Consistent Faces in Stable Diffusion

Learn how to generate consistent faces using Stable Diffusion. Using the power of stable diffusion's controlnet you can use the same face over and over. Whether it be for realistic images or cartoon style characters.
Want to advance your ai Animation skills? Checkout my Patreon:
/ sebastiantorresvfx
www.sebastiantorresvfx.com/dow...
Links from the Video ###
RealisticVision Model:civitai.com/models/4201
Random Name Generator: www.behindthename.com/random/

[@sharonryanabel]

Honestly this is the best tutorial I’ve seen for roop and consistent faces ! , thank you for the clear concise and high quality video !

[@sebastiantorresvfx]

😁 I’m really happy to hear that, sometimes I wonder if I’m been clear enough since there’s so much information to deliver on any given topic.

[@HisWorkman]

Thank you so much, this is probably one of the most useful SD tutorials I have watched in a very long time!

[@sebastiantorresvfx]

Thank you 🙂 that’s my aim.

[@siltoruz3502]

Very nice techniques showcased!! Interesting and unique approach to this problem that i had not seen in any video before.

[@sebastiantorresvfx]

Thank you @siltoruz3502, I appreciate the kind word. I’m trying not to tread on anyone else’s videos, unfortunately it’ll probably happen eventually 😂

[@maDM72]

Best tutorial I’ve seen so far. So straight to the point and in depth with details! Do you mind making video how to setup the environment that you are using?

[@sebastiantorresvfx]

Thank you 😁 Do you mean a tutorial on how to install automatic 1111?

[@Elwaves2925]

It's good to see other recommending this way of doing things. I've been telling folks about it for awhile, after I accidentally stumbled on it when I spelt a celeb name wrong, then wondered why each subsequent generation was wrong but the same. Then I used Roop after that once the penny dropped.

[@sebastiantorresvfx]

Happy accidents are a great way of learning new tricks! Can’t be afraid of goofing up, I’ve most of my unconventional tricks are from moments I goofed and something Great came out of it so it lead me down new paths. Which is why I always recommend, learn what I teach then go experiment. 😉

[@kritikusi-666]

nice guide. No BS, straight to the point. Thanks for the reference guide btw. I learned something new today.

[@sebastiantorresvfx]

Fantastic! 😃

[@kritikusi-666]

@@sebastiantorresvfx I tried doing portrait shots of myself, but the eyes...I cannot get to stay on par, they somewhat look fake. Be cool to see some future tutorial on how to achieve portraits.

[@sebastiantorresvfx]

There’s a great Lora model called perfect eyes. Use it with inpainting for best results. I’ll put out a short covering this on the weekend because it’s an amazing tool that’s helped me

[@jean-francoiscartier658]

Excellent! I would definitively like to watch a tutorial for a cartoon as well :) Thanks again

[@JayC216]

Quick, simple and strait to the point 👍

[@jimli3604]

this is what I'm looking for, thanks man.

[@sebastiantorresvfx]

Glad to hear that 👍🙂

[@ChristianIce]

Never thought about the random generated name. Makes sense! Perhaps you can improve it by selecting the best and most consistent results and cooking up a Lora using the same name.

[@sebastiantorresvfx]

Exactly, makes the process of elimination much more efficient when working on LORA data sets

[@stickman-1]

Mind Blown.

[@op12studio]

in my experience faceswaplab tends to do better than roop so if anyone wants an alternative give it a shot. Great tutorial regardless! Awesome job

[@sebastiantorresvfx]

Yes I’ve been looking into this might need to do an update soon 😉

[@DanielSchweinert]

Thank you! Great idea!

[@sebastiantorresvfx]

Thank you Daniel 😀

[@pongtrometer]

Excellent overview and technique, cheers

[@sebastiantorresvfx]

Thank you 😊

[@tr0ublem4kerWZ]

Love the tip! Keep it up the good work!

[@sebastiantorresvfx]

Thank you @tr0ublem4kerWZ, your support is much appreciated 😃

[@mada_faka]

THank you for best tutorial sd, you amazing sir, hype for next tutorial

[@sebastiantorresvfx]

Appreciate the kind words, I’ll keep the videos coming out 😁

[@lqdlqd1385]

Very good tutorial. Thank you :).

[@sebastiantorresvfx]

Much appreciated 😃

[@redslashed]

Thank you gonna try afterwards.

[@sebastiantorresvfx]

Let me know how you go. 😃

[@jamesalxl3636]

this helped me so much thank you

[@sebastiantorresvfx]

Glad to hear that 😁

[@carlosnascimento1155]

excellent tutorial! original and innovative! congratulations! Could you do something similar for comics?

[@sebastiantorresvfx]

That’s definitely on the table!

[@nhituenguyen8848]

Thank you, bro, you are the genius

[@sebastiantorresvfx]

Thank you 😁 if you enjoyed it, that makes my day!

[@thekotfather]

easy! thanx man!

[@sebastiantorresvfx]

You're welcome! 😁

[@DericKVP]

Bro thank u sooooooo much I work's bro I look up so much tutorial and this is the only one that work

[@sebastiantorresvfx]

Excellent ! Glad I could help! 😃

[@johandc1710]

Great video. Will you be doing any tutorials that are for complete beginners in stable diffusion?

[@sebastiantorresvfx]

Sure will, stay tuned. 😃

[@ParvathyKapoor]

Great tip! any idea for custom face expressions?

[@sebastiantorresvfx]

Expressions are limited by the model or Loras you use. Controlnet can only take you so far, if the SD model doesn’t know what a specific expression is meant to look like it’ll try it’s best but doesn’t always have great results.

[@ParvathyKapoor]

can u do a video with control net expressions? coz for the pose is easy. but face expressions cant @@sebastiantorresvfx

[@sebastiantorresvfx]

it’s something I’m considering for sure

[@brandonb9267]

Is it possible to create the face grid from a stable diffusion portrait I’ve already generated? I’ve tried messing around with img2img and reusing the original prompt but the faces on the grid was drastically different.

[@CrhystyannDavidBernalChaves]

Muchas gracias por este tema, es posible ver algo asi pero con comics?

[@sebastiantorresvfx]

Te aseguro que cómics me ha inspirado mucho a aprender stable diffusion, así que seguro que pronto me meteré en eso.

[@ViratxDoodle]

Thanks for this, which SD model do you prefer when creating cartoon characters?

[@sebastiantorresvfx]

The answer is a long winded one that I’ll touch base on soon. I’ll be releasing a similar video but for cartoon/comic characters.

[@ViratxDoodle]

@@sebastiantorresvfx Thanks, you've read my mind. :) A video would be very helpful.

[@ImSimpai]

Is there a video for different positions and povs ?

[@sb6934]

Thanks!

[@sebastiantorresvfx]

You’re welcome 😁

[@alexmehler6765]

thats pretty cool , now we need consistent cloths and accessories for comics, storyboards, animation , whatever

[@sebastiantorresvfx]

We’re getting there soon. The latest video on my channel begins covering this process and the next couple will continue the method. Good to see I’m not the only one who’s eager to see how ai can help produce comics. 👍

[@alexmehler6765]

@@sebastiantorresvfx sure, ai is here to stay now, and i think there are many people who might be excelelent storietellers with interesting content, way different from the general comic and manga categories. a comic artist might be hindered by the sheer amount of drawing labour and can not take the time to build better plottwists and themes in general. so it can benefit versed comic artists as well and inspire people from other fields (writers mainly) to create a visual peace

[@sebastiantorresvfx]

Couldn’t have said it better myself. I know a lot of artists are hesitant at the moment believing that large companies like marvel and dc will adopt it to cut costs. My point of view is the indie artist who now takes three to six months to produce one single comic on their own being able to compete on the same level as the company’s who release one or even two comics of a title per month.

[@_Shimona]

Hey, I'm unable to sign up for you newsletter. I only see two tabs, "About" and "videos"?

[@jason-sk9oi]

Brilliant 👍👍

[@sebastiantorresvfx]

Thanks Jason 🙂

[@haris7255]

Thank you so much. Could you show how to generate different expressions?

[@sebastiantorresvfx]

Expressions are dependent on what the model has been trained on. There’s a few tricks that one can use with the help of control net and reference images of the expression you’re after.

[@ediraven]

Hey Man! I justr found out your channel, You earn 1 new suscriber. Thanks for the tuto for roop. Disculpa por preguntar, vi tu nombre y parece latino. Hablas español? Saludos desde Mexico!

[@sebastiantorresvfx]

Hola Ed, soy argentino y uruguayo, pero vivo en Australia. Me encanta que tengo más suscriptores hispanohablantes.

[@ediraven]

@@sebastiantorresvfx Excelente, Seb! Estoy viendo más video en tu canal. Gracias y saludos hasta Australia!

[@yesaultube2810]

Thanks for the tutorial Sebastian. Beside few hours I played with simple things, I have 0 knowledge in both stable Diffusion and ControlNet Canny. I'm trying to follow you guide to use reference image to generate multiple perspectives. And it doesn't work. I see a gap in the video @ around 5:13 where suddenly 9 separate images appear. But what exactly when before that is unclear. Could you advise on exact settings of ControlNet Canny, and why it will not generate 9 images as your demo shows. Thanks.

[@tylerschmale2139]

did you ever figure this out? same issue, one face generates with artifacts from the controlnet left over rather than 9 separate ones

[@VirselClemend]

a extra tip: use [(Name1|Name2):(Reference_person:1.32):0.5]:1.2 and with that u can create originals characters

[@sebastiantorresvfx]

Love it! Awesome tip!

[@ghostbusterz]

Why 2 names? Is reference_person an embedding?

[@bennyrawood]

Good info on the name generator. Thanks. I can't get 'roop' to appear however. Also, where do I get the 'canny' model? 'Canny' shows up as a preprocessor but in models to the right there is 'none'. Also the 9 grid anatomy poses do not show up on a generated image. I read all the comments and I saw the possible roop work around. Other than that, I couldn't follow your tutorial all the way. Hopefully you make another one soon!

[@sebastiantorresvfx]

Quick google search for controlnet uncanny models with bring up their huggingface downloads. I’ll be putting out a video on installing automatic 1111 and the important extension soon.

[@FantasyArtworkAI]

I'm new and a bit confused at this, was the grid of poses just to test that the portrait was good? Or just to create that portrait you used in Roop later?

[@sebastiantorresvfx]

I create the grid to have the face from all Angles: that way if I need to replace the face at a certain angle I can switch out the corresponding angle or the closest one.

[@FantasyArtworkAI]

@@sebastiantorresvfxThat makes sense, thanks!

[@sebastiantorresvfx]

You’re welcome 😀

[@NE0561]

Hi Sebastian! These programs you use are only for apple?

[@sebastiantorresvfx]

Actually I use these on Windows. I’m not sure what the process or comparability is with apple.

[@marschantescorcio1778]

Thanks for the tips. When you're generating your first batch of fictional portraits have you tried prompting with a specific hairstyle? I usually run a few batches pulling from a wildcard text file to find a style I like most. Guess the only caveat is that you'd have to use the same hairstyle in future prompts up until the point you decide to train a Lora.

[@sebastiantorresvfx]

You can have different hairstyles when you train your Lora you just need to specify that’s one of the things that can be altered about the character with the descriptions when training. It just depends if your character will have different styles or not: usually for a cartoon character or comic book character you’ll end up using the same style in every shot so helps to only train it with one look so the model doesn’t confuse what it’s meant to reproduce

[@lilillllii246]

Is it possible to apply clothes and have them look exactly the same when they are slightly different?

[@sebastiantorresvfx]

Clothing is a bit difficult, you can train a Lora with a specific item of clothing but it’s quite a bit of work. Aitrepreneur has a great tutorial on just that.

[@cbccbd]

Also for the names you could use extension Clone Cleaner

[@sebastiantorresvfx]

Haven’t used it yet, thank you for the recommendation I’ll check it out.

[@ChloeLollyPops]

I love the video, thank you. I'd like to find a discord channel to work with others on these type of projects.

[@sebastiantorresvfx]

There’s quite a few around, it’s a really booming community 😃

[@Rafaelgodoyebert]

After installing roop, even after restarting the SD, the option does not appear, any idea?

[@svensebastianhorner]

Is there a way to do the second approach with a img2img prompt? The goal would be to get the 9 different angles of a character already created in one image, so it can be further defined and refined from all sides as a reference. I tried the steps, but the result seems to be influenced by the face_grid.png beyond the perspective, so the character is changed completely.

[@sebastiantorresvfx]

Another way I sometimes use is to make a picture with multiple angles of a model or actor. Run it through an open pose controlnet only. As long as you have a name and distinctive descriptions in your prompt, they should all come out looking like the same person from different angles.

[@KenapciCeviri]

Hi! Great content. I have a problem that I changing the name but keep getting the same face everytime in grid. I did exactly what you did but everytime same result.

[@sebastiantorresvfx]

Did you reset the seed to -1?

[@KenapciCeviri]

Never changed it @@sebastiantorresvfx

[@mrskull3607]

great tuto, thanx man, any body how have issue with roop, i did install it but its not showing the extension in the UI

[@sebastiantorresvfx]

Sometimes it can be a case of restarting the ui. If that doesn’t work make sure your copy of a1111 is up to date. Delete the extension manually and reinstall

[@bettymac8248]

Anyone else not seeing the Roop controls after you have Applied and restarted the UI??

[@StreetArtistsOfTheWorld]

Hey! Good video! I'm a noob at this so this is great! Can you have a more detailed video of comic creation? That part went really quick :(

[@sebastiantorresvfx]

I will have soon. 😃

[@StreetArtistsOfTheWorld]

​ @sebastiantorresvfx Yay!

[@svensebastianhorner]

Using ControlNet v1.1.426 with the face_grid.png from your website and your instructions, I don't get the 9 faces. Sometimes it's 2, sometimes 3. But they don't align with the ControlNet Image at all. I don't know what I'm doing wrong. Tried v4.0, v5.1 and v6.0 Realistic Vision inpaint models

[@sebastiantorresvfx]

This means your controlnet model isn’t engaging. If you go back through the control panel logs for that specific generation it’ll tell you if the canny model, (or lineart,softedge) engaged properly or not. If they didn’t it’s usually a compatibility issue.

[@ClaudioMalagrino]

Hi. I'm doing a graphic novel based on my books. I still have a 3D animation. My main challenge here is to keep the consistency between planes. I have some interesting concepts generated but I need the same environment and characters over the different planes of composition. I've been trying to work using layers, but I'm not sure if it works. Could you give me a hint here? Thanks a lot!

[@sebastiantorresvfx]

Okay so I’m gonna assume you have 3D experience. In this case I would generate a bunch of images of the environment at different angles and project them onto some simple geometry. This way you can just render out the backgrounds isolated from your character and combine the two in post. This way you can get any angle you want and it won’t change from panel to panel. I should probably make this part of my comics series I have in the works. The first ones coming out soon. Very soon! 😁

[@ClaudioMalagrino]

Thank you,@@sebastiantorresvfx

[@angloland4539]

❤

[@algotrader9024]

Do I need to have a powerful GPU to generate images? Why not use dreamstudio instead? Thank you ..

[@sebastiantorresvfx]

I think you should use what ever you have available to you. If I didn’t have a powerful GPU I’d be using online GPU servers. As for why use a1111 and comfyUI. There’s just more freedom using local installs.

[@johnclapperton8211]

Great content :) For me it would be easier to see with dark mode for the UI. Also I couldn't see the prompt, I don't think you said what it was; was it just the random person name?!

[@sebastiantorresvfx]

Hahaha I know, I try to keep the UI looking straight out of the box as much as possible so newbies don’t get confused. Which specific prompt are you after? I can post it on here for you. Just let me know the timecode.

[@johnclapperton8211]

@@sebastiantorresvfx The first prompt, right after the demo of the random name generator; I can see it was short, was it just the name?

[@sebastiantorresvfx]

RAW Photo, Véssica Marise Kool, Young Woman, beautiful, brown hair, 8k, Just change the description to what best describes your character along with the name

[@user-uf1di5ze1z]

do i need visual studio for this? cause roop doesnt show up after installation

[@sebastiantorresvfx]

I suggest having it installed anyhow. Later on today I’ll be releasing my new video and the link to visual will be in the description of the video if you can’t find it by then.

[@PranavGogwekar]

Can I do the same with ai sites like nightcafe, fotor, dall e etc?

[@sebastiantorresvfx]

Not sure what their process for this kind of thing is to be honest. I believe MJ has a way to do it but the others I wouldn’t be able to say.

[@ninatko]

Hi, where is this GUI from, how can I set it up on my localhost? Thank you

[@sebastiantorresvfx]

Yes I’ve been asked this a couple time actually. While automatic 1111 is a pretty quick and easy install I think I should show that process too. Stand by 😉

[@streamtabulous]

wish i could get this to work in easy diffusion, mind you i though watching this you take the 10 face angles and use kohya ss to make a lora of the face, they you could just use that anytime you wanted. only catch is just head shots it only works on close up prompt, but with a few of the standing, sitting etc you did, if you put them in also say 20 images total you have a lora that you could use any time you wanted. i was hoping this would work in Easy Diffusion for that purpose to make lora models in kohya from the generated images

[@sebastiantorresvfx]

Yeah unfortunately I’ve tried to make Loras just from head shots it was just too limited in what the Lora could reproduce. This method in the video is more of a precursor so you can produce the different angles and shot types at varying distances so your Lora can be more well rounded when trained. I hope that makes sense?

[@streamtabulous]

@@sebastiantorresvfx yer does I'm new but learning fast, i found you need full body shots high detailed and found not to got over 640x1280 portrait for anything head shot distance etc, bigger it is a hot mess. So lora from distance, seems better if you do multiple lora, so one for close up only, another for full body, another from back where head turned to face backward and load the one needed for the prompt. Making one lora with all dose not really work its a mess I found. So lora for what you need get better results. over all with how new this stuff is im blown away by it all.

[@joespantosso]

Canny doesn't work for me. I installed controllnet, but the model doesn't load and the shapes aren't being applied. Someone got a fix for my problem?

[@sebastiantorresvfx]

Check the control panel and see if it’s properly engaging when you run the generation with Canny activated. If it isn’t then it could be a matter of the canny model being used. Unfortunately there’s so many at the moment it’s getting confusing

[@ALDUIINN]

which is the control net you are using ? i didn't said

[@sebastiantorresvfx]

You can any of the canny controlnet models for the grid. Just make sure you’re using 1.5 models with 1.5 checkpoints and/or SDXL models with SDXL checkpoints or they won’t work properly.

[@leonamnagel9999]

I installed Roop but it doesn't show up in my Automatic1111, I already have Visual Studio Community installed, do you know what it can be?

[@sebastiantorresvfx]

I’ve had this issue with other extensions in the past. I went into the extensions folder of automatic 1111 deleted the folder of the specific extension and reinstalled it. When that didn’t work I’d delete it and install it manually from its respective git hub link. Hopefully that helps. Also make sure your automatic 1111 is up to date because the first time I installed roop it kicked up a stink and I had to bring A1111 back up to date.

[@leonamnagel9999]

let's try@@sebastiantorresvfx

[@gomezabel8880]

This is not exactly related to your tutorial although this was a great tutorial, but Did github remove controlnet from their repository? I finally installed Automatic 1111 and about 12 extensions but controlnet was not available. I'm a newb so maybe I'm losing my mind and not seeing it, can someone comment and tell me controlnet extension for Auto1111 is there? Anyone can comment on this question, thanks

[@sebastiantorresvfx]

Do a goggle search for, lllyasviel/ControlNet-v1-1. You’ll find the models in the files tab. Hope that helps 🙂

[@MyneBestOfHardStyle]

can you do the same with SD, but without running it in localhost and without any "extras" ?

[@sebastiantorresvfx]

Interesting question, you can run it on the Runpod server but it’s a paid service, relatively inexpensive but something to keep in mind. I think it starts at like 30c an hour. You don’t need to use the Roop extension if you don’t want, you can use the character naming convention. But in saying that you will need to generate more images and have a good eye for detail to pick out the ones that don’t match. Hope that helps.

[@MyneBestOfHardStyle]

@@sebastiantorresvfx so this will not work on generators like Leonardo, NCS, dezgo, or others? I've been trying this trick of using the same name to get same face, with deepdreamgenerator (not google!), and it works well with the "photoreal" model they have (SD 1.5). For a while i always thought that this "trick" of using the name was something that i "get lucky" as result, but it looks like this is a consistent trick. Maybe the same way we can create new faces by mixing faces from celibrities.

[@sebastiantorresvfx]

I don’t see why it wouldn’t work on Leonardo from what I understand it was built on the stable diffusion models. As for the extensions I’m not sure what systems Leonardo and the others have built in. I would just use the unique names in that case. As for combining celebrities faces that could be done by setting each persons names to 0.5 in the prompt. Though I’m not sure how consistent the results will be. It’s kind of a shot in the dark.

[@eskelCz]

Nice. Can you do it with your own face? I’m trying to generate my portrait and I’m looking for the best method.

[@sebastiantorresvfx]

100% you can, that’s the beauty of this system as long as you have the references you can apply the face to anything then train the LORA with your face if that’s what you want to do. I’ve done that with a couple friends who wanted me to make them a LORA of them.

[@eskelCz]

@@sebastiantorresvfx Cool, thanks. I feel like that would be a popular video, if you could make a short tutorial :)

[@sebastiantorresvfx]

I’ll definitely get around to it at some point. 😁

[@TeluguNarrativeHub]

I need a tutorial for character turnaround. Front view,side vide view,back view with Same cloths,face. I tried Lora. Not that much great.

[@sebastiantorresvfx]

Okay that’s actually not too different to this method. Bare with me because I’ll be touching base with the whole turn arounds work flow soon, the results are amazing.

[@matthewanacleto7885]

Is that a standalone version of Stable Diffusion? I don't recognize the UI.

[@sebastiantorresvfx]

Just a standard install on automatic 1111 with a few too many extensions I was trying out.

[@FlashRey]

Why do we need to generate a random name? Can we just make up one? I mean is this step necessary?

[@sebastiantorresvfx]

You can make up a name, I’ve done it, works fine. This is more so you have full control over the character’s ethnicity and by extension their overall look.

[@CodyMapping]

the control net thing is not working for me, it just makes normal images (portraits of one person) instead of those 9 images like in the grid

[@sebastiantorresvfx]

What settings have you got? As for controlnet Remember to set the preprocessing to none and controlnet is more important.

[@CodyMapping]

I think the problem was that I used the wrong control net model@@sebastiantorresvfx

[@sebastiantorresvfx]

Yes that’ll definitely do it, make sure you only use canny. You can also use HED and scribble but I’m not keen on their results sometimes

[@CodyMapping]

@@sebastiantorresvfx i also realized that it only works with realisticVision as Stable Diffusion XL just makes normal images

[@sebastiantorresvfx]

SDXL doesn’t support controlnet on automatic 1111 yet. Only with comfyUI. It’s still pretty early days though

[@tanjirokit7477]

Where can I download your grid file? I'm getting error 404 when I try to download. thx

[@sebastiantorresvfx]

Odd, just checked the links. All active. Email me and I’ll hook you up.

[@crow-mag4827]

The newsletter signup says "This list cannot accept any more subscribers.", How can I get the Face Grid file?

[@sebastiantorresvfx]

try again, It will work now. Im just updating the newsletter.

[@ian9895]

@@sebastiantorresvfx There is not link to sign up for the newsletter, on your website there is no option other than about and videos

[@AlexVladimir91]

I want to generate my face in the grid image, but I don't know how, help

[@sebastiantorresvfx]

you don’t need to, that’s only if you’re generating someone who doesn’t exist: you can just create a grid with some photos you take and use those for the Roop extension so it uses your face in the generations

[@danielt7661]

Can roop be used for fixing hands? :D

[@sebastiantorresvfx]

Only face replacement. Hands are still a dicey subject. Some checkpoint models work better than other with certain hand gestures. It’s a matter of trying different one. I like to use a photo editor and a 3D hand, apply the replacement hand and use inpainting to blend it all properly.

[@danielt7661]

thanks I'll try that! @@sebastiantorresvfx

[@FMFvideos]

Is it possible to consistently create a non human character? For example a robot or a dog?

[@sebastiantorresvfx]

Unfortunately the answer is no and yes, no in that this system I’m showing can’t do that; it’s made purely for human faces. And yes in that it’s possible to create consistent non humans with the use of a LORA file, which means you’d need to have some level of artistic ability to create the dataset for the Lora file, be it in 2d or 3d. If you’re able to do that then you’d be able to get those character’s consistently. Hope that makes sense.

[@FMFvideos]

@@sebastiantorresvfxThanks! I'm new to stable diffusion, I'll do some more research

[@shemkipruto4394]

How and where do you download and install the software

[@sebastiantorresvfx]

I’ll be uploading a Stable Diffusion install video this week 👍 in the meantime if you’re keen to give it a go, Google “automatic 1111” and check out their GitHub. It does require some slight knowledge on installing python apps though.

[@shemkipruto4394]

@@sebastiantorresvfx noted thanks

[@stepfury]

I can't catch up where you create the 9 pictures with different angle and then crop to use only one picture for the roop. What is the purpose to create that 9 pictures?

[@ChristianIce]

Being a single generation it will output a consistent result, the 9 faces will have same seed and noise, so they will be more similar.

[@sebastiantorresvfx]

It’s all about having the options for the different angles. I’ve found Roop has better results that way 👍

[@stepfury]

So I keep all of them and then choose one of the angles that fit the most for rooping the picture I am working on. correct?@@sebastiantorresvfx

[@sebastiantorresvfx]

That’s right, you got it 🙂

[@stepfury]

@@sebastiantorresvfx Thankyou very much

[@futurepresident2598]

I just create a face and faceswap it in the whole batch

[@waleedkh9769]

What does Roop do?

[@sebastiantorresvfx]

Roop is an extension that finds faces in your image and replaces them with a face you load into the extension.

[@fullviewstatus3438]

Can you share your pc spec.

[@sebastiantorresvfx]

I don’t mention it on my video’s simply because I don’t want people to think you need a specific kind of computer to use SD. But I have a 3960x thread ripper and 3090 GPU. Both of which are about 3 years old now. So definitely nothing current and top of the line.

[@Serenergize]

roop does not show up after install, even after restart

[@sebastiantorresvfx]

You may need to uninstall it manually and install it manually, though I would probably go with something like adetailer now since it’s more updated. It can do a lot more than Roop could.

[@Serenergize]

@@sebastiantorresvfx Thanks i just figured it out as you replied. turns out i had the wrong control net installed or rather not the correct one from the list. Thank you for your videos! you are the king of stable diffusion on youtube and my go to youtuber!

[@Kelenn35]

Thanks but the roop extension is a big pain to install

[@sebastiantorresvfx]

I totally understand the first time I installed Roop it caused a slew of problems I ended up having to delete my whole install and start again. Which I wasn’t willing to do for another few months 😂 there’s a new face lab extension which is a little more robust but I haven’t had time to play with it yet. Let me know if you do.

[@dronearon3085]

you can use grid also for roop..option 0,1,2 ..for other technique like for swap faces ..kzbin.info/www/bejne/d5XKn55jmrCdpLc

[@hankkingsley9183]

Not a huge fan of tutorials that just give you instructions without explaining the "why". For example saying "you need Roop" but not explaining what Roop does, what it is for, why you are performing that step

[@twilightfilms9436]

Room has tons of issues for professional use, but the most annoying is that it can’t sustain a batch sequence. It goes crazy, out of control, and no matter what you do, it doesn’t respect the prompt…..

[@sebastiantorresvfx]

Yeah Roop isn’t meant for image sequences, it falls apart for that kind of workflow. This is more so for single images or multiple images that one can use to build a LORA model. With a few more steps after that you’d be able to use the LORA to generate image sequences with the face you made from Roop.

[@ArnoldVeeman]

Try to make "revealing" art in Midjourney is a pain 😂

[@sebastiantorresvfx]

😂 brilliant

[@pempers1885656]

Why is the website an IP adress

[@sebastiantorresvfx]

Basically, It’s the local address to access the environment you’re running automatic 1111 in.

[@JarppaGuru]

actually i see van damme with arnold face. not look that girl it look that real on photo with that girl lol

[@sebastiantorresvfx]

😂 😂 that’s gold!

[@KS-tz6dl]

knew this shit would be to good to be true ffs

[@ghostbusterz]

Roop fails for anything besides a blank expression.

[@sebastiantorresvfx]

Really? Haven’t found that to be the case. If promoted with a specific expression like smiling laughing etc it’ll copy the expression over to the new face. Just did a quick test to be sure.

[@ghostbusterz]

@@sebastiantorresvfx I usually get weird anatomy. Could be any number of factors really, hard to tell.

[@sebastiantorresvfx]

Sounds like it could be the model you’re using to generate more so than Roop itself, Roop only replaces the face after the image is generated.

[@bigdeadlykicksandkickflips]

yo.. what the fuck is STABLE DIFFUSION? I look it up and its like its NOT a program? But you said "in" Stable Diffusion. HOLY FUCK. eVERYTIME i've looked it up. its like this.

[@sebastiantorresvfx]

lol you’re just a bundle of joy aren’t you. stable diffusion is the type of ai models used, in Automatic 1111 and ComfyUI programs.

[@UNLKYHNTR]

Clothing brand models crying in vain

[@sebastiantorresvfx]

lol yes and no. I’m a photographer for a fashion influencer and she wants me to train a model on her likeness so she can monetise her image more efficiently 😆

[@Fumas12]

​@@sebastiantorresvfxwell only until the labels themselves create "their offical brand face" and replace all but the most exclusive VS models through AI. Their only silver lining right now is that we can't really reliably put the designer's pieces on them. Once we're able to train a very specific piece of clothing or accessory, they're done for

[@kkryptokayden4653]

Ai don't have aging problem so yeah it's a plus

[@a_alhamed]

What is the URL of Stable Diffusion?

[@simonhick9124]

why do you look so sad

[@sebastiantorresvfx]

🥴I have no idea what you mean. Maybe it was a really hot day and I had done one too many takes 😂

[@JefferyWWright]

Installing roop is a comical nightmare, unbelievably vague instructions and requires downloading several gigabytes of microsoft bloatware, developers need to learn how to use language and articulate instructions more clearly than "Run this command: pip install insightface==0.7.3" where? In what program? Their official documentation says, not kidding: "For rest of the errors, use google. Good luck." Not ready for prime time.

[@sebastiantorresvfx]

Yep, I’m gonna going to lie the first time I installed Roop I almost threw the monitor at the wall because it broke my entire installation. In saying that I need to make an updated video for this because Roop is now basically ‘reactor’ and it’s a lot easier to install from the extensions tab.

[@JarppaGuru]

why need use computer fix and lie photos. amateur go outside shoot photo and its BUUTIFULL right from camera. no need photoshop lol. why pros allways fix things they did not get on image first place lol

[@sebastiantorresvfx]

😂 😂 the Mona Lisa supposedly has 30 layers of paint and 3 different versions beneath the final version. Old school photoshopping 😂 fixing images goes with the territory, cameras don’t always capture what’s in our imagination.

[@abakedpotato1486]

You're the amateur clearly never heard of shooting photos in RAW format which NEED software like Photoshop. Go educate yourself on photography before you open your mouth lil bro

[@joynigam4151]

Gotta ask you man, whats the point of this? I want to have a good look at AI and yet I fail to do so. Need your opinion.

[@sebastiantorresvfx]

I think everyone will get what they need out of it. In my case I use it when I need a vfx asset, or something to fix a photo so I use inpainting to create that asset. It’s like using photoshop to touch up something but not needing to go on a deep dive into Google to find it 😆 Probably won’t be for everyone but it can be handy.

[@joynigam4151]

@@sebastiantorresvfx and what's your opinion on AI replacing jobs?

[@sebastiantorresvfx]

I suppose it depends on what jobs. After Jurassic Park a lot of people were terrified that CGI took over their jobs. In reality it created a completely new industry and different jobs. I believe ai will be no different, it’ll change some jobs, make some obsolete and create a completely new division of jobs in all industries.