Stable diffusion thread

[AlmightyOnes]

use 繁体
so they know is not ah tiong channel

Why not use both lol..
I saw people 'spamming' it that way =x

 

[zheng]

do u guys find it hard to generate single person in 16:9 aspect ratio?

The AI seems to keep wanting to generate more than 1 person when i use a wide ratio, even when i use strong prompts like (1girl:1.9)
Most of the time it still generates more than 1 girl.

16:9 is problematic... bec model is train on 512 1:1

gen on 16:9 is like 2 block of 512... so alway get 2 heads or 1 head with extra body haha

gen in 1:1 and just outpaint to 16:9

 

[order66]

you tried the Beetlejuice method?

what;s that

 

[order66]

16:9 is problematic... bec model is train on 512 1:1

gen in 1:1 and just outpaint to 16:9

yeah guess have to do it that way

i was trying to avoid that by forcing the AI to generate 16:9 with strong prompts, but the results are extremely unreliable.

 

[zheng]

Why not use both lol..
I saw people 'spamming' it that way =x

yah if u want to capture ah tiong audiences also

 

[order66]

#比基尼 #清纯 #天真 #小妹妹 #骆驼脚趾

 

[x1243x]

what;s that

you repeat the same prompt 3 times.. apparently some ppl say it works if you die die can't get the effect you want.. but the above suggestion to up scale seems better

 

[order66]

gonna upload another cat video to my channel to appease my subs, before i start uploading NSFW stuffs.
Cat pics harder to generate compared to cute girl pics...

 

[doogyhatts]

Top geographies​

Views · Since uploaded (lifetime)

	United States		42.9%
	Canada		2.9%
	Russia		2.3%
	United Kingdom		2.1%

All angmohs...

I need asian viewers who can appreciate my TW XMM pictures.

Mine is 65-70% Japanese viewers, so my keywords are mostly in Japanese characters.

 

[order66]

Mine is 65-70% Japanese viewers, so my keywords are mostly in Japanese characters.

gonna put japanese tag in my videos too!

#痴漢電車

 

[doogyhatts]

Wah this swimsuit channel started on 7th June 2023, already 163K views!

 

[order66]

Wah this swimsuit channel started on 7th June 2023, already 163K views!

gonna start all out generating bikini, swimsuit and tehkor pictures now!

 

[doogyhatts]

Bikini video done liao, so next step is ZG video, then open legs after that
that's the standard progression right?

You have generated so many pictures for a 3:45 min video.
Then repeat with different environment, different poses, new characters, new clothing, etc.

Your channel is already on the YPP?

Speaking of which, I have yet to figure out how to generate new images for the same character with a different pose, from an existing image.

 

[zheng]

You have generated so many pictures for a 3:45 min video.
Then repeat with different environment, different poses, new characters, new clothing, etc.

Your channel is already on the YPP?

Speaking of which, I have yet to figure out how to generate new images for the same character with a different pose, from an existing image.

need to use controlnet for that

 

[doogyhatts]

need to use controlnet for that

I am going to try reference_only and see if I can generate new character bodies with similar clothing first.
Then use Panohead to rotate the head and Roop to transfer the face.

 

[zheng]

something new drag gan

 

[order66]

You have generated so many pictures for a 3:45 min video.
Then repeat with different environment, different poses, new characters, new clothing, etc.

Your channel is already on the YPP?

Speaking of which, I have yet to figure out how to generate new images for the same character with a different pose, from an existing image.

nope, my views are all from years ago

Need 4000 watch hours for last 365 days, which i dont have
have to start all over again

 

[doogyhatts]

something new drag gan

I tried it yesterday or the day before.
They have not added the functionality to import an image.

LeiaPix depth converter cannot be used liao because they decided to limit usage to only people whom have bought their products.
No wonder the recent videos by ai_gyunyu are of lower quality. He had been doing mov2mov using his own generated videos.

Now need to check the 3D Photo Inpainting technique again.
It is a more precise depth-based animation technique compared to what LeiaPix depth converter is using.
I remember generating some videos nicely out of it before, but it is awfully slow. And sometimes, the process may stall due to out of memory.

 

[zheng]

I tried it yesterday or the day before.
They have not added the functionality to import an image.

LeiaPix depth converter cannot be used liao because they decided to limit usage to only people whom have bought their products.
No wonder the recent videos by ai_gyunyu are of lower quality. He had been doing mov2mov using his own generated videos.

Now need to check the 3D Photo Inpainting technique again.
It is a more precise depth-based animation technique compared to what LeiaPix depth converter is using.
I remember generating some videos nicely out of it before, but it is awfully slow. And sometimes, the process may stall due to out of memory.

can try controlnet depth

 

[doogyhatts]

can try controlnet depth

I was referring to this one instead.
It can auto generate four types of videos.

The control-net depth is for posing.