Stable diffusion thread

[BlackWing1977]

https://community.amd.com/t5/ai/int...-with-amd-xdna-super-resolution-a/ba-p/697374

https://www.amuse-ai.com/

Got people try this??

 

[doogyhatts]

Testing character head motion with speaking animation in landscape format.
Hedra + DynamiCrafter + LivePortrait


Anyone can do game similar to Star Control 3 liao.

 

[doogyhatts]

Nvidia research has solved the problem with character inconsistencies for text-to-image.

ConsiStory: Training-Free Consistent Text-to-Image Generation​

https://research.nvidia.com/labs/par/consistory/

It turns out it was 6 months ago that was released with no code.

 

[doogyhatts]

Virtual tabletop game editor.

 

[doogyhatts]

New image model called Flux by the original team for Stable Diffusion.
12B parameters size. Need min 16gb vram. Recommended 24gb vram.
Midjourney quality.
HF demo.



I tried the HF demo, it couldn't do the same quality for the monsters that I had been generating using PixArt-Sigma.

 

[doogyhatts]

Someone actually got it to work with 8gb vram as well, but need 32gb sys ram.

 

[doogyhatts]

The hands and feet came out almost perfect!

 

[doogyhatts]

More comparisons.

 

[doogyhatts]

Black Forest Labs is also working on a video model.
https://blackforestlabs.ai/wp-content/uploads/2024/08/tv_no_screen_3.mp4

 

[focus1974]

Someone actually got it to work with 8gb vram as well, but need 32gb sys ram.

wa.. thanks for the link
i got 8gb vram and 64gb sys ram.
Can try.
but does it do NSFW asians?

 

[doogyhatts]

wa.. thanks for the link
i got 8gb vram and 64gb sys ram.
Can try.
but does it do NSFW asians?

I think it can, but the results are likely to be undertrained.

 

[focus1974]

I think it can, but the results are likely to be undertrained.

ok thanks. then not worth the time and effort to setup

 

[BlackWing1977]

@doogyhatts What do you think of the Amuse 2.0 Beta... ??

Finally downloaded and installed it... omg.. excessive moderation on generated images despite it is all local... nonsense... now waiting for a more simple solution to get rid of it...

 

[doogyhatts]

@doogyhatts What do you think of the Amuse 2.0 Beta... ??

Finally downloaded and installed it... omg.. excessive moderation on generated images despite it is all local... nonsense... now waiting for a more simple solution to get rid of it...

I didn't try Amuse 2. Got new models such as Flux anyway.

I am very comfortable with PixArt-Sigma right now because it can generate monsters, which passed YT censorship.
YT algorithm doesn't like AI girls, but actively recommends videos with AI monsters on both the homepage and the up-next sections.

 

[BlackWing1977]

I didn't try Amuse 2. Got new models such as Flux anyway.

I am very comfortable with PixArt-Sigma right now because it can generate monsters, which passed YT censorship.
YT algorithm doesn't like AI girls, but actively recommends videos with AI monsters on both the homepage and the up-next sections.

I not posting to anywhere lar.. is ownself see ownself song nia...

 

[doogyhatts]

I not posting to anywhere lar.. is ownself see ownself song nia...

I see.
Now I know what Amuse 2 is.
https://community.amd.com/t5/ai/int...-with-amd-xdna-super-resolution-a/ba-p/697374

You are using an AMD hardware.

 

[BlackWing1977]

I see.
Now I know what Amuse 2 is.
https://community.amd.com/t5/ai/int...-with-amd-xdna-super-resolution-a/ba-p/697374

You are using an AMD hardware.

Yeah using AMD hardware but I dun have the new NPU AI thing so cannot use super resolution, I only have the GPU.

It will from time to time prompt you to download certain model if the generation requires.. I think got download juggernaut XL then later got SDXL or something because I copied the prompts from some AI image websites.. but results very different though as those websites uses multiple models...

 

[doogyhatts]

Yeah using AMD hardware but I dun have the new NPU AI thing so cannot use super resolution, I only have the GPU.

It will from time to time prompt you to download certain model if the generation requires.. I think got download juggernaut XL then later got SDXL or something because I copied the prompts from some AI image websites.. but results very different though as those websites uses multiple models...

You can just use Stability Matrix and install the ComfyUI package. Very easy to setup.
If you are on windows, there is support for DirectML.

 

[doogyhatts]

ok thanks. then not worth the time and effort to setup

Can try using the HF demo. The dev one is better than the schnell one.
https://huggingface.co/spaces/black-forest-labs/FLUX.1-dev

 

[focus1974]

Can try using the HF demo. The dev one is better than the schnell one.
https://huggingface.co/spaces/black-forest-labs/FLUX.1-dev

not bad.
i tried "a chinese female model seductive look fashion"

i think the result is more or less quite similar to sdxl.

but even sd1.5 can do it if you setup peturbed attention and self-attention guidance together (not including the text part. that one sd1.5 cannot do).


now which webui can run locally using GTX 4060?
i am still using sd forge coz i dont like comfyui.
swarm ui is also not very userfriendly in my opinon.