Stable diffusion thread

[doogyhatts]

Kuai Shou now have their own image model, except it uses Unet as opposed to DiT architecture.
In the meantime, their Kling platform is now available for limited use.

 

[doogyhatts]

AI girls can now be upgraded with LivePortrait.
Just use existing IG stories with facial expressions as the driving video.

 

[doogyhatts]

Kling AI's web platform also has T2I.
Testing biomechanical creature again without IP adapter.

Using other people's generated image from MJ as prompt.

 

[Prismatic]

wa.. making money with instagram model ..

is it easy?
what you use? SDXL or just SD1.5?

Instagram is ridiculously behind on the monetization boat lah. It's only good as a convenient online gallery. Anyway, the only way to make money off AI models is nakeed models through OF or Patreon.

I'm using IG mainly as a showcase. I'm trying to create more realistic (not hyper-realistic) images that are more true to life. Actually, most of the time, people knows immediately it's AI, because the face is ridiculously perfect. So have to introduce a bit of imperfection.

So far for cinematic shots, I feel that MJ excels in this area. SDXL1.5 is good too, but sometimes some aspect like the focusing, bokeh looks a bit weird.

One of my most popular shot below.

 

[focus1974]

Instagram is ridiculously behind on the monetization boat lah. It's only good as a convenient online gallery. Anyway, the only wa

One of my most popular shot below.

very nice. ..
the face is unique or just MJ generated?

 

[doogyhatts]

Instagram is ridiculously behind on the monetization boat lah. It's only good as a convenient online gallery. Anyway, the only way to make money off AI models is nakeed models through OF or Patreon.

I remember reading a post on reddit that someone managed to make 5k usd from offering her AI artist services on Fiverr.
And some of her clients will ask for nude AI artworks.

But judging from the layout and sheer number of freelance artists on Fiverr, I don't think it is possible now for anyone to simply join in.

 

[doogyhatts]

So I found out that Lumina team will be integrating SD3's 16ch vae into their upcoming model.
Not sure why they are not using AuraDiffusion's 16ch vae.

 

[doogyhatts]

AnimateDiff LoRA for walking, trained using Motion Director.

 

[doogyhatts]

I am going to start a new channel on animals to see how well AI animals perform.
Because Kling AI has animal motions!!!

 

[AZE]

Kuai Shou now have their own image model, except it uses Unet as opposed to DiT architecture.
In the meantime, their Kling platform is now available for limited use.

According to their paper, Hunyuan actually uses unet style DiT, a sort of hybrid system.
A paper published at similar time by researchers with support from the Huawei's Noah Ark's labs(the one supporting the release of the original Pixart), also investigate about that.
Conclusion ish Unet-DiT ish better if implement properly to capture the unique architecture benefits of both.

 

[AZE]

So I found out that Lumina team will be integrating SD3's 16ch vae into their upcoming model.
Not sure why they are not using AuraDiffusion's 16ch vae.

If use SD3's VAE means under their license already(which ish different from the original open licenses of SD1,2,XL).
Means SAI can issue take down notice or retract the license at any time and require the users/dev to delete all derived models and materials, however unlikely it may be.
It ish better for creators and devs to not use that, or at least create two different models in parallel, one SD3 based and others open license based, just in case.

 

[doogyhatts]

Biomecha test again using Fooocus and Level4-alpha model. No image prompt.

 

[AikiBoy]

plug in an anime figurine lora...
from this

To this

 

[doogyhatts]

Incorrect generation for Dryad.

 

[AikiBoy]

Biomecha test again using Fooocus and Level4-alpha model. No image prompt.

U used what to make this?

 

[AikiBoy]

Oh thanks! I didn't actually notice it the first time! Managed to remove it in ver 2.

wao, what you use to make these?


next thing i want to learn is the background and surrounding environment.

 

[doogyhatts]

U used what to make this?

Fooocus and Level4-alpha model

 

[AikiBoy]

Fooocus and Level4-alpha model

u use so many app.

 

[AikiBoy]

Instagram is ridiculously behind on the monetization boat lah. It's only good as a convenient online gallery. Anyway, the only way to make money off AI models is nakeed models through OF or Patreon.

I'm using IG mainly as a showcase. I'm trying to create more realistic (not hyper-realistic) images that are more true to life. Actually, most of the time, people knows immediately it's AI, because the face is ridiculously perfect. So have to introduce a bit of imperfection.

So far for cinematic shots, I feel that MJ excels in this area. SDXL1.5 is good too, but sometimes some aspect like the focusing, bokeh looks a bit weird.

One of my most popular shot below.

really got onlyfan's AI model meh... how to make $$ oh. SG got people make AI model n earn in OF?

 

[Prismatic]

I use Midjourney (paid services) and Fooocus. Fooocus is mainly for the body poses and faces because can load LoRAs into it. Then finish off with MJ for the photographic finish. You need to be precise with your prompts. Not too much, not too little.