Software Engineering Courses (SECourses)•13mo ago

Most CogVideoX videos 👎 There was no human in the prompt, no opening of the door or anything like t

Most CogVideoX videos

There was no human in the prompt, no opening of the door or anything like that. I only wanted a camera rotation...

KKallamamran Most CogVideoX videos 👎 There was no human in the prompt, no opening of the doo...

Furkan Gözükara SECourses•12/12/24, 1:04 PM

can you give image and prompt

Furkan Gözükara SECourses•12/12/24, 1:04 PM

i can test

Furkan Gözükara SECourses•12/12/24, 1:19 PM

download speeds of runpod

Furkan Gözükara SECourses•12/12/24, 1:19 PM

man everytime i regret immediately this is L40S community machine

Furkan Gözükara SECourses•12/12/24, 1:50 PM

without my special download script it downloads with 40 mb second

Furkan Gözükara SECourses•12/12/24, 1:59 PM

this model really supper

FFurkan Gözükara SECourses this is newest : CogVideoX1.5-5B-I2V

wardensc2•12/12/24, 2:31 PM

Do you have any tips for prompt to make less distort people face or distort video

Furkan Gözükara SECourses•12/12/24, 2:33 PM

Wwardensc2 Do you have any tips for prompt to make less distort people face or distort vide...

Furkan Gözükara SECourses•12/12/24, 2:34 PM

not yet atm

KKallamamran Yeah... 8GB is low :/ With the correct model maybe 🤨

Drocsid Imperator Maximus•12/12/24, 2:48 PM

sad noises

FFurkan Gözükara SECourses can you give image and prompt

KallamamranOP•12/12/24, 2:50 PM

My bad. Used a prompt generator which added the "woman" :/

Furkan Gözükara SECourses•12/12/24, 3:05 PM

just perfect

KallamamranOP•12/12/24, 3:13 PM

Trying a camera rotation around a car. As usual I get like 5 degrees camera motion

KallamamranOP•12/12/24, 3:15 PM

Gotta try Hunyuan or what it's called

KallamamranOP•12/12/24, 3:15 PM

Sana only generated black images, so f**k that as well

Furkan Gözükara SECourses•12/12/24, 3:27 PM

Furkan Gözükara SECourses•12/12/24, 3:28 PM

pay attention to small details

Furkan Gözükara SECourses•12/12/24, 3:28 PM

like flying birds

KallamamranOP•12/12/24, 3:28 PM

Yeah, but try making a larger rotation around it

KallamamranOP•12/12/24, 3:29 PM

This is just... Zzzzzz

Furkan Gözükara SECourses•12/12/24, 3:29 PM

i tried simple prompt

Furkan Gözükara SECourses•12/12/24, 4:18 PM

published : https://www.patreon.com/posts/112848192

Patreon

1-Click CogVideoX1.5-5B-I2V Image to Video Installers For Windows, ...

Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon

J2_Nedo•12/12/24, 4:33 PM

i have to test this later at home to compare to Hunyuan Videos

Furkan Gözükara SECourses•12/12/24, 4:34 PM

raw video requires 19 GB VRAM - 1360x768px - 81 frames - 16 fps

Furkan Gözükara SECourses•12/12/24, 4:35 PM

used image for comparison if anyone wants - A majestic golden airship flying fast on a majestic city inside clouds

screencapture-5b455ba3e58ea0b66b-gradio-live-2024-12-12-19_16_22.png

Furkan Gözükara SECourses•12/12/24, 4:35 PM

@J2_Nedo if you test let me know posted image and prompt above

J2_Nedo•12/12/24, 6:53 PM

i added " camera flying close to the airship"

Furkan Gözükara SECourses•12/12/24, 10:01 PM

8x rtx 4090 generating CogVideoX1.5-5B-I2V videos

Furkan Gözükara SECourses•12/12/24, 10:01 PM

Furkan Gözükara SECourses•12/12/24, 10:02 PM

screencapture-0a326b3ffcb94df320-gradio-live-2024-12-13-01_01_36.png

Furkan Gözükara SECourses•12/12/24, 10:14 PM

wow cool

Furkan Gözükara SECourses•12/12/24, 10:29 PM

�

𝕏4C•12/13/24, 3:47 AM

Hello @Furkan Gözükara SECourses i have a question regard flux lora training , where can i see config file used for Best_v4-000150, i found that perform best

becruily•12/13/24, 9:20 AM

In Supir is it possible to increase VRAM usage? With my 3090 it only uses about 12GB, its tilled I think
but I wanted to know if I can increase this so that its faster

Zhigarev•12/13/24, 12:08 PM

Tell me the most important parameters in descending order that make LoRA training more effective and faster. As I understand it, for characters, activity settings are important, while for training styles, it’s necessary for the model to memorize general style features. Accordingly, different settings for LR, Dim, and Alpha are needed—not 128 and not LR 0.0005. Or am I mistaken? Which setting greatly affects the training time? I just want to understand if, for example, image quality improves by 10-15%, but the time doubles, then I’ll use those settings

Punceka•12/13/24, 12:14 PM

Hi. For the last 2 weeks I've been training loras of people with your recommended settings. The end results was fine. This time, I'm trying to train a lora of a gas mask and no matter what I do, what recommended settings I use, it is always distorted. A lora that weights over 3gb can't replicate a simple gas mask.
For example this one here. A prompt - A man wearing a GP5M4sk
The result. Why Flux always adds a hood/a hat even if there's none in my training images or modifies the mask with objects that were trained with flux model and not mine?