VP of Technology at @theblaze | AI Filmmaker | Ex-Disney
Georgia
Joined January 2008
- Tweets67.8K
- Following1.5K
- Followers42.1K
- Likes75K
Pinned Tweet
Episode 4: Mary Sue Thinks the Villain Is Misunderstood
Produced by @capcutapp using Seedance 2.5
This is what peak performance looks like.
This is a screenshot of Remake Master, an Ai movie tool I vibe-coded in one hour and it is the single best tool I've ever used in 20 years of digital content creation.
From day 1 there hasn't been a single bug problem, it's more stable than Photoshop, it runs lean and allows me to speed-iterate new movie clips because every video is still connected to the prompt, source plates, character sheets and the original graphic novel PDF.
What's amazing is it knows the story I'm working on. It was trained on my own animation style from timing, to staging to my 2D art that focuses on strong silhouette and exaggerated pose-to-pose performance. It read all of the personality traits of every character in the graphic novel and pulled every line of dialogue from the book and puts it in every clip.
I could put this out to the public for free and it would work, but it wouldn't help you. It is specifically designed to make just this project because I created it while working. It aped what I wanted in any given moment, I hate key-commands so everything is mouse-driven. I hate pull down menus and tool bars so it's all written for a cave man. When I need a new feature I tell it, "Now make an icon that tells me there is take 2 for take two available for this shot." I don't tell it how to do it, I just tell it what I want and it works.
I can talk to it while I'm working, "Get rid of the previous take, remind me later that this doesn't hook up and I'll need to regenerate".
This is the first project I ever worked on where the computer is invisible. I don't feel like I'm crippled by someone else's UI design I have to spend months learning, this one became an instant extension of me from day one. If you had to train on this tool it would take you weeks to reorient your head into my preferences, or you could just write your own perfect tool in one hour that works the way you do.
On my next Ai feature movie I wouldn't use this again. I would dump it and just make a new one that is designed for that project. It's that easy.
This is a fun style.
Josh Daws retweeted
The best thing to do with Ai is to tell it what you want. Don't try to break it down into code. We are so used to computers not working that we translate our wants into computer-speak to try to get something to work. Ai is an interpreter to break down your desire for use into code. So start with big picture things, "I want to write a book in 3 months" or "redrawing storyboards is taking me 6 months" then have it give options you can try to get to what you want. Try not to limit what it can do from the start, think of what you want and usually it will rise up to the task and do things you didn't think Ai could pull off yet.
Celebrating my son’s 18th birthday (a few days early) with some bluegrass at Everett’s Music Barn. This place is a great celebration of American culture. I love it.
Josh Daws retweeted
.@DreamLabLA is in the studio cooking up heat 📷
Real actors. Live sets that change on the spot. A crew working with AI in the room.
@FortuneMagazine went inside Dream Lab LA to see hybrid production in action. 📷
Josh Daws retweeted
9 months of @thisisoddkid and we already have a slate of 10 projects in the works.
Some are in very early development, some are in development for streamers, some are fully funded, others are in negotiations.
It’s been a pretty wild ride so far, and if there’s one thing I’m becoming increasingly convinced of, it’s that AI is a catalyst, not magic.
It makes exploring ideas faster, experimenting cheaper and bringing worlds to life more accessible than ever.
But it doesn’t tell you which ideas are worth pursuing, which stories are worth telling, or where you should spend your time and energy.
At the end of the day, it still comes down to what you believe deserves to exist.
And then doing the work to make it happen.
I let this one run overnight.
“Same process. Make an 80s sci-fi action film about an animatronic teddy bear fighting robots with laser guns. Under a minute.”
This was made 100% with local models. Every frame and every sound in this cartoon came from AI models running on an NVIDIA DGX Spark on my desk. No cloud generation, no subscriptions.
"Santa and the Little Lost Star"
All I typed was "make a Christmas short under a minute" (plus "Santa" and "1930s cartoon style"). I didn't write the story, plan the shots or edit it. Opus 5.5 came up with the story, wrote the narration, generated every shot on the Spark, rejected the bad takes, mixed the sound and cut it together.
It took about 4 and a half hours today. The first full cut was done in an hour. The rest was agent directed polishing: 28 video renders, about 3 hours of GPU time, for 22 seconds of finished film.
It was all made in Imagination Studio, a free, open-source local film studio built by @SamJWasserman. This thing is ridiculous. One installer downloads every model you need, and an AI agent can drive the whole thing.
What made what (all running locally):
Video: MiniMax H3 (FastH3) with the Studio 1939 LoRA by @OdinLovis for the golden-age hand-painted look
Title card: MOSS-Image
End card: FLUX3-K
Keyframe fixes: Qwen-Image 2.1
Narrator and Santa's voice: Fish Audio Drama 3
Score: MiniMax Music 3
The "Merry Christmas everyone" song: YuE2 Sound effects: Stable Audio Open
github.com/drowzeys/Imaginat…
This was made 100% with local models. Every frame and every sound in this cartoon came from AI models running on an NVIDIA DGX Spark on my desk. No cloud generation, no subscriptions.
"Santa and the Little Lost Star"
All I typed was "make a Christmas short under a minute" (plus "Santa" and "1930s cartoon style"). I didn't write the story, plan the shots or edit it. Opus 5.5 came up with the story, wrote the narration, generated every shot on the Spark, rejected the bad takes, mixed the sound and cut it together.
It took about 4 and a half hours today. The first full cut was done in an hour. The rest was agent directed polishing: 28 video renders, about 3 hours of GPU time, for 22 seconds of finished film.
It was all made in Imagination Studio, a free, open-source local film studio built by @SamJWasserman. This thing is ridiculous. One installer downloads every model you need, and an AI agent can drive the whole thing.
What made what (all running locally):
Video: MiniMax H3 (FastH3) with the Studio 1939 LoRA by @OdinLovis for the golden-age hand-painted look
Title card: MOSS-Image
End card: FLUX3-K
Keyframe fixes: Qwen-Image 2.1
Narrator and Santa's voice: Fish Audio Drama 3
Score: MiniMax Music 3
The "Merry Christmas everyone" song: YuE2 Sound effects: Stable Audio Open
github.com/drowzeys/Imaginat…
Good word here.
Young men, this message is for you.
You are inheriting a difficult world. The economy is stacked against you, institutions increasingly treat you as the problem, and it is easy to become cynical and discouraged.
But God has not made a mistake in creating you. You CAN build, lead, cultivate, and leave something better for the next generation.
Obi-Wan @SamJWasserman: You've taken your first step into a larger world
The future of AI is local.
Josh Daws retweeted
If your version of Jesus…
+Supports abortion.
+Celebrates lifestyles that Scripture calls people to repent of.
+Affirms the idea that we can redefine the gender God created.
+Claims there are multiple ways to Heaven.
Then you’re following a false Jesus.
You’re not worshipping the Jesus of Scripture. You’re worshipping your own ideologies and simply attaching His name to them.
