Archived: IO and IO — Bonus Video
Posted 10 February 2026
Archived 13 May 2026
BONUS DROP: The Bloopers Reel (and the brutal math behind one “simple” video)
This wasn’t on the schedule. It happened because I kept saving the failures.
If you’ve ever wondered what AI video creation actually looks like from the artist side—this is the honest version. Not the polished “wow” moments. The weird elbows. The impossible joints. The suddenly-melted faces. The lighting that decides to become a different century mid-shot. The render that nails the mood… and then gives one of my twins a third 'something' like it’s a gift.
So I’m posting an impromptu monthly bonus: part blooper reel, part walkthrough—a look at what I go through to get one single finished video out the door.
The workflow (no mythology, just numbers)
Each finished piece is built in fragments—clip by clip—because text-to-video behaves like a temperamental animal.
90–120 finished clips per video (sometimes more, depending on pacing and coverage)
~20 render attempts per clip is normal
That’s 1,800–2,400 attempts for a clean 90–120-clip video!!!
And that’s before I’m even happy with continuity, body logic, lighting consistency, and the “this is actually my world” feeling
And yes: all strictly original text-to-video content. No recycling someone else’s footage. No shortcuts. No “train it on a movie and pretend it’s art.” Just me, the prompts, the revisions, and the relentless stupid labor.
What this bonus shows
What “bad renders” actually look like (and why they fail)
The kinds of errors that happen over and over (hands, torsos, anatomy, motion logic, camera drift)
Why a 12-minute film isn’t “one render,” it’s a constructed object—assembled like architecture
I love these characters. I’m not pretending otherwise. I love the fact that without fail they truly appear to love each other and they laugh at everything. Good company... But the point here is the work: the errors, the fixes, the discipline it takes to keep a visual world coherent when the tool keeps trying to hallucinate its way out of it.
Comments (4)
I have been playing with text to picture generation. I like to create American military men and you would think the ai would know that each man wear dog tags! Nope. The same with the hair style, beard, and chest hair. I’ve also learned ai doesn’t know what a cock or dick is, I have to use the correct term penis. Instead of fuck I have to use the term sodomy. Also I get freaks with multiple limbs and multiple cocks. Just some of my experiences.
I had no idea. Thank you so much for your original work and keeping it sacred!
Thanks for that interesting bonus video. It's amazing the quality of finished production you manage to create with all the AI randomness and limitations to deal with! I guess the difficulty and results also depends on the AI software package you use. I've seen some other creators get really good outcomes for anatomy and stuff like BJs, but usually the clips are very short like 15 secs so they've prob had to try loads of times to get that one which looks perfect. The marvel of your work (and a couple of similar creators) is going to the trouble of editing it all together fluidly to make a longer movie of 10 - 15 mins, plus with a story, music and voices. So appreciated! 😁♥
The fifteen second ones are the one's I talk about in the video. They are new skins on a 3D model, basically - one basic piston gesture - in-and-out of whatever orifice, altered with AI. Not really flexible. Pick hair, pick face, pick position...Hit go.