As someone who just started to explore this and also as someone who's using Automatic1111, this is very helpful. Want and hope to see more videos like this. Subbed ^^
@tr0ublem4kerWZ
Жыл бұрын
Very helpful. After using sd for almost a year, still learned 3 more tricks. Lovely! Keep the good content coming!
@ARHAT-No1
Жыл бұрын
Thank you for your effort. Very useful and interesting analyzes and tips that save a lot of time. I really like to watch. 👍
@jonevansauthor
10 ай бұрын
The notification sound alone has saved me a tonne of wasted time when you don't realise it finished generating - which is a real problem with an aged 1080 because it takes a while to do anything.
@sledgehog1
11 ай бұрын
Hi, for the models not being in your main storage I've been using an additional setting on the boot up settings and that will load the models from the given folder, for example from an external hard drive. I don't know if it also loads up the vae as I've been using models with embedded vaes, though... Also, I don't know if loading external models impacts the image generation speed.
@nodewizard
Жыл бұрын
Gooday fellow Thaumaturge. Can you make a video comparing some GPUs for SDXL. I have an RTX 3070 (8GB of vram), RTX 3080TI (16GB of vram), 4060TI (16GB of GDDR6 vram) and a 4080 (16GB of GDDR6X vram). I have found that the 4060TI and 4080TI zip through SDXL very fast along with the 3080TI. However, friends with a 4070 (12GB of GDDR6X vram) are struggling with SDXL, Refiners enabled and Hires Fix. Can you shed some light on these problems people are experiencing. Just how much vram does SDXL require on Auto1111 (not ComfyUI)? Thanks
@siliconthaumaturgy7593
Жыл бұрын
Too much VRAM required for SDXL on A1111 compared to Comfy. I couldn't get past 1800x1800 ish without VRAM errors with a 3090, which was a lot lower than I could do for SD 1.5. Though that was at SDXL launch. I haven’t tested with the recent A1111 update yet to see if they optimized it. 10GB VRAM is the suppose minimum for SDXL, so they should be able to run it, but with A1111 not being optimized, they might be stuck with Comfy for now As for hardware testing, I would love to, but unfortunately, I don't have a bunch of different GPUs to test.
@dawnstroup4698
11 ай бұрын
🙋 *PromoSM*
@puciohenzap891
4 ай бұрын
This is incredible stuff.
@SilverVon
Жыл бұрын
I have played with A1111 for couple months and only 2 tricks from this video that I was aware of. Thanks for this tutorial, very useful.
@Ikarusu
Жыл бұрын
hey, i had a question but what is the current method to create a similar effect of hi-res fix in img2img ? specifically how hi-res fix can add more detail to the image as the resolution grows.
@siliconthaumaturgy7593
Жыл бұрын
Only difference between img2img and hires fix is that img2img by default uses steps * denoising as the actual number of steps
@MrPicklesAndTea
Жыл бұрын
notification sound, and transparency for inpainting, are both useful. I might one day find continuous generation useful, as an alternative to batch generation.
@evanstastny964
Жыл бұрын
The cntrl+up/down trick (cmnd+up/down) is a neat little time saver. Also will be experimenting with the precision inpainting. That one looks like it could be fun.
Пікірлер: 19