nemu , SFW

https://redd.it/1p4xrzc
@rStableDiffusion

Qwen Image -> Controlnet -> SDXL: Killer combo?

https://redd.it/1p54z0k
@rStableDiffusion

Neon Genesis Evangelion LoRa - Flux1 Dev

https://redd.it/1p5grs3
@rStableDiffusion

"Prison City" Short AI Film (Wan22 I2V ComfyUI)
https://youtu.be/DqpOvi1ZOyk

https://redd.it/1p5j7ig
@rStableDiffusion

600k 1mp+ dataset

https://huggingface.co/datasets/opendiffusionai/cc12m-1mp\_plus-realistic

I previously posted some higher-resolution datasets, but they only got up to around 200k images.
I dug deeper, including 1mp (1024x1024 or greater) sized images from CC12M, and that brings up the image count to 600k.

Disclaimer: The quality is not as good as some of our hand-curated datasets. But... when you need large amounts of data, you have to make sacrifices sometimes. sigh.



https://redd.it/1p5kjuk
@rStableDiffusion

Is 16+8 VRAM and 32 GB of RAM enough for Wan 2.2?

Just bought a 5060 TI and will be using my 3060 TI in secondary slot.

So I have this question.

I don't really want to buy more RAM right now because I'm on an AM4, I was thinking of upgrading the whole system at the end of next year.

https://redd.it/1p5h1zp
@rStableDiffusion

Hunyuan 1.5 step distilled loras are out.

https://huggingface.co/Comfy-Org/HunyuanVideo_1.5_repackaged/tree/main/split_files/loras

Seems to work with T2V 720p model as well but obviously might be different results than using the 720p lora when that comes out. Using it with euler/beta 1 str, 1CFG 4-8 steps works.

I get gen times as low as (non-cold start after model is loaded and prompt is processed)

>6/6 [00:28<00:00, 4.81s/it]
>Prompt executed in 47.89 seconds

With a 3080 and the FP16 model, 49 frames 640*480, no sage or fast accumulation as the individual iterations are quite fast already and the vae decoding takes up a decent % of the time.

https://redd.it/1p5ou1s
@rStableDiffusion

Switching to Nvidia for SD

So right now I have a 6950xt (went AMD since I didn't really have AI in mind at the time.) and I was wanting to swap over to a Nvidia GPU to use Sable Diffusion. But I don't really know how much a performance bump I would get if I went budget and got something like a 3060 12gb. Right now I've been using one obsession to generate images and getting around 1.4it/s. I was also looking at getting a 5070 but am a little hesitant from the price (I'm broke).

https://redd.it/1p5pvu1
@rStableDiffusion