Software Engineering Courses (SECourses)•3y ago

I asume you didn't train yet with Pixart? They just updated their repo and now collaborate with the

I asume you didn't train yet with Pixart? They just updated their repo and now collaborate with the LCM Team

(

New) Nov. 30, 2023.

PixArt collaborates with LCMs team to make the fastest Training & Inference Text-to-Image Generation System. Here, Training code & Inference code & Weights & Demo are all released, we hope users will enjoy them. Refer to docs for more details. At the same time, we update the codebase for better user experience and fix some bugs in the newest version.

BBenjamin I asume you didn't train yet with Pixart? They just updated their repo and now c...

Furkan Gözükara SECourses•11/30/23, 9:30 PM

i havent trained yet

Furkan Gözükara SECourses•11/30/23, 9:30 PM

they ddont have any example how to train lol

FFurkan Gözükara SECourses check this out : https://www.linkedin.com/posts/furkangozukara_stable-diffusion-...

Kallamamran•11/30/23, 10:12 PM

Pixart is fast! Same time to render a 512x512 as 2048x2048.... Wierd!

Kallamamran•11/30/23, 10:13 PM

Always seem to get this "dotted" images though

KKallamamran Always seem to get this "dotted" images though

Furkan Gözükara SECourses•11/30/23, 10:14 PM

use my auto installer i have no such issue

Kallamamran•11/30/23, 10:14 PM

I have used your installer

Furkan Gözükara SECourses•11/30/23, 10:14 PM

i made amazing comparison list here

Furkan Gözükara SECourses•11/30/23, 10:14 PM

https://twitter.com/GozukaraFurkan/status/1730316580494393635

KKallamamran I have used your installer

Furkan Gözükara SECourses•11/30/23, 10:14 PM

what is your prompt i can try

Kallamamran•11/30/23, 10:15 PM

Ford mustang driving fast down a runway

Kallamamran•11/30/23, 10:15 PM

Just tried something

Kallamamran•11/30/23, 10:15 PM

Nothing fancy, nothing well thought through

Kallamamran•11/30/23, 10:16 PM

SA-Solver almost seem better?!

Kallamamran•11/30/23, 10:17 PM

I don't get how it works

512x512 takes the same time as 2048x2048 and produces the same image

Kallamamran•11/30/23, 10:18 PM

Also resolution has NO impact on VRAM

Kallamamran•11/30/23, 10:18 PM

This is somewhat impressive

Furkan Gözükara SECourses•11/30/23, 10:22 PM

ye this prompt is broken :d

Furkan Gözükara SECourses•11/30/23, 10:22 PM

KKallamamran This is somewhat impressive

Furkan Gözükara SECourses•11/30/23, 10:22 PM

accurate

Furkan Gözükara SECourses•11/30/23, 10:23 PM

try this prompt

Furkan Gözükara SECourses•11/30/23, 10:23 PM

beautiful, vibrant phoenix portrait in the dark sky spreading wings, highly detailed, hyper realistic, best quality

Kallamamran•11/30/23, 10:23 PM

Furkan Gözükara SECourses•11/30/23, 10:24 PM

i think because when you increase resolution

Furkan Gözükara SECourses•11/30/23, 10:24 PM

it still generates 1024x1024

Furkan Gözükara SECourses•11/30/23, 10:24 PM

and upscale haha

Kallamamran•11/30/23, 10:24 PM

but when you generate 512... It downscales?!

Kallamamran•11/30/23, 10:24 PM

It takes the same time, so

Furkan Gözükara SECourses•11/30/23, 10:24 PM

i didnt test 512 yet

Furkan Gözükara SECourses•11/30/23, 10:25 PM

ye it doesnt generate higher res

Kallamamran•11/30/23, 10:25 PM

It's the same, just better quality

Kallamamran•11/30/23, 10:25 PM

Explains things

Furkan Gözükara SECourses•11/30/23, 10:25 PM

testing right now

Kallamamran•11/30/23, 10:25 PM

I really, REALLY wish we had access to Midjourneys upscale function!

Kallamamran•11/30/23, 10:26 PM

256x256 didn't downscale though

Kallamamran•11/30/23, 10:26 PM

Horrible quality

Kallamamran•11/30/23, 10:27 PM

512x512 is really sharp though

KKallamamran I really, REALLY wish we had access to Midjourneys upscale function!

Furkan Gözükara SECourses•11/30/23, 10:27 PM

that would be nice

Kallamamran•11/30/23, 10:28 PM

SDXL Turbo then? What's your impressions

Kallamamran•11/30/23, 10:28 PM

I have seen some models already out on Civitai

Kallamamran•11/30/23, 10:29 PM

I think speed is not the biggest problem we have though. It's quality

Furkan Gözükara SECourses•11/30/23, 10:30 PM

it is lower quality

Kallamamran•11/30/23, 10:32 PM

So when do we get models with higher quality I wonder. Focus seem to be on speed

AIGambino•11/30/23, 10:36 PM

I'm trying to do an exorcist style image, with the girl kind of upside down, but the face wont attach or make the same pose, any advice on how to do this

Kallamamran•11/30/23, 10:36 PM

I'm impressed with PixArt, but I can't really use it for anything, so I'm sticking with A1111 and SDXL for now

AAIGambino I'm trying to do an exorcist style image, with the girl kind of upside down, but...

Kallamamran•11/30/23, 10:40 PM

Stabel Diffusion sucks at up-side-down!

Kallamamran•11/30/23, 10:41 PM

Try in Dalle-3 and then inpaint in A1111

KKallamamran So when do we get models with higher quality I wonder. Focus seem to be on speed

Furkan Gözükara SECourses•11/30/23, 11:19 PM

i agree. all recent ones focuses on speed

MMishroom Hi! I have some issues with Dreambooth Kohya, I tried to start learning but it s...

Evan•12/1/23, 12:42 AM

has anyone tried realistic vision v6 yet?

Rae•12/1/23, 1:02 AM

@Dr. Furkan Gözükarahigher quality than what? good prompts can generate insane photoreal pictures with 1.5, SDXL is even beyond that, what else in quality do we need? we need easier ways to finetune, tag, and prompt and generate.

Rae•12/1/23, 1:03 AM

it looks like most attention is at getting these photos to videos, every single day you can see the videos getting better and better.

I asume you didn't train yet with Pixart? They just updated their repo and now collaborate with the

Similar Threads