Stable diffusion thread

[codemasterguy]

Now doing images for 3rd video. Going to see if it can latch onto Honeyworks popular song.

Stable diffusion can make video??? How do you do it?

 

[Perisher]

Midjourney and stable-diffusion are the winners when it comes to image-rendering.

Limited right? Need to pay to generate…

 

[zheng]

Limited right? Need to pay to generate…

mj v5 for sure beats sd many times but not free...
hope sd 3.0 will get better... 2.0 has almost no adaptation... many still stays at 1.5

 

[doogyhatts]

Stable diffusion can make video??? How do you do it?

Put images together as a video as per normal.
Stable diffusion just generate the images.

 

[doogyhatts]

Twitter now have a lot of AIgravure content created using SD.
A few accounts have now breached the 50K followers goal.

Interestingly, my first YT short can get more viewership than the long-form video, although still lower than Tiktok video.
So that means I need to create more of such short-form videos.
But the fastest to gain followers is still Twitter, so eventually I will have to do AIgravure.

For now, I am still doing lower-tier content.

 

[x1243x]

Twitter now have a lot of AIgravure content created using SD.
A few accounts have now breached the 50K followers goal.

Interestingly, my first YT short can get more viewership than the long-form video, although still lower than Tiktok video.
So that means I need to create more of such short-form videos.
But the fastest to gain followers is still Twitter, so eventually I will have to do AIgravure.

For now, I am still doing lower-tier content.

not bad the hands look quite good

 

[Gaiason]

fantasy test...

 

[Mystyque]

fantasy test...

This is nice (ignoring fingers) even if not sexy

 

[thisisnotme1212]

haves tos trains thes models ownselfs?
cans trains withs idols?
withs kpops idols faces cans relates mores

 

[doogyhatts]

haves tos trains thes models ownselfs?
cans trains withs idols?
withs kpops idols faces cans relates mores

No need to train the models ourselves. Just use ChilloutMix.

Jpmm faces. The creator is a Japanese AI/ML expert.
https://forums.hardwarezone.com.sg/threads/atbs-created-by-ai.6874205/page-3#post-146820551

 

[zheng]

testing prompt for some futuristic space theme

 

[doogyhatts]

Today I woke up to my first YT anime short at >585 views and gained 8 subscribers!
So doing YT shorts is a good way to gain followers even if few people only view the long-form videos.
And it now did better than my Tiktok videos.

In the meantime, I tried PixAI's variations. It tries to maintain the chio appearance of the girl with slight changes and gives you 3 more upscaled images.
This reduces the average cost of 1 upscaled image from 2900 to 2075 credits.

Two images showing the difference between PixAI and PW.

 

[doogyhatts]

Improved my techniques in PW to get an even more chio girl than before. But there isn't a way to get variations.

Villager girl kym?

 

[zheng]

Improved my techniques in PW to get an even more chio girl than before. But there isn't a way to get variations.

Villager girl kym?

looks good
throw in some keywords like
cinematic photography, bokeh, depth of field,

 

[doogyhatts]

The AiTV_ channel now has a few videos that reached 1M views liao.
So from now onwards, all my videos will have the "AI Try On" substring in the title.
See if it can allow the YT algorithm to recommend both my shorts & longs to their viewers.

Some progress on my YT. The first short reached 860 views liao despite having a glitch on frame 0.
Gained another 3 followers. Two more shorts taking off soon after a very slow start.

I noticed most of the viewership still correlate to US timezone.
Asia is now the content producer for such content and the US side is the consumer.

Everyday I try to upload at least 1 long and 3 shorts to YT. Maybe doing 8 shorts would be better.

Last night managed to get a side view. But there is something wrong with the strap.

 

[x1243x]

Improved my techniques in PW to get an even more chio girl than PixAI.

Villager girl kym?

you

The AiTV_ channel now has a few videos that reached 1M views liao.
So from now onwards, all my videos will have the "AI Try On" substring in the title.
See if it can allow the YT algorithm to recommend both my shorts & longs to their viewers.

Some progress on my YT. The first short reached 860 views liao despite having a glitch on frame 0.
Gained another 3 followers. Two more shorts taking off soon after a very slow start.

I noticed most of the viewership still correlate to US timezone.
Asia is now the content producer for such content and the US side is the consumer.

Everyday I try to upload at least 1 long and 3 shorts to YT. Maybe doing 8 shorts would be better.

Last night managed to get a side view. But there is something wrong with the strap.

looks good. the hand looks normal too.. didn't really notice the strap until you pointed it out.

 

[doogyhatts]

Output from PW's new Anime v2.5 model.
Image width has been temporary reduced to 512 when this was generated.

 

[li9ht5peed]

stable diffusion cannot generate photorealistic images of wrecked cars, deaths and suicides, with details of broken body parts. i have little interest in images of healthy people

 

[accucore12]

what gpu u using and how long each generation?

 

[zheng]

what gpu u using and how long each generation?

im using 2060rtx
small preview can go 4 image per min
high quality can go up to 5min per image