Software Engineering Courses (SECourses)•3y ago

any sd. 2.1 512/768 model suggestion?

Furkan Gözükara SECourses•1/30/23, 10:34 PM

xyz_grid-0008-1283833964-photo20of20testsd21_asdf20very20high20detail20photorealistic20cinematic20lighting20awesome20quality20anime20style203d20render20rpg20game20charact.png

CCoqueTornadoMusic any sd. 2.1 512/768 model suggestion?

Furkan Gözükara SECourses•1/30/23, 10:34 PM

i dont know any :d

Furkan Gözükara SECourses•1/30/23, 10:34 PM

looks like 1.5 best

CoqueTornadoMusicOP•1/30/23, 10:35 PM

ah, interesting

CoqueTornadoMusicOP•1/30/23, 10:35 PM

https://civitai.com/models/5395/mangled-merge

Mangled Merge | Stable Diffusion Checkpoint | Civitai

Introduction:Mangled Merge V1 is a merge of 11 different v2.x 768x768 models currently available. All keywords from these models will work to help guide the model but not as powerfully as the individual models they were merged from. However, this is counterbalanced by the amazing versatiliy the v2.1 model gains from merging them all together. Ma...

CoqueTornadoMusicOP•1/30/23, 10:35 PM

this looks fine

Furkan Gözükara SECourses•1/30/23, 10:35 PM

i will test cfg now

CoqueTornadoMusicOP•1/30/23, 10:36 PM

: )

CoqueTornadoMusicOP•1/30/23, 10:36 PM

try it with another cooler model of 2.1 (the equivalent of Protogen)

CCoqueTornadoMusic ok

Haashim•1/30/23, 10:37 PM

2.1 base model was really good for me

Haashim•1/30/23, 10:37 PM

Just requires different prompting

CoqueTornadoMusicOP•1/30/23, 10:37 PM

different prompting?

CCoqueTornadoMusic https://civitai.com/models/5395/mangled-merge

Colored Tonalities•1/30/23, 10:37 PM

mangled, yeah

CoqueTornadoMusicOP•1/30/23, 10:37 PM

but we are in the marvelloous world of TI

Haashim•1/30/23, 10:37 PM

It’s a different text encoder

CoqueTornadoMusicOP•1/30/23, 10:38 PM

TI can be swapped easily

CoqueTornadoMusicOP•1/30/23, 10:38 PM

ah, yeah, I know... so all the prompts I have wont work properly there... ok... good new

Haashim•1/30/23, 10:39 PM

But for me the photorealistic outcome in 2.1 base was better then protogen

Haashim•1/30/23, 10:39 PM

In 1.5

Furkan Gözükara SECourses•1/30/23, 10:39 PM

cfg effect

xyz_grid-0009-1283833964-photo_of_testsd21_1.5_very_high_detail_photorealistic_cinematic_lighting_awesome_quality_anime_style_3d_render_rpg_game_characte.jpg

Furkan Gözükara SECourses•1/30/23, 10:39 PM

for textual inversion

Furkan Gözükara SECourses•1/30/23, 10:39 PM

i think you cant change cfg :d

HHaashim But for me the photorealistic outcome in 2.1 base was better then protogen

CoqueTornadoMusicOP•1/30/23, 10:39 PM

Protogen x3.4 (Photorealism) ?

Furkan Gözükara SECourses•1/30/23, 10:39 PM

probably it is trianed on 7

FFurkan Gözükara SECourses cfg effect

CoqueTornadoMusicOP•1/30/23, 10:40 PM

any of them looks like you more than the 90%?

FFurkan Gözükara SECourses probably it is trianed on 7

Haashim•1/30/23, 10:40 PM

O wow. Can we set this parameter ?

CCoqueTornadoMusic any of them looks like you more than the 90%?

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

not yet

HHaashim O wow. Can we set this parameter ?

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

not sure :d

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

the prompt has dramatic effect on sd 2.1

FFurkan Gözükara SECourses not yet

CoqueTornadoMusicOP•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

CoqueTornadoMusicOP•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

CoqueTornadoMusicOP•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 ...

CoqueTornadoMusicOP•1/30/23, 10:43 PM

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

CoqueTornadoMusicOP•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

CoqueTornadoMusicOP•1/30/23, 10:46 PM

good question

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

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

same

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

for style TI is said to be better

Furkan Gözükara SECourses•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...

CoqueTornadoMusicOP•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?

any sd. 2.1 512/768 model suggestion?

Similar Threads

Similar Threads

Similar Threads