Software Engineering Courses (SECourses)•10mo ago

I will give you the original prompt from last video which worked:

Ccangir01 I will give you the original prompt from last video which worked:

Furkan Gözükara SECourses•4/1/25, 12:09 AM

sure

cangir01OP•4/1/25, 12:09 AM

Here is the video without sound and prompt

00006.txt1.65KB

cangir01OP•4/1/25, 12:10 AM

another similar video:

00002.txt1.57KB

Ccangir01 another similar video:

Furkan Gözükara SECourses•4/1/25, 12:11 AM

testing without teacache

Furkan Gözükara SECourses•4/1/25, 12:11 AM

Furkan Gözükara SECourses•4/1/25, 12:12 AM

cangir01OP•4/1/25, 12:12 AM

There are a lot of results, which were also not good, but the promlem is the text in most cases not the visuals, as you know transformer models have still difficulty to create letters, sometimes they are really good but never perfect, always small mistakes

cangir01OP•4/1/25, 12:13 AM

Without cache in 9 min is fast enough I think.

Furkan Gözükara SECourses•4/1/25, 12:14 AM

ye this is on rtx 5090 and 50 steps

Furkan Gözükara SECourses•4/1/25, 12:14 AM

lets see

cangir01OP•4/1/25, 12:14 AM

I have 4090

cangir01OP•4/1/25, 12:23 AM

Bu arada,, as I saw information about you, I realized that you are from mersin? I lived there 5 years many many years ago

Ccangir01 Bu arada,, as I saw information about you, I realized that you are from mersin? ...

Furkan Gözükara SECourses•4/1/25, 12:25 AM

true

Furkan Gözükara SECourses•4/1/25, 12:25 AM

it worked great

Furkan Gözükara SECourses•4/1/25, 12:25 AM

700 seconds

Furkan Gözükara SECourses•4/1/25, 12:25 AM

Furkan Gözükara SECourses•4/1/25, 12:26 AM

dmn i am surpised how good text it can write

Furkan Gözükara SECourses•4/1/25, 12:26 AM

now will test with teacache

Furkan Gözükara SECourses•4/1/25, 12:27 AM

teacache super fast haha

cangir01OP•4/1/25, 12:28 AM

But this means, cache lowers quality significantly with new version, untill v49, there was no quality difference with or without cache no difference anly it was 2 x times faster with cache. Now it is 4 times faster but quality decreases, especially with text it gets not usable as you experienced yourself

Furkan Gözükara SECourses•4/1/25, 12:31 AM

ye 0.15 is too big

Furkan Gözükara SECourses•4/1/25, 12:31 AM

testing 0.1

Furkan Gözükara SECourses•4/1/25, 12:31 AM

cangir01OP•4/1/25, 12:32 AM

and with 0.05? just for comparison

Furkan Gözükara SECourses•4/1/25, 12:32 AM

i can test more tomorrow probably

cangir01OP•4/1/25, 12:32 AM

ok its too late sure

Furkan Gözükara SECourses•4/1/25, 12:32 AM

i got to sleep

cangir01OP•4/1/25, 12:33 AM

Iyi geceler, yarin görüsmek üzere

mihutJedi•4/1/25, 6:38 PM

quick question: how does the batch process work for the gradio app for wan, I mean how to name the images in the input folder, and where to add the prompts, is there any details on this somewhere?

ᴏ

ᴏᴍɴɪ•4/1/25, 8:00 PM

@Furkan Gözükara SECoursesHi- the pickle versions from the Wan install ex. the xlm-roberta model is safe since its official Wan repo? Going forward would you be able to always use the safetensor versions of models?

MmihutJedi quick question: how does the batch process work for the gradio app for wan, I me...

Furkan Gözükara SECourses•4/1/25, 10:07 PM

it is like this. image1.png, image1.txt if not it will use prompt box

Furkan Gözükara SECourses•4/1/25, 10:07 PM

image names can be anything

ᴏᴏᴍɴɪ <@205854764540362752>Hi- the pickle versions from the Wan install ex. the xlm-ro...

Furkan Gözükara SECourses•4/1/25, 10:07 PM

yes it is from official repo you can see it

Furkan Gözükara SECourses•4/1/25, 10:08 PM

wan 2.1 used that i dont know if safetensors replacement would work i didnt test

Weatherby•4/1/25, 11:58 PM

Has anyone had any luck setting up deep seek R1 locally? (I mean really locally, not with 16x A100's)

FFurkan Gözükara SECourses yes it is from official repo you can see it

ᴏ

ᴏᴍɴɪ•4/2/25, 12:03 AM

I know just wondering. why they still use pickle when safetensor eases peoples mind haha.

ᴏᴏᴍɴɪ I know just wondering. why they still use pickle when safetensor eases peoples m...

Furkan Gözükara SECourses•4/2/25, 12:08 AM

i, agree

WWeatherby Has anyone had any luck setting up deep seek R1 locally? (I mean really locally,...

Furkan Gözükara SECourses•4/2/25, 12:08 AM

you need like 0.5 bit :d

Furkan Gözükara SECourses•4/2/25, 12:09 AM

16 bit version is 680 GB

Furkan Gözükara SECourses•4/2/25, 12:09 AM

1 bit is 42 gb haha

Furkan Gözükara SECourses•4/2/25, 12:09 AM

i got to sleep now see you later hopefully

Erik•4/2/25, 3:21 AM

Guys what's the current best face swap option?

FFurkan Gözükara SECourses i got to sleep now see you later hopefully

Weatherby•4/2/25, 3:25 AM

Take care!

J2_Nedo•4/2/25, 9:42 AM

VACE Preview released ! https://ali-vilab.github.io/VACE-Page/

VACE

All-in-One Video Creation and Editing

EErik Guys what's the current best face swap option?

Furkan Gözükara SECourses•4/2/25, 11:12 AM

facefusion and visomaster

Weatherby•4/2/25, 9:45 PM

So, where are we with upscalers? Anything better than SUPIR yet?

WWeatherby So, where are we with upscalers? Anything better than SUPIR yet?

Furkan Gözükara SECourses•4/2/25, 9:47 PM

nothing better than SUPIR yet for image upscale

Weatherby•4/2/25, 9:48 PM

That's what I figured, but I haven't been keeping close tabs on that segment. I won't complain, though, SUPIR works well, especially with some reall nice sdxl checkpoints.

Inoshi•4/2/25, 11:06 PM

https://github.com/hacksider/Deep-Live-Cam

GitHub

GitHub - hacksider/Deep-Live-Cam: real time face swap and one-click...

real time face swap and one-click video deepfake with only a single image - hacksider/Deep-Live-Cam

Inoshi•4/2/25, 11:07 PM

https://www.reddit.com/r/StableDiffusion/s/juhlgOZb8w

From the StableDiffusion community on Reddit

Explore this post and more from the StableDiffusion community

I will give you the original prompt from last video which worked:

Similar Threads

Similar Threads

Similar Threads