Computer Impulse Buy / AI Ganbang / Soaking Discussion

(post deleted by author)

Iiz it a butt cheek tho?

1 Like

yes

1 Like

Couldn’t figure out how to chunk the model above into frame batches so memory didn’t run away. Gonna install comfyUI again and check out SeedVR2. Thank God for Chinese AI projects or open source would be fisted to death by our oligarchs already.

I’m inside the computer~! :face_with_spiral_eyes:

ComfyUTI not too bad but was indeed annoying to set up.

2 Likes

would

Nice dude wrote do a json workflow for this that works good.

If anybody want lemme know.

maybe we should vibe code a sane UI over this shit and make a $1B startup
do you know how to sell thingz ?

2 Likes

I was a VP of a software company for 4 months once. It was a super unpleasant gig.

Would be pretty easy to run this model buried in the cloud with an app store based gui boomers could use on their phones.

They have old media and all the money they stole from younger generations. Would probably work.

I’m too lazy for all that sheeyat tho.

Randomly I suggest this on the Topaz forum not long ago.

2 Likes

Figuring out how to run this model without blowing my memory tricky. I think I’ll get it down today then have sit back and savor nominal differences to the solutions I paid for. :sunglasses:

The tech wizards workflow busts (non jizz) on two points. You need path based video import nodule in place of regular and you need meta batch chunker.

So dis shit. Enjoy!

This will save you a bunch of hours of tinkering. Works as good as any paid solution but requires some brain juice to setup and make work well.

Used FFMPEG to split the video file into tiles and processed them separately since the chinamen haven’t shipped VAE Tile shits in the SeedVR2 yet. Worked good!

I spent some time trying to understand the frame batch function on the seedVR node and mechanics described in paper. The temporal flutter is related to a lack of overlap function. The first and last 15 frames of each batch see less information to turn into temporal smoothing as they near cutoffs. I’ll see if i can find a solution there. Until then, I extended my frame batch size to 197 so the tails at beginning and end are statistically less of the overall frame pool. This is why the longer batch size was looking better.

So this overlap function I need doesn’t exist in comfyui world. I’m going to try to vibe code it into the meta frame batch nodes.

K nar pray for me!

1 Like

Ho vers la flamer cummah by tomorrow.

Too much detail poss.

1 Like

3 Likes

I’m too lazy to edit out wanda from his dialogue or vid stuffs. This labor intensive as fuck but interesting.

1 Like

This thread I guess.

I develop the thingies i imagine using vibe coding THO should be called brute force coding cuz I try endlessly until something works like I expect and this permit me to proceed without learning shit or remembering sheyat.

1 Like

Sacred relic. The tearing in ultra fast spots wasn’t the interlacing or pull down but that he likely went faster than the interlacing of the monitor they recorded at the time which was filmed.

#fasterthankinoscopes

1 Like