Thanks! Nice video. Now I wondering how to fine-tune the svd using my own video dataset. Any source to learn about it?
@cliffordmwenda90
9 ай бұрын
Great video!
@thomasgoodwin2648
9 ай бұрын
I haven't tried it yet, but then my hardware's barely up to pushing out SD1.5, and struggles with SDXL. I'm curious, but really don't feel like waiting hours for a 50/50 chance of producing something watchable, let alone artistic. Most of the good stuff I have seen appears to have been done on rather high end hardware, so I'm guessing a lot of cherry picking is still going on. Having said that, let me also add that SD is freaking amazing! It truly is a completely new artistic medium. Even though my published works so far have been largely lost in the river of flowing published porn, I do think think it's potential as a new tool to study and express the living condition through art is historically unparalleled. The occasional love I do get in response to posted images is enough to encourage me to continue to push the boundaries in my own (hopefully interesting) voice (still trying to learn what that one is, but it seems MAD magazine may come to haunt my old age as it did my youth😉). Is this tech disruptive? OH YEAH! Here's what I see coming. The movie industries in general, and specifically the porn industry, are doomed. Why sift through mountains of someone else's fantasies when there is this magic box under the desk can deliver up whatever your naughty little heart desires within seconds. Also, AI 'Actors' will look and sound perfect with complete multi-modal support. So who really wants to pay to watch a bunch of whiny, flawed, real humans when "my personal hopes, fears, and dreams" can be made manifest at the touch of a button for free? (very 'Brave New World') Well speaking of manifest(os), I seem to have gotten a bit very long winded here, so I'll shut up for now. ☮ n ❤Louis, 🖖😎👍 (P.S. Using ComfyUI as my front end forced me to learn a lot about the models and how they work and interact. Enough at least that I could follow along and grok the 'technical' bits.😁Think I'll give myself a gold star for that. 🌟 Yay!) A1111, Dall-E, etc., are all functional but not very educational. Comfy puts you under the hood. Expect, respect, and grow to ❤the steep learning curves.
@catspyjamass
9 ай бұрын
Is there any AI platform that allows you to upload an image (e.g. a watch) and then ask the system to generate 10 images of random people wearing that watch?
@thomasgoodwin2648
9 ай бұрын
To some degree, Stable Diffusion can already do that. It's a bit tricky,, and results might need a fair bit of cherry picking, but it can be done a few ways. The quick and dirty way would be to use RemBG (Remove Background) or similar segmentation tool to isolate the image of the watch. Then simply mask it, and outpaint the scene around it. The best practice way to accomplish this would require a few pictures of the watch. Then you could just train a simple 'bolt on' model such as a LoRA (LOw Rank Adapter) or Text Inversion. Once the watch is reduced to Latent Space, you could do pretty much anything you want with it. Turn it blue, give the dog a watch for a face, melt the watch over ice cream... whatever a twisted (or even normal) mind can come up with.
@catspyjamass
9 ай бұрын
@@thomasgoodwin2648 Thanks. Yes assume that the watch photos would be available as product pack shots (from various angles) on a white background.
Пікірлер: 7