Відео

use s key to zoom

Did what you said, minus the xformers, since I'm funning on CPU and have gotten a bit of improvement.

One thing that sped my sdxl up was switching from my high res external monitor to output only to my laptop screen which is lower resolution. Or i guess if i lowered the resolution on my external monitor that would help too. Laptop rtx 4070.

awful technique

I can't draw on the image (upper right).

tnx alot, make more videos :D

Doesn't work on ForgeUI+Flux.

What's the difference between inpaint and inpaint sketch?

wow! I only have 2GB VRAM and this tip really made a difference on my Stable Diffusin WebUI Forge performance ! Thanks a lot !

cool, finally i dont need to wait 20+ minutes

do you have a link to the anime model used at 10:28? I searched citusMIX but cant find anything.

I use this utility ALL the time, here's a few tips for the highest quality you can achieve in the end results if you wanna be a try hard like me (depending on your GPU strength, you'll need to adjust. I have a 4090): TL:DR: I know this isn't practical for most people, but if you're reading this and like me, you want the BEST possible quality in the end result and you got the gear, this is how. 1. 128x128 Latent tile width/height settings with a 64 latent tile overlap (the suggestion of 8 in the video is probably more performant) and use the other recommended settings. 2. Set an xyz script with denoising and in the parameters put 0.1-0.5(+0.05). This will generate 9 different images and takes a while, even on my 4090 it's around 8-10 minutes. You can probably just do 0.1-0.5(+0.1) and be fine with 5 images for the rest. 3. The reason you do the previous step is because each level of denoise will produce a slightly different picture. Some with better eyes, some with better clothes, etc. 4. Take the lowest denoise photo and put it into a photo editor of your choice as a base(I use Photopea). 5. Import the other photos and set them as a raster mask layer and hide all. 6. Use a brush with black and white colors set and a lowest hardness setting to start blending in different layers. This will allow you to take the best from every single gen and combine them into a completely flawless end product. 7. Afterwards you can adjust color balancing, hue, contrast, and brightness settings to clean up the picture. Many pieces tend to gen with a slight red or green hue that needs to be fixed. 8. If you do this and it needs a lot of editing, it can introduce what's called color banding. To fix this, export your final result and import it into img2img. Turn off TiledDiffusion/VAE, set denoise to 0, turn off xyz script, set scale to 1, and re-gen. This will smooth out any artifacts introduced by editing but maintain the original picture at original size. Also, if you DO have high end gear, turn off Fast Encoder/Decoder. If you zoom in on a final product, it appears as though the entire gen was made with tiny little dots. If you need the performance advantage of the coders, you can fix this issue afterwards with step number 8 and it will remove them.

When I enable region prompt control, I always get this error "TypeError: expected Tensor as element 0 in argument 0, but got DictWithShape". It doesn't generate image, but when I uncheck it, it works - ofc without the region control.... Any idea how to fix that? Thank you

I wished I had watched your tutorial before going down the rabbit hole 😅 thank you so much for the great tutorial. Perhaps it would make your videos a bit longer, quick explanations for the terms like VAE, textual inversions, etc. could make this seem more approachable imho

Great video. Whats your opinion about ComfyUI, I am curious why you never make any videos on it.

oh no,i don't konw why my computer can't download the inpaint model id,i followed your video properly😢,i need help😢

What if I’m getting a “package modification failed”? Please help.

if you have multiple areas that need to be fixed, can you use latent upscaling and then downscale the image so it's more workable, and then use latent upscaling again?

is this channel abandoned?

Thanks for the video! Do you know if there is a way to train your own models and use them with this program? I mean like uploading 100 pics of a particular person's face to get a model. Can you do that with this?

Thank you very much sir

Such an amazing video! I just started with SD 2 weeks ago, and this video is what made my images reach a much higher quality level. Thanks!

Excellent guide, you made it so easy to understand! Thanks!

Sir, I have a important request. How we can re-design from a low q. To high. No upscale way. I want AI can known the objects and re-design it on high quality, I would to big thanks

man i searched everywhere to find out how to get the brush color toggle for inpainting and only this video explained how to manually get the extension installed by going into the available tab. it says that the canvas zoom is "built-in" in my A1111 but it doesnt update and it didnt came with the brush color tool. this video really helped, thanks so much bro

Thank you ❤, it's work perfect

If i want to add a Lora like "add more details" and "Sharpness tweaker" will it work on this tile diffusion method?

Thank you for this tutorial it was very useful because you showed things step by step, and didnt skip anything. Only problem I have, is that I cant find anything else than the inpainting model I put in ControlNet in the models folder, is that normal?

Tried going to the hugging face page but the file is gone. Or at least if it's there it's very different than the one you showed.

Trying to do methord 2 but I get botched results. Think this needs an updated guide, that or my setup requires different parameters

Did things change or do I need to do something else? I have downloaded everything and put it into the folder but the pre processor doesnt have the one I downloaded. It only has auto, clip_h, pulid, face_id_plus, face_id, clip_sdxl_plus_vith and clip_g

13:20 this advanced part was very helpful. I always struggled with preventing artifacts.

When doing this, Inpaint just stretches the imagine. Any idea what I'm doing wrong?

This is great, I tried it myself and of course screwed something up. When I find a background generation I like, I then try to add chracters to the background generation, while keeping the same seed number, but the background changes even when keep the same seed number. Not sure how to proceed without relying on rng. Any thoughts?

is there a video on installing the segment-anything model ?

There is no "SEARCH" option for Extensions.

It doesn't work with pony or XL modelssssssss, can you use your alchemy to make it work?

太棒了！我昨天遇到了这个问题，无限延长的手臂，感谢您的视频！

Really ty again !

Does anyone know why im getting little bumby circles artifacts on upscaled images when using? using method 1. EDIT: Its something with FAST DECODER, with it off I get slower gen but I must

You say "I like to keep my extensions up to date all time", so just add : for /D %%I in (extensions\*) do git -C "%%I" pull at the start webui-user.bat. That will update all start, no need to restart :)

HI how do you generate the heatmap?

This video is a Gem, well done Sir 🔥

FYI, After Detailer now is called ADetailer if you cannot find it in the extension list

so nowadays, is it normal for a 25-30sec generation SDXL image on a 3080 10gb? I was generating @ 720x1280 cheers.

thanks teach

this video is gold standard for anyone starting with automatic1111

It worked at first then all of a sudden it just stopped doing anything for the eyes. I can detect the eyes but it's like it doesn't even change them after running the adetailer. The eyes are the same afterwards, anyone run into this issue?

THis is insanely helpful