I hit one that sounded good. It felt like it needed lyrics.

I sat for about 7 hours trying to write a decent poem that said something interesting while complimenting the vibe of the riff.

About 2 hours into that 7, I had written the chorus, which I thought was perfect. "It's not the world that you were promised, It's not the world that you once knew, Now it's just people in sunglasses, turning horses into glue"

Then I tried to write some verses for it. I wrote about 20, and they were all terrible except one, which I liked. I used that one.

I tried to write 2 more verses and failed. I got half of one right, but I was getting tired and so I just used that one.

It was day 3 of part time work getting this put together, only a few hours a day to dedicate. I used the instrumental track version to drive about 170 vocal takes with the AI singer performing my lyrics.

Day 4. I wanted to make a video for it, but didn't have any ideas right at that moment. The song had a lot of energy, maybe something fun and simple would work. Let the music carry the experience and the video is just going along with it.

Day 6. doing about 4 hours work on average over 3 days I've got about 500 plus shots. I put them into Save Point's Vortex extrapolator module, the one that directs and edits video, and let it make about 60 versions of the complete video. I didn't use these, except to just watch and learn how different footage from the available library interacted with the music. Sometimes a certain scene would just hit right with a section of music, other times I was just getting to know all the available shots I had to work with.

Day 7. I go through all the footage and just delete any shots that are bad or dull. That cuts it down to about 45 minutes of raw footage, I put that footage into Resolve with the finished track.

Day 11. I've finished cutting and editing the footage down to a little over 3 minutes. I've added 3 new parts to the song, and carefully tracked out the audio inside Resolve. I hand cut every clip, then adjusted start times to get better sync with the music.

I spent 10 minutes doing the title, selecting from a list of several hundred fonts till I found one that fit, then scaled, spaced, positioned it, and placed it on the timeline where I thought it worked best.

I set up the encoding settings for the render, rendered it, Took the finished file, put it into Topaz AI video upscaler, brought it up to 4k with some motion fixes, then uploaded it.

So that is the workflow for what I'd describe as a "small hybrid project".

indietalk

Message History

Delivery option(s)