Robots & Cyborgs in Stable Diffusion! Custom Model is totally awesome!

[@dadlord689]

Wow! Specialized models might be the right way to get exactly what you would like to see. This is really interesting.

[@sedetweiler]

I am excited to see what people come up with!

[@dadlord689]

@@sedetweiler I have found for myself that it is not that easy to achieve what I want , but I believe that it is really just another tool for a good artist. I just may expect to be able to pay for affordable art, just like in some assets store, but that will be specified for my needs. From my experience it is too much easier for experienced person to do a simple task then to me went through all that initiation and small pitfalls that are destroying motivation. I hope for more decentralized production and marketplace, less expansive and hard workflows. Because at this point we can't hope to see some movie based on some book that many people loves. We sort of have to consume what is available. And people are not willing to fund projects because of bad reputation (too many failed crowdfunding projects being done). So this so called "AI" technology is really handy. Yet I am not a fan of decisions-making AI, because it is too much.

[@max477]

Any word documentation for stable diffusion for prompt with example would be great....!

[@AntiPattern123]

Everything looks awesome with the Prompt "Giger painting" at first :D

[@3oxisprimus848]

That intro was Fire! Amazing content (as usual)

[@sedetweiler]

Glad you enjoyed it!

[@mr-nobodyz-g7q]

This is so dope. Thanks for helping me install locally

[@GyroO7]

Can you do a tutorial about installing stable diffusion infinity

[@rickymcdowell199]

Is the Ai creating is own original images bases on those artists/styles or is it pulling images from the internet bases on the tags

[@ksp-crafter5907]

It creates completely new & unique images that don't exist until that!

[@corpostar]

@@ksp-crafter5907 New but very similar, I can imagine the legal problems they could have if used in commercial projects, before the nearby AI regulations. They are easily identifiable and will probably never be used for major commercial projects.

[@lynnqi6451]

WOW! Amazing!

[@tmaxstudio55]

Hey, Scott. It would be nice if, maybe, you put these videos in a playlist? Just a suggestion, if you like. I'm enjoying your tutorials, btw. Thanks.

[@sedetweiler]

Thank you for the idea, and the support! I thought I had them in a playlist, but I will check on that.

[@s1gma9000]

learning alot, keep 'em coming! enjoying the vids

[@sedetweiler]

Thanks, will do!

[@AscendantStoic]

Yeah, that's an awesome model and nice results you got from it, and don't underestimate anime models like Waifu Diffusion, both 1.2 and 1.3 can produce very interesting results as the infusion of anime art affects all human characters generated even realistic ones creating an interesting semi-realistic look, specially when combined with classical artists like Anders Zorn, Sophie Anderson and William Adolphe Bouguereau, and maybe some pin-up artists like Gil Elvgren.

[@sedetweiler]

I'm all excited to try them. 8 did have some luck initially (a few weeks ago) but I think getting those artists right will help a ton. Thank you!

[@entrypoint2009]

Are you using Image2image and inpainting? I ask because I can't make a real human face using this model like the ones you show, like a cyborg.

[@sedetweiler]

not on these pieces, this is how they came out of the oven.

[@gr33nDestiny]

@@sedetweiler I cant figure out how to get a human face either, I tried many prompts, Its also difficult to get a full body

[@bobhawkey3783]

Amazing. So mad I have to go away for a few days so I can't try this! Thanks so much.

[@sedetweiler]

It will be here when you get back. 😉

[@WillButtlerYeets]

HOW DO I MAKE A MODEL? I know how to make embeddings via textual inversion or dreambooth, but how do I make a full model like this?

[@sneedtube]

Sorry for the stupid question but I'm a big noob: if I trained a model (of myself) and I want to use it in conjunction with this new robots/cyborgs model, how should I go about that? Should I train a third model merging the two? There must be a simpler way...

[@sedetweiler]

Probably best to do with an embedding for your face and then you can use a few of them in combination rather than training a new hypernetwork.

[@patruff]

More fluid!! I've given her all she take captain!

[@mac24seven]

So, just to confirm... To install a new model, imply download it and stick it in the stable diffusion model folder. No renaming or anything?

[@sedetweiler]

Correct. Just place it in that folder.

[@mac24seven]

@@sedetweiler I've been looking for this answer for a while! Thanks so much! I just bought a 12gb 3060 today and can't wait to try this out :)

[@hatuey6326]

hi simple question how you can create human face with this models plz ? because i strugggle with it

[@widowido2972]

Merci pour cette vidéo !

[@gearboxxxrox1]

Im not sure whats happening but i always seem to be getting this repetition patter of some sorts. I have my setting exactly the same as yours and seem to get unfavorable results. Any ideas?

[@Michael-Humphrey]

im hooked

[@redplanet76]

Mega dope

[@sedetweiler]

I think so as well!

[@lithium534]

Hi. I really like the image at time stamp 3:41. I have been trying to create that exact same pose before this video with no avail. Would you be willing to share the seed so that I can other version of it? I know it's asking a lot. Nonetheless thank you for the video and great ideas for prompts and paint up.

[@CamiloMonsalve]

AMAZING! 😱

[@sedetweiler]

I love it as well!

[@Coltography]

Could you do a video on converting diffusion models to ckpt?

[@sedetweiler]

That is probably the hardest way to add custom images. Is there a reason you want to do it using that method?

[@Coltography]

@@sedetweiler I've seen a python script floating around that lets you do it, I ask cause when browsing modelverse they're not all ckpt files and need to be converted apparently. It's on my list of "need to learn" for next week haha

[@johannes.gamperl]

Hey, is there a way to create a complete image. In most pictures, the hair or the upper area is cut off at the top.

[@sedetweiler]

It really depends on how the images were trained. It is an issue for everyone right now because all of the models are trained on 512x512 and that often means the top of the head has to go. You can try outpainting it to get some additional room.

[@alanokamura6250]

Hey there, another awesome and informative video! Hey if I’m using Dream Studio any tips on how to install this model? Does this only work if you have a local install? Thanks.

[@etiennevanier]

Cool! I was looking for better cyborgs for a hobby pen and paper RPG project!

[@sedetweiler]

I think you have an infinite number of them now. 😂

[@MrLaura34]

has an error to put a stable studio robot CKPT file, it is an error. missing a cmd?

[@trent797]

Incredible stuff...been binging all your content and really appreciate what you do! Question: have you mostly switched to running Stable Diffusion locally? I've only been looking at this stuff for ~ 1 week, but it seems like running Stable Diffusion locally is way more cost effective if you want to iterate over tons of ideas.

[@sedetweiler]

Yup! That's what I do.

[@omarmolina_videoeditor]

Someone with the same issue? Cannot retrieve the public link of the file. You may need to change the permission to 'Anyone with the link', or have had many accesses.

[@jeremiahdelk8306]

how do you download models?

[@shadowmead8506]

I can see my model in the stable diffusion checkpoint box but when i select it the v1-5 pruned shows up instead? why? lol

[@blackvx]

Scott, don't you miss your old job? 😂😉 Thanks!

[@sedetweiler]

I do sometimes. 😂

[@devnull_]

Can it actually generate anything else than those so boring "mean" looking stealth plane / bucket head hybrids (and borg clones)? Anything more happy and positive perhaps? More of the same Artstation "concept art" derivatives. How about full body shots?

[@sedetweiler]

I think it was trained on most headshots, so not a lot of options there. I am hoping this gets updated if it's successful enough.

[@Pauluz_The_Web_Gnome]

Hi Scott, what gpu do you have? It's darn fast, isn't it?

[@sedetweiler]

I have a 3080 and love it!

[@Pauluz_The_Web_Gnome]

@@sedetweiler Woow cool! Unlucky I can't afford it, but I have ordered a 3060, and hope that will work gpod and fast also!

[@chacecampbell2697]

@@Pauluz_The_Web_Gnome I have a 3060ti and it works great! Just a bit slower than his, and I can train a textual inversion set 10,000 steps in just over an hour. Especially with how quickly this stuff is being optimized that card will keep you happy.

[@Pauluz_The_Web_Gnome]

@@chacecampbell2697 Hi thanks for joining in...The 3060 OC is a bit slower then the ti version right? But still fast enough?

[@chacecampbell2697]

@@Pauluz_The_Web_Gnome I'm not entirely sure as I'm pretty new to the world of understanding anything about computer hardware, but that's my understanding. The extra 6gb of VRAM I'd hope would make up for any loss of speed though with the ability to make larger images even if they take a bit longer. Just make sure to add --xformers to the bat file so it runs even more smoothly, I noticed and immediate speed boost just after adding. I also have 34gb of ddr4 ram, after having a damn aneurysm over which to get with my budget and that's how I rationalized going with the ti and less vram, the performance in other creative apps and file systems is night and day, though I was on an ass budget office pc before so it's hard for me to gauge against anything substantial.

[@Gamezmodz444]

Siiiiick

[@sedetweiler]

That's what I thought! I had to share this!

[@TadeuLevy]

I check news from SD literally every day to see if they added AMD cards support. Doesn't seem like it will ever come. Until then, I'm stuck with Midjourney (which serves me well, but SD is free and open-source, and I don't really like the way MJ's developers are handling it)

[@bilalalam1]

I am a paid MJ user , the AI generated by MJ is two steps ahead of SD - but with SD requires a lot of tweaking and probably can get results like that , and even better - so MJ is more like out of the box apple product with little customization and SD is more android do it yourself 😅

[@rodriguez4363]

You should at least try using stable diffusion on colab or some websites

[@bilalalam1]

@@rodriguez4363 you are right - I downloaded 1111 yesterday and loved tweaking my images by MJ - more so good for in painting than new generations so far

[@TadeuLevy]

@@bilalalam1 with SD I can batch-render stuff overnight and wake up with about 500 images

[@bilalalam1]

@@TadeuLevy that’s exactly what I am doing right now - made an awesome initial image on MJ - didn’t like the head gear - put it in new inpaint 1.5 - brush applied - new command - 50 images ordered - all look amazing - now in crisis which one to choose and which one to let go :)

[@BastianSteffens]

Dont know hoe a artist could compete with this. It takes a few seconds per pictures.

[@zemu252]

Live theft record we can even see the victims names lol

[@PromisedTriggeringExcalibait]

How can you steal something that they didn't make?

[@zemu252]

Well go and publish a marvel movie then cuz it aint stealin if they didnt make

[@MichaelJVanecekWrites]

I officially hate you. I'm going to put that in my byline. Michael J Vanecek hates Scott Detweiler. As payback, try this prompt - nousr spaceship, star wars, 4k Bwahahahahahaaaaa. Cough. Actually, go crazy with anything after nousr spaceship - some really cool and funky results.

[@MichaelJVanecekWrites]

OMG - try something like nousr armor (scarlett johansson:1.0), 4k, by __artist__ too (provided you have wildcard prompts script active) Goodbye having a productive day. :)

[@itslocane412]

Hey Scott, is there a way to do this but also include the body? I can't seem to get a full body render

[@sedetweiler]

The issue is that the model was probably trained on the upper body. It isn't your fault.