Software Engineering Courses (SECourses)•3y ago

not yet

HHaashim O wow. Can we set this parameter ?

Furkan Gözükara SECoursesOP•1/30/23, 10:40 PM

not sure :d

Furkan Gözükara SECoursesOP•1/30/23, 10:41 PM

the prompt has dramatic effect on sd 2.1

FFurkan Gözükara SECourses not yet

CoqueTornadoMusic•1/30/23, 10:41 PM

that is the thing... dreambooth make magic

Haashim•1/30/23, 10:41 PM

These S/R tests are important then

Bassna•1/30/23, 10:41 PM

What's the best way to go about getting a similar style to clipart images, I've collected a small set and tried one DB model but I think I overtrained it quite a bit. Was a few weeks ago. Was going to give it another shot. DB again, maybe an embedding? Here's my source

CoqueTornadoMusic•1/30/23, 10:42 PM

nah, been trying with Fátima and Émily with S/R tests and not luck... also prompting can help... it is quite a hard work... but not really worked

CoqueTornadoMusic•1/30/23, 10:42 PM

maybe using inpainting xD...

CCoqueTornadoMusic Protogen x3.4 (Photorealism) ?

Haashim•1/30/23, 10:43 PM

Yeah. That one or dreamlike photoreal was my favorite in 1.5

BBassna What's the best way to go about getting a similar style to clipart images, I've ...

CoqueTornadoMusic•1/30/23, 10:43 PM

probably the TI thing we are talking about it was made for things like this...?

CoqueTornadoMusic•1/30/23, 10:44 PM

or even better, just try the newest extension custom-diffusion-webui and let us know the results...

Haashim•1/30/23, 10:44 PM

Yeah probly 2.1 TI style training

FFurkan Gözükara SECourses cfg effect

Haashim•1/30/23, 10:45 PM

You used the same process as TI video? Or new findings

CCoqueTornadoMusic or even better, just try the newest extension custom-diffusion-webui and let us...

Bassna•1/30/23, 10:45 PM

Sorry I missed that discussion, I'll try and find it above. I haven't messed with TI stuff yet so I guess down that rabbit hole I go

HHaashim You used the same process as TI video? Or new findings

CoqueTornadoMusic•1/30/23, 10:46 PM

good question

HHaashim You used the same process as TI video? Or new findings

Furkan Gözükara SECoursesOP•1/30/23, 10:47 PM

same

Furkan Gözükara SECoursesOP•1/30/23, 10:47 PM

i am working on prompts now

BBassna What's the best way to go about getting a similar style to clipart images, I've ...

Furkan Gözükara SECoursesOP•1/30/23, 10:47 PM

for style TI is said to be better

Furkan Gözükara SECoursesOP•1/30/23, 10:47 PM

you tried?

BBassna Sorry I missed that discussion, I'll try and find it above. I haven't messed wit...

CoqueTornadoMusic•1/30/23, 10:47 PM

the TI thing was made for styles they said in the papers. So.. But it is possible to achieve the 90% of similarity in humans

Bassna•1/30/23, 10:47 PM

I haven't tried anything with TI yet, been using DB for everything. I guess today's the day though :e_kekw:

Bassna•1/30/23, 10:49 PM

https://www.youtube.com/watch?v=dNOpWt-epdQ

YouTubeSoftware Engineering Courses - SE Courses

How To Do Stable Diffusion Textual Inversion (TI) / Text Embeddings...

In this video, I am explaining almost every aspect of Stable Diffusion Textual Inversion (TI) / Text Embeddings. I am demonstrating a live example of how to train a person face with all of the best settings including technical details.

If I have been of assistance to you and you would like to show your support for my work, please consider bec...

Bassna•1/30/23, 10:49 PM

This still the way to go about it?

Furkan Gözükara SECoursesOP•1/30/23, 10:50 PM

this is interesting

Furkan Gözükara SECoursesOP•1/30/23, 10:50 PM

Furkan Gözükara SECoursesOP•1/30/23, 10:50 PM

CoqueTornadoMusic•1/30/23, 10:52 PM

negative prompts

Furkan Gözükara SECoursesOP•1/30/23, 10:52 PM

how 75 goes to 150

Furkan Gözükara SECoursesOP•1/30/23, 10:52 PM

weird

FFurkan Gözükara SECourses Click to see attachment

Colored Tonalities•1/30/23, 10:53 PM

part of a token?

Colored Tonalities•1/30/23, 10:53 PM

CoqueTornadoMusic•1/30/23, 10:54 PM

mayb a bug

Furkan Gözükara SECoursesOP•1/30/23, 10:54 PM

by the way

Furkan Gözükara SECoursesOP•1/30/23, 10:54 PM

negative prompt

Furkan Gözükara SECoursesOP•1/30/23, 10:54 PM

seriously removes your face

Furkan Gözükara SECoursesOP•1/30/23, 10:54 PM

from results

CoqueTornadoMusic•1/30/23, 10:54 PM

yep

CoqueTornadoMusic•1/30/23, 10:55 PM

in TI I used to avoid too many negative prompts

CoqueTornadoMusic•1/30/23, 10:55 PM

(if that is the context you are talking about now heheh)

CCoqueTornadoMusic yep

Furkan Gözükara SECoursesOP•1/30/23, 10:55 PM

yep

CoqueTornadoMusic•1/30/23, 10:55 PM

yes, prompts like "ugly" can make the face another one

CoqueTornadoMusic•1/30/23, 10:56 PM

maybe it is because TI is about tokens, prompts

CoqueTornadoMusic•1/30/23, 10:56 PM

so one can change the whole thing

CoqueTornadoMusic•1/30/23, 10:56 PM

I dunno really

Furkan Gözükara SECoursesOP•1/30/23, 10:57 PM

TI affected more than dreambooth that is for sure

Furkan Gözükara SECoursesOP•1/30/23, 10:57 PM

i am gonna test just like this now

CoqueTornadoMusic•1/30/23, 10:59 PM

it is quite delicated the thing of TI

CoqueTornadoMusic•1/30/23, 11:01 PM

i have found that if you train it in the vanilla, it works better in the vanilla. But not too much in protogen. And if you train it in protogen, it works better in protogen. But not really amazingly. Just an 60%/70% ... it is quite hard... I need like 10 or 20 more tests... trial and error. Maybe you, the community can help

Haashim•1/30/23, 11:03 PM

The Clip interrogator V2 is helpful for prompting

not yet

Similar Threads

Similar Threads

Similar Threads