FULL Perspective Control - ControlNET in Automatic 1111 - Stable Diffusion

[@masterjx]

Followed a hard guide on the internet to install stable diffusion and they didn't even go over xformers. Learned about it a week later and this little line of code literally sped up my renders by like 45% and I'm not joking. Some renders was taking 3 minutes (I use a lot of lora's) and this cut it down to like 1:30 minutes some even faster. Thank you!!!!

[@OlivioSarikas]

#### Links from the Video #### Join my Live Stream: kzbin.infopuxTMSqC1bc SketchFab Japanese Alley: sketchfab.com/3d-models/japanese-street-at-night-fb1bdcd71a5544d699379d2d13dd1171 Buy me a Coffee: www.buymeacoffee.com/oliviotutorials Join my Facebook Group: facebook.com/groups/theairevolution Joint my Discord Group: discord.gg/XKAk7GUzAW

[@santosic]

I can't believe I never thought of using Depth to create scenery in a certain perspective!! Wow. That is actually a really good use of that feature. I've spent a long time trying to get the camera in the right spot, and I could have just done it this way. Thanks for the clever tip!

[@martin-cheers]

I second that.

[@cekuhnen]

I work with blender and from the 3D scene extract the depth image Super useful

[@zoybean]

Woo! I asked if you could post something like this before and you did it, awesome! Thank you so much!

[@OlivioSarikas]

Thank you. Did you talk to me about that in my last live stream?

[@zoybean]

@@OlivioSarikas Yep, that was me!

[@jaredbeiswenger3766]

Wonderful tip. I've been trying to do this with sketches, but I'm excited to free myself from the tunnel background with buildings on either side stretching infinitely into the distance.

[@MarkDemarest]

FIRST, and I #CantWait to get into it! 💪🧠 -Thanks, Olivio!! 🎉

[@aggressiveaegyo7679]

It's incredible. I didn’t even think that you can just take a screenshot of any scene in a movie. Or a video game. Take a recognizable map and create your own version. I think that we all have already taken a picture of our apartment and played with interior design, for example, classicism or Victorian style.

[@nathanielblairofkew1082]

what you really need is a nested-type default slider. nesting sliders or somewhat similar will be absolutely neccessary to handle future complexities before they are simplified

[@bryan98pa]

So interesting video. The way we can control the weight and that's affect the final results is something that I learned today👍

[@OlivioSarikas]

Thank you :)

[@RetzyWilliams]

Wow, such a great idea. Awesome! 👏

[@Aristocle]

If I want to fill an empty room with furniture using this technique, but leaving the position of the fixtures and walls unchanged, how should I set up the problem? If I want you to suggest the random arrangement of the supplies (brainstorming).

[@dreamzdziner8484]

I knew those sliders in Controlnet could do wonders :-) Gr8 video mate 👌

[@temarket]

hey man, your channel rocks!

[@paulstrife]

Was there ever a technique to have the exact same scene from multiple angles? It is important for my needs to show the same objects. :(

[@nonameishere7234]

Thanks for sharing the tip. You're awesome ;)

[@Moedow]

How does the guidance parameter defines how long it’s gonna be used when the guidance start parameters does that already?

[@niteshghuge]

Hey Olivio, I am trying to install the AUTOMATIC1111 but issue is coming my laptop is getting hanged please tell me what is the system requirement to install the A1111 in local machine

[@coda514]

Great video, informative as always.

[@OlivioSarikas]

Thank you very much :)

[@audiogus2651]

Heck yah, great video! I have been using game screenshots for this sort of thing too, works great!

[@OlivioSarikas]

Great Idea! :) Thank you

[@aronhommer1942]

is it also possible to create an output that doesnt looke cartony with this method?

[@fluffsquirrel]

Maybe that's just the model he's using?

[@OlivioSarikas]

yes, of course. I just used anime here, because it is easier to prompt for

[@fluffsquirrel]

@@OlivioSarikas Thank you!

[@BlackMita]

What’s with the annotation result preview? I’ve never seen that in 1111 and I don’t get how it relates to the output

[@ryry9780]

Not all that different from a project I did before -- taking a picture of a person from the internet and turning it into an anime-style fanart. The only things that the original picture and the final product have in common are the pose of the character and the camera angle. ControlNet Depth and Canny had been very important, along with ControlNet Clip Image (style).

[@cekuhnen]

Oliver does controlNet generate a depth map ? I was under the impression that you need to supply it with it.

[@animestories5084]

Any way we can turn the image in different (like fill in the blank) angles and get consistency so it can be used for 3D scenes? For example: You take an image and keep it as a texture on the model, then move the angle a bit so that the depth map can read it a different way, but now (or still) you're still having an imgtoimg feature involved, which can be tested to stay texturally consistent. Of course, the UV maps will update, but you have now textures that can be used for 3D animation. Please reply to let me know lol.

[@entrypoint2009]

Using the Guest mode gives very good result.

[@OlivioSarikas]

Thank you, i will try that

[@pedrodeelizalde7812]

Hi, i have controlnet but my settings are different, for start there is no annotator result next to the image, also the prepocressor options there is three options : depht leres, depht midas and depht zoe, then below options are Control Weight, starting control step, ending control step, preprocessor resolution, remove near %, remove background %. Anyone know why my setings are not the same as him?

[@kikeluzi]

Same here... And also... I think it's not working :c

[@pedrodeelizalde7812]

@@kikeluzi It did work for me using depth leres. But i had to play with settings to get it...

[@kikeluzi]

@@pedrodeelizalde7812, I'll try this one then. Thank you!!! 😁I was using "depth_midas" * edit: I just needed to download a model first... ;u;

[@FikaBakilli]

Hello! As always, everything is top notch. For which you have a lot of respect. In the video you mentioned that the resulting images can be converted into 3D. Could you show us how all these pictures could be converted back into 3-D model. )))

[@OlivioSarikas]

Hi, thank you. I think you misunderstood me. I don't know how to turn them back into 3D. It can be done to a certain degree, by modeling the same scene, but only for a kind of zoom effect, not an actual 3D space, as far as i know

[@CHACHILLIE]

You can camera project them back onto the 3D model

[@user-jk9zr3sc5h]

I always wondered what those sliders would do-

[@petzme8910]

Can I add 1girl in the prompt? If I want the girl stand in the middle of the street 😊

[@OlivioSarikas]

yes, you can also use multi-controlnet to pose her exactly in a specific position

[@digitaltutorials1]

It's interesting because control net pulls depth out of 2D but with SD in blender the depth is 100% calculated so it's more accurate to use the SD plugin (which I haven't tested yet but I assume it is underdeveloped compared to auto's).

[@cobr3545]

Control Net uses a rendered depth map, same as blender. It's not estimated from 2D if you supply one.

[@cobr3545]

Control Net uses a rendered depth map, same as blender. It's not estimated from 2D if you supply one.

[@JulienTaillez]

what a brilliant idea !

[@OlivioSarikas]

Thank you :)

[@RiyadJaamour]

Hi Olivio is there still a way to use stable diffusion with all changes and additions in google colab? or even if there is an alternative to google colab other than local installation would be great! ty

[@elarcadenoah9000]

do u have a video with stable diffusion xl inseting text?

[@OlivioSarikas]

Nope, not yet. I could make one

[@elarcadenoah9000]

@@OlivioSarikas cool bro you are the chosen one

[@sb6934]

Thanks!

[@tarekramadan1867]

I have used kinda the same method but for interior architecture scene with images i generated i will send it

[@facex7x]

hey i dont have "depth" under model for control net. how did you get that?

[@OlivioSarikas]

maybe your controlnet version is outdated or you didn't download the depth model

[@The-Inner-Self]

Have you experimented with video using the 3d models to walk through the scene and then plug into depth control map?

[@stillfangirlingtoday1468]

This is probably a stupid question, but do you think it's possible for the AI to generate the same image but will different lighting? It would be awesome.

[@OlivioSarikas]

try rendering the scene first, then using canny as the controlmap with that image input and change the daylight description. See if that helps

[@stillfangirlingtoday1468]

@@OlivioSarikas Oh, I will definitely try it! Thank you for replying!

[@NasserQahtani]

كم انت جميل في طرحك

[@sirmalof3255]

Where are you from Olivio. I can just listen to your accent for hours and hours. It is so cute :)

[@OlivioSarikas]

Thank you. I'm from Vienna. Well, i'm from Germany, but i live in Vienna :)

[@sakifishmam1436]

It's now banned anyone know any alternatives?

[@kikeluzi]

I didn't find anything about that. ControlNET was banned? 🤔 how do you know and where to find?

[@michail_777]

Hi Olivio. I don't remember the name, but Stable Diffusion has an Expansion that bends the image and creates a corridor. You can create streets and whatever you want.But I have another question. If you know how to achieve stability when you create animation with ControlNet script (img2img) or when you process batch at once. Just the face can already be stabilized, but the clothes, they are always different. It's kind of similar, but not. If you know how and what to do.Show us, please.

[@smortonmedia]

I wish MidJourney could do something like this... generate depth maps or use a depth map as perspective control

[@audiogus2651]

It used to come up quite bit in the weekly chats a few months ago and they did say it was being looked into.

[@OlivioSarikas]

Yes, MJ really needs stuff like that ASAP

[@Rasukix]

also worth mentioning you used a different seed for each render

[@Hazzel31337]

meta has a new ai called segment anything, goes nicely with ai generated images to cut out elements, maybe worth content for you

[@OlivioSarikas]

I will have a look at that. thank you

[@LouisGedo]

👋

[@HarryJPotter]

cleaver girl....

[@User-eg4ws]

so you basically stole an art for a sample? Aight

[@dezenho]

do you lost your time with ai now ....life time not worth ....you are old ....control it ....you are not happy with this.... pay atencion what your life is now.....

[@Thagnoth]

Yes, instructor… I shall abandon all my enjoyment of technology… Thank you for showing me your text-only Amish hypnosis technique……