Stable diffusion thread

[doogyhatts]

Open Sora project has progressed to v1.2.
Need to rent either an A40 or an A100 then can use it to generate video at 720x1280 resolution for 4 second duration.
Otherwise, just use their Gradio demo and can input reference image!
It uses Colossal-AI, which only has Linux support.

 

[AikiBoy]

Open Sora project has progressed to v1.2.
Need to rent either an A40 or an A100 then can use it to generate video at 720x1280 resolution for 4 second duration.
Otherwise, just use their Gradio demo and can input reference image!
It uses Colossal-AI, which only has Linux support.

4 second.... enough for a youtube ad.

 

[doogyhatts]

As for Lumina, have to wait until they finish researching the 16-channel VAE that SD3 uses.
In the meantime, Kijai has provided a ComfyUI wrapper for it.
Maybe worth trying out their Next-SFT model first and the Gemma-2b LLM that it uses.

More images and prompts using Lumina.

 

[zheng]

comfyui oso gone case liao

 

[doogyhatts]

Open Sora HF gradio demo seldom works now. Other users reported that their initial outputs are not good.
The installation on a remote server instance requires NVidia Apex, which takes like 25 minutes to install.

Luma Dream Machine is in high demand so the free tier is queued all the way at the back.

Haiper discarded their 2 seconds animation and set a minimum of 4 seconds, of which is far more unstable than their 2 seconds one.

Pika is getting better at reducing blurriness. I discovered that I had to ignore the negative prompt in order to maintain correct colouring as the input image.

 

[zheng]

trying dream machine... wolf transform into a man... this is what i get

https://streamable.com/nuczki

 

[doogyhatts]

From discord, RunDiffusion team has begun training their dataset for PixArt-Sigma.

 

[AikiBoy]

today playing with openpose control net.
Can start AI onlyfan liao LOL

 

[doogyhatts]

 

[doogyhatts]

Diffutoon.
Conversion to anime is now a lot better!
Requires 22gb vram.

 

[doogyhatts]

In the meantime, the free tier for Luma Dream Machine is becoming unusable.
Only two concurrent generations can be in the queue and the free queue takes forever to move.

 

[AikiBoy]

Diffutoon.
Conversion to anime is now a lot better!
Requires 22gb vram.

is it me or most of these ppl who came out the articles are ... chinese

 

[doogyhatts]

is it me or most of these ppl who came out the articles are ... chinese

Majority of the human capital for research scientists and engineers for AI and ML are from China.
They are everywhere from the universities to the big-tech and start-up companies.

For example, Pika.art is founded by two Atbs whom dropped out of Stanford.
PixArt team is also from China.

 

[zheng]

is it me or most of these ppl who came out the articles are ... chinese

many Taiwanese and China oso

 

[AikiBoy]

Majority of the human capital for research scientists and engineers for AI and ML are from China.
They are everywhere from the universities to the big-tech and start-up companies.

For example, Pika.art is founded by two Atbs whom dropped out of Stanford.
PixArt team is also from China.

many Taiwanese and China oso

where all the amdk.



comfyui files so big. heng i got new hdd from amazon sales last time... else gg

 

[doogyhatts]

where all the amdk.

comfyui files so big. heng i got new hdd from amazon sales last time... else gg

AMDK went to Stability.ai, OpenAI, Anthropic and LumaLabs.

 

[AikiBoy]

AMDK went to Stability.ai, OpenAI, Anthropic and LumaLabs.

no wonder the model i generate, some look like tw actress ruby lin.

 

[zheng]

 

[doogyhatts]

Have to do inpainting for the sara dish on the top of her head.

 

[zheng]

Have to do inpainting for the sara dish on the top of her head.

pixart?