Stable diffusion thread

[zheng]

nice but lack realism if u are trying for photorealism style... this one look too cg/3d, lack skin textures all that

 

[zheng]

testing chillioutmix model with kr, tw likeness lora

 

[HushiePuppies]

interesting.. is it worth paying extra for the gpu to play with stable diffusion?

just get a 4080 16gb la. it's less than $2k

https://www.amazon.sg/Graphics-IceStorm-Advanced-Lighting-ZT-D40810J-10P/dp/B0BKK371SB/

 

[HushiePuppies]

courtesy of my gpu. i use ControlNext extension as well to get the pose I wanted.

Fully text-to-img generated using ControlNet_scribble model with input of me (guy) wearing a checkered shirt.

Prompt: Highest Quality masterpiece, (Split lighting photography), Highly detailed photo of attractive latino woman with luxuriant blonde hair wearing checkered oversized shirt, portrait, perfect face, alluring eyes, (highly detailed skin), toned body, (key light:0.4), split lighting portrait, medium depth of field, photographed on a Canon EOS R5, 100mm lens, F/4 aperture, (hyperdetailed, intricate details), sharp focus, 8k, 9mm film grain

Steps: 25, Sampler: DPM++ SDE Karras, CFG scale: 7.5, Seed: 4206853858, Size: 600x800

 

[HushiePuppies]

I used a selfie photo of my wife taken in our bedroom then let AI "draw" out it out (left) before generating a photo-realistic image based on that sketch (right). My wife see liao stunned.

 

[GaoProsper]

So many abusing it for their own desire...and its quite easy to setup and run the models. Once you see the community site, you will notice so many creators with ..horn-y imaginations.

 

[zheng]

try openpose already?

 

[orpisia]

What on earth stable diffusion means

 

[doogyhatts]

I haven't figure out how to generate this kind of anime style yet.
I am noticing those creators can build communities around such art easily.

Spoiler

 

[3XTR3M1ST]

Non-upscaled images that I generated from PixAI.
Not bad, got Korean and Japanese Doll Likeness LoRAs.
Got 10K free credits daily. Every day, I can generate about 4 upscaled images under low priority.
But the upscaling is limited to 1.3 only.

got upscaling limit for cup size?

 

[doogyhatts]

got upscaling limit for cup size?

Yes, it is dependent on the prompt.

 

[zheng]

for those who wanna play play for free can try- https://leonardo.ai/
many free token... can go there steal their prompt too

 

[zheng]

new midjourney v5 is super awesome... getting more realistic and better hand

 

[YouHeypiCanLiao]

Thought got new stable to share in edmw..

Stable your head

 

[doogyhatts]

Discovered that I can use PW to generate sfw girls.
No tokens required except they use a different model compared to PixAI, and has upscale 4X.
But PixAI got their own Pixai Real v2 model which I can only generate 3 upscaled images every day.

The upscaling method in both apps are different too.
PixAI can make the girl prettier (and look different) after upscaling to 1.3, while the one from PW looks the same after upscaling.

This one is from PW.

 

[Elnoxv]

interesting.. is it worth paying extra for the gpu to play with stable diffusion?

Not just for stable diffusion. The AI community is booming, there are many others AI model that’s pretty exciting to play with.

e.g, alpaca Lora and llama, GPT3 like that can be used into consumer machine. (7B and 14B 4bits)

It’s worth the investment if you play with AI models.

 

[ygsub41]

Thought school holiday over alrdy?????

 

[hwarzoner]

Interesting very nice

 

[doogyhatts]

Trying out the Grapefruit anime model on PW.

 

[zheng]

Trying out the Grapefruit anime model on PW.

what is PW?