any generator on web ui ?

[ashen moat]

im so noob installing stuff on my old-end computer 😦

[pine breach]

If your computer is old then it probably won't run the webui very well.
You can use a free google colab instance instead. Here's the one that most people around here use:
https://colab.research.google.com/drive/1wEa-tS10h4LlDykd87TF5zzpXIIQoCmq

[ashen moat]

ty so much ❤️

[versed moss]

yeah, unless you have one of those pcs with multiple SLI gpus, doesnt seem a good option to run SD locally

and with SLI, dont know if deepspeed is implemented

[mint lodge]

huh? isnt sli for linking multiple gpus?

from what I know you can fun automatic1111 with 4gig vram and 8gig ram, although at a slow speed. I think it would be better than having to launch google colab every time right?

[pine breach]

The colab T4s are actually like 2060 tier in terms of it/s and the extra vram isn't that useful unless you're using controlnet in which case it's a bit faster. Local SD is still better or equivalent on midrange hardware for everything but training.

[obsidian wraith]

Without training actually 3050ti is enough right ?

local way

[versed moss]

if its above 6GB vram, you're good

4-6GB have issues but work

[obsidian wraith]

Everythings is okay until I train

😄

I can get 2 4 6 8K İmages

Also fast generate but

cant do train

I can handle this on colab

Thanks for ur respond

[versed moss]

training needs 8+

some reports of LoRA on 6 GB

[pine breach]

Lora works fine on 6GB on 20 series and above. I'm not sure about 10 series because of xformers

[versed moss]

16 series is fucked

[pine breach]

Absolutely

[versed moss]

there is no way nvidia made a gpu architecture that bad, twice

[obsidian wraith]

3050Tİ 4GBVRAM ? Could handle lora ? training locally

[pine breach]

you could try but if it works it will be right on the edge. Colab makes far more sense

[obsidian wraith]

Okay sir thanks for information ❤️

[mossy drift]

The now broken web UI I used to use used https://huggingface.co/hakurei/waifu-diffusion-v1-3/resolve/main/wd-v1-3-float16.ckpt as the model... Using the web UI @pine breach posted at the head of this discussion, which model will produce brighter / more anime stylings? (The default one is kind of flat)

[versed moss]

you probably confused webui with model

but right now, the best WD model is this one:

-rs wd

[mild ridge]

Waifu Diffusion

Current models can be found here:
https://huggingface.co/waifu-diffusion/wd-1-5-beta3
Waifu Diffusion 1.5 is a Stable Diffusion 2.1 model, it might be different from what you're used to, so make sure to give the release notes a read!