Stable diffusion thread

[zheng]

testing new model

 

[doogyhatts]

PW no longer free for larger images liao.
Must pay in crypto or CC.

Lite plan gives only 11200 credits. Each generation task in fast queue cost between 28 to 50 credits. (20 steps at 50 credits)
Cost is dependent on image size and iteration steps.
Enough for 16 days on cheapest, if each day generate 25 images, total 400 images per purchase.

 

[PrincZe]

PW no longer free for larger images liao.
Must pay in crypto or CC.
Lite plan gives only 11200 credits. Each generation task in fast queue cost 50 credits.
Enough for 9 days, if each day generate 25 images.

How much would that be


Read HWZ Forum Rules!

 

[doogyhatts]

what gpu u using and how long each generation?

I would worry about the size of the vram and not the speed of the generation task.

To answer x1234x question on whether 12gb ram is worth or not? I would say YES!

According to the documentation here,

• The maximum size that can fit on 6GB GPU (RTX 2060) is around 1152x1088.

https://github.com/dieghobonilla/stable-diffusion-optimized

Why is this so?
Because a larger vram allows you to generate those landscape images, so your overall video quality goes up.
But your inference speed will decrease due to memory optimisations, so it will take a while to generate the large image.

And JoeMultimedia in his latest video showed that he upgraded to a RTX4070 12gb, he was previously limited to a RTX3060 8gb.

 

[doogyhatts]

How much would that be

Lite plan is $2.99 per month. 11.2K credits.
Plus plan is $19.17 per 3 months. 90K credits.
Ultimate plan is $89.94 per 12 months. 600K credits.

In crypto or CC.

 

[bbfa gene]

Thank me later
https://civitai.com/
prompt for prawns
or if lazy can explore the fantasy of other likeminded guys.

 

[zheng]

I would worry about the size of the vram and not the speed of the generation task.

To answer x1234x question on whether 12gb ram is worth or not? I would say YES!

According to the documentation here,

• The maximum size that can fit on 6GB GPU (RTX 2060) is around 1152x1088.

https://github.com/dieghobonilla/stable-diffusion-optimized

Why is this so?
Because a larger vram allows you to generate those landscape images, so your overall video quality goes up.
But your inference speed will decrease due to memory optimisations, so it will take a while to generate the large image.

And JoeMultimedia in his latest video showed that he upgraded to a RTX4070 12gb, he was previously limited to a RTX3060 8gb.

my home pc 2060 8gb
den my office pc one 3060 12gb - i tested there is like many times faster already

if want 40XX go for 16gb better

 

[zheng]

Thank me later
https://civitai.com/
prompt for prawns
or if lazy can explore the fantasy of other likeminded guys.

ya everyone know about thats just not talking about pornz here
u can join 4chan for that


anyway realistic-vision model new update look damn good...

https://civitai.com/models/4201/realistic-vision-v20

 

[zheng]

anyway 8gb isnt that limiting with a few combo of upscaling technic...
i can actually get up to very high quality/detailed 4k images

 

[x1243x]

ya everyone know about thats just not talking about pornz here
u can join 4chan for that


anyway realistic-vision model new update look damn good...

https://civitai.com/models/4201/realistic-vision-v20

looks really good..

 

[doogyhatts]

Lite plan is $2.99 per month. 11.2K credits.
Plus plan is $19.17 per 3 months. 90K credits.
Ultimate plan is $89.94 per 12 months. 600K credits.

In crypto or CC.

Updated for correctness.

Free tier still available but subjected to max users of 500 at any one time and can only generate lower resolution of 512x768.

 

[doogyhatts]

Testing output from Pixai Real v3.0.
Looks even chioer than before!
But got hand problems.

PixAI now has upscaling to 2.6, means can now do landscape of width 1992.

 

[Gaiason]

my home pc 2060 8gb
den my office pc one 3060 12gb - i tested there is like many times faster already

if want 40XX go for 16gb better

moi also gpu 8gb and set the xformers seems faster a bit.. but still have prob to test run dreambooth..

 

[potatohead]

imagine 10 years from now, you and your wife/gf wear a special headset. she can imagine piaking her favourite korean boy band group while you can imagine piaking japanese av girl but real life is piak each other.

maybe by then the scenarios u described and many more are juz 1 pill away ... no other gears needed.

 

[zheng]

moi also gpu 8gb and set the xformers seems faster a bit.. but still have prob to test run dreambooth..

just do lora training loh think 8gb vram can already

 

[YouHeypiCanLiao]

https://civitai.com/models/11096/irene@Jian19

 

[YouHeypiCanLiao]

https://civitai.com/models/13125/aespa-karina@sexeyXR @chsjason @Hisashiburi

 

[Jian19]

https://civitai.com/models/11096/irene@Jian19

Look weird sia

 

[zheng]

testing new model again - OrangeChillMix.. also testing seed travel script for morph like animation

 

[bbfa gene]

Can SD expert verify that this is really AI_ART?
https://twitter.com/ai_gyunyu