I have 12 GB and it tried to allocate 12.38.
I have 12 GB and it tried to allocate 12.38.
PGood night everyone, I have a question, can I add a different language to the Elevenlabs site myself and train it?
![Digital [Starburst]](https://cdn.discordapp.com/avatars/220961909447458817/8ecbbb3132fc6e44312f3d6d2ae3df69.webp?size=40)
DOkay I understand. The instructions say the longer the video the more VRAM it uses. So I split the video I'm trying, and scaled it down. I'll see if I can get it to a size the software can handle.
![Digital [Starburst]](https://cdn.discordapp.com/avatars/220961909447458817/8ecbbb3132fc6e44312f3d6d2ae3df69.webp?size=16)
Freduce densepose duration
SAnyone around tha tknows bout creating diffrent -virtural drives for multipule versons of python??
DI got it to work with a 640X480 video that is 30 seconds in link, but I didn't let it complete. I let it run for hours (fell asleep waiting), and it said it was at 03:43:?>37:?:? hours, and I said no way am I waiting that long for it to process a single video. The video isn't all that fantastic anyway. It was just a test.
Ddensepose resulted in 900 frames.
DI noticed that for every 30 seconds of video, I got 900 frames.
DThen I took a look at the example video. It is only about 2 seconds long. If this works the way I think it does, a 5 second video is going to take about 1 hour per 1 second of video. Is there much you can do about the processing time?
DOkay a 10 second video creates 300 frames, so lets see how long it says that will take.
DI'm not even sure what to expect it to look like when it is done.
Fit is about fps
DOkay. So what do I do with that to keep it from using too much VRAM?
DI download a video off a tube website that allows it using a downloader. I clip it to 10 seconds regardless of how long it is, but if I understand you correctly, I don't have to cut the video that short. Thing being that at 10 seconds it generates 300 frames and the ETA is around 9 hours to process it on a 4070ti.
DI have 12 GB of VRAM.
DI know you mentioned 16 is ideal, but I couldn't afford that, so I'm sort of making due with what I have.
Fonly way currently is making densepose video shorter
TWhy don’t I have a batch button?
TFbecause my auto installer uses my own gradio
Fnot llava gradio app
TWhat I installed is the automatic installation package released by your patreon.
Fcan you show me your install folder screenshot
T
T
T
F1 sec let me verify something
DOkay.
Fprobably download command didnt work
DFit must be due to a library missing
TI have reinstalled the computer system and installed the dependencies. I have two computers with the same problem
DI'm not getting the impression that SD is really built for video work. I've generated some mp4's I sent my family that made them motion sick. Other than that I haven't gotten it to do much.
Fi think it is because
Fyour folder has space character
DThe idea is cool, but I don't think the technology is there yet.
Fi will dm you now
Fi sent you gradio file put into this folder
Fye not very good
DI decided I didn't have the patience to wait nine hours to see what happens with a ten second video. When I checked the benchmark rank for graphics cards about the only thing better than what I have was a 4080. It had more VRAM, but cost nearly double the 4070ti.
DBefore video generation can really be used for anything useful they are going to have figure out a way to speed it up somehow with lighter hardware.
Fye it is slow
ADo the following work for both dreambooth, Lora, Textual inversion:
Basic Captioning
BLIP Captioning
GIT Captioning
WED14 Captioning
Manual Captioning
Or are the specific to each?
Basic Captioning
BLIP Captioning
GIT Captioning
WED14 Captioning
Manual Captioning
Or are the specific to each?
Fyes all works
Fit just different results
Aare there strength and weaknesses for them or do most just use BLIP?
Fyes each one fits
Fdifferent cases
Fi try them
JHello Dr. Furkan, i followed your tutorial to get a lora trained using a rtx 4090 gpu, but at my studio we are wanting to use multi-gpu training for sdxl, ive run into issues trying to configure kohya ss to do this properly (it ends up running into a memory issue after maxing out the memory of the first gpu), would you be willing to help me with this issue?