Music Video Test - 90s References, MC Hammer Pants
Please enjoy my 1990 music video made with the help of RuneXX workflows. Details below
I had to generate about 150 clips for this 1 minute video. LTX is terrible at fast motion, faces at a distance and fast moving faces at a distance! I spent a lot of time testing 3 samplers passes, open that 3 would fix all the errors.
Whats very interesting is that I didn't have to use any vocal separation. All of this is direct music audio. LTX 2.3 knew when not to sing (for the most part) . No need to use Mel Band! I prompted "Dancing is synced to the music"
I'm using BF16 models with a RTX 5090 and 128 GB of RAM on an 3 year old Threadripper.
Music made with Suno
Image variations with Klein 9B
Video made with LTX 2.3 at 1080p
About 10 minutes per 10 second clip.
I most all sort of experiments, create custom nodes and workflows if you want to follow me on Linkedin
I have a cool Sigma curve editor I'm going to share. Its great for fine tuning motion.
I also spent a lot of time making prompt instructions. Everytime I asked an LLM to improve my instructions it would make it worse. Very odd. I can post my instructions in a comment if anyone's interested.
⢠LinkedIn: https://www.linkedin.com/in/matthew-hallett-041a3881
⢠My AI site mostly about architecture: https://www.hallett-ai.com
⢠Instagram: https://www.instagram.com/hallettvisual
⢠Facebook: https://www.facebook.com/hallettvisual
Wow.. that looks amazing ;-)
And quite professional and production ready looking ;-)
Well done ;-) I wanted to try out something similar myself, but havent gotten around to it yet (I added Ace Step and HeartMula in my comfy setup already hehe.. but yes Suno is better)
And yes there was a post here with a guy asking about 3-pass sampler, and tested it a few times. It does help with details and motions.
And I have followed you on YT for a long time ;-) watched some tram video and i remember your 2 guys in the car etc ;-)
Will check out your other links too
That was fun and catchy! I hope you'll consider redoing it whenever 2.5 drops, would love to see how it would look/sound with the (hopefully) improved model.