스테이블 디퓨전 3.5 미디엄 모델 개요 및 비교

### 새로운 스테이블 디퓨전 3.5 미디엄 모델 살펴보기: 기능 및 커뮤니티 반응
* 🎥 이 영상은 시청자가 **새로운 Stable Diffusion 3.5 Medium 모델**을 사용해보고 재미있게 즐기면서 자신의 의견을 제시하도록 권장합니다.
* 💬 크리에이터는 서포터들에게 감사를 표하며 동영상을 제작하는 데 있어 **서포터들의 역할**을 강조하고 커뮤니티 내에서 **새로운 모델이 어떻게 받아들여지고 있는지에 대한 피드백**을 장려합니다.
* 📊 3.5 미디엄 모델은 **25억 개의 파라미터**를 지원하여 이전 3.0 모델보다 크게 개선되었지만, 생성된 이미지에서 **인체 해부학적인 문제**를 여전히 해결해야 하는 과제를 안고 있습니다.
* 🖼 이 모델은 고품질 이미지를 생성하고 프롬프트를 정확하게 따르는 능력으로 유명하지만, 특히 **인체 해부학 렌더링에서 3.0 모델과 몇 가지 문제**를 공유하고 있습니다.
* 💡 3.5 미디엄은 스테이블 디퓨전 XL에 비해 특히 GPU가 약한 사용자에게 화질과 속도 면에서 더 뛰어난 **'스테이블 디퓨전 엑셀 킬러'**로 불립니다.
* 🔍 크리에이터는 커뮤니티의 반응이 명확해질 때까지 **3.5 모델에 대한 추가 동영상 제작을 자제**할 계획이며, 이는 해당 모델의 장기적인 사용에 대한 불확실성을 나타냅니다.
* 🔧 새 모델에 대한 설치 옵션이 제공되며, **Patreon 후원자를 위한 자동 설치 프로그램**이 포함되어 있어 기능을 살펴보고 싶은 사용자도 쉽게 이용할 수 있습니다.
* 🚀 전반적으로 **3.5 미디엄은 가능성을 보여주지만**, 그 수용과 사용은 궁극적으로 커뮤니티의 참여와 더 나은 성능을 위해 모델을 미세 조정하려는 의지에 따라 달라질 것입니다.

3.5 중형 모델은 안정적인 디퓨전 XL보다 뛰어난 이미지 품질과 신속한 정확도를 제공합니다.

hello humans, my name is K. Overload, and Stability AI just released another new AI model called Stable Diffusion 3.5 Medium, only one week after the release of the 3.5 Large. This model is aiming to outperform the 3.0 Medium model released a few months ago and to bury the Stable Diffusion XL model once and for all. So, if you have a weak GPU and you want to run the latest state-of-the-art text-image model, well, this model is maybe for you. But how does it actually perform compared to the 3.5 Large, 3.0 Medium, and Stable Diffusion XL? And should you switch to this newly released model? Well, this is what I'm going to showcase in this video. So, that being said, sit back, relax, and let's go. All right, so to install the new Stable Diffusion 3.5 Medium, you can either use the updated automatic installer that is available for my Patreon supporters for both the local install and RunPod, or you can just follow the manual.

installation guide that is available in the description down below, and the preview simple only one 3 five workflow that I showed you in my previous video will actually work with this medium model as well because it is kind of essentially the same type of models. And all you have to do is basically go into your manager, click on update all, then load the model, and you're good to go. Oh, and also speaking of workflow, My ultimate only one workflow V2 for flux is normally coming this weekend, so don't forget to subscribe and follow me on Patreon so that you get notified once it's released. Okay, so now let's begin. So just like the 3.5 large, what makes this 3.5 medium model so special, and are there any differences between the 3.5 large and this 3.5 medium? Well, first, if you remember correctly, the 3.5 large was an 8.1 billion parameters model, whereas this new 3.5 medium is a 2.5 billion parameters, meaning that the model is

much smaller and thus requires much less vram to run and is much faster to generate an image. So, if you have a weak GPU and you cannot run models like flux or stabil 3.5 large at a high speed, well, this 3.5 medium is probably your best bet. Because if you can run stable diffusion excel at a good speed, since it is pretty much more or less the same sized models, you'll be able to run the 3.5 medium at the same type of speed. The second good part of this model is that, again, just like the 3.5 large, it is really, really good at following your prompt. So, again, I'm going to use the same prompt that I used for the previous model, which was basically a photography of a gorgeous blond woman, real estate agent with blue eyes, red lipstick, wearing an orange suit with diamond earrings, sitting on a chair in our office. So, now if I can generate and we get something like this, which is once again exactly what I asked for, a blond

woman, blue eyes, red lipstick, diamond earrings, orange suit, sitting on a chair, basically a great image, perfectly following my prompt with a pretty good image quality. Like once again to remind you, if I actually choose the 3.5 large and I generate the same image, we get something like this, a very similar-looking image, although it's a little bit of a higher quality compared to the 3.5 medium. However, that same image generated with the 3.5 medium was generated like three times faster, so that's definitely something to take into consideration. And basically, that's just it. The 3.5 medium is basically like a 3.5 large, just a little bit faster with a little bit less quality, but usually holds more or less the same qualities as the 3.5 large. So just once again, the ability to be trained and fine-tuned much easier compared to a model like Flux, since this is not a dist model where you can also use the negative prompt, the CMG scale, just like I showed in my previous video, just like the 3.5.

it has a huge array of styles right out of the box. So, like for example, if I put something like an illustration of a young woman inside of space element, if now I press generate, it gives me something like this, which looks absolutely fantastic, just really, really beautiful. And it knows like a bunch of different styles and artists that you can use in your prompt. Like for example, this is the same illustration, but this time in the style of Frank Franzetta, which once again looks really, really good. So once again, just like the 3.5 large, it has a lot of different styles inside of the model right out of the box. So that in the future, if you want to F-tun it on a different style, this should be very, very easy. However, there is also one other thing that makes this model actually much better than the 3.5 large, and that is the fact that the 3.5 large had a limit of 1-megapixel resolution, whereas the 3.5 medium has a limit of 2.

megapixel meaning that you can generate high-resolution images right out of the box, just like flux or stable diffusion Excel. So, for example, if I input something like 1792 for the width, and I can generate, I get this type of really cool image, which looks kind of pretty cool, and this was generated right out of the box. If you CH the same thing with the 3.5 large, if you remember correctly in my previous video, that same image with the same parameters would give you something like this. Oh my God, this is horrifying. So yeah, it really gives you this really horrible atrocity, really coming from a nightmare, because it is above one megapixel in resolution, which is something that the 3.5 large does not support, whereas the 3.5 medium has no problem generating those types of large resolution images, and all of that without any fine-tunes, right out of the box. So yeah, I mean, that's really, really cool, and yes, once again, I know you're going to ask, just like the 3.5.

large the 3.5 medium is a truly unsensored model, so you can generate your bazongas and buas without any issues. However, that being said, there's still a few problems and a few cons of using this model. Because yes, this model is truly the stable diffusion Excel killer. There is absolutely no doubt in that. Because if you, for example, generate an image with the 3.5 medium with this types of amazing quality, and you take the same prompt and put it inside stable diffusion Excel, this is the type of images that you would get. So yeah, really not the same types of quality. This is stable diffusion Excel, and this is the 3.5 medium. Two models that should be more or less of the same size parameters, and no matter what prompt you choose for the generation, the 3.5 medium will always outperform the stable diffusion Excel. And mean, this is stable diffusion Excel. This is the base model that was used to fine-tune models that could generate images like these. So imagine what we could do with a model like the 3.5.

medium. This could be huge, so yes, I'm going to say it and say it again: the 3.5 medium destroys the stable diffusion XL on every point. It is just the more superior model, even though it has pretty much the same size and requires more or less the same amount of VRAM to run. So yeah, the 3.5 medium definitely wins. However, that being said, the 3.5 medium is not a perfect model, and it still has the same amount of troubles with having another to me that the 3.5 large has or that the 3.0 model had as well. Like, for example, if I try to generate the super infamous photo of a woman laying in grass, and I can generate once again, I get this very weird types of image with some weird anatomy, very reminiscent of what the 3.0 model could do. So like, once again, if I generate another image, it still looks kind of weird. Let me actually generate a few of them at the same time so we can compare it with

a 3.0. So, there you go. This is the 3.5 medium, and pretty much none of the images look really good. They all look kind of weird. And if I do the same thing with the 3.0 this time, you can see that here again we get the same types of weird images that have become so infamous in the community. And yet it still managed to generate like two better images than the one generated with the 3.5. So, yeah, in terms of the quality of the anatomy, the 3.5 medium is really not that much better compared to the 3.0. If I, for example, generate another image, like for example, a photo of a woman doing yoga in a busy park in a dynamic pose, and I click generate, which as you can see once again generates those horrible atrocities, almost with the same types of quality that the 3.0 used to do. And actually, if I do the same thing with the 3.0 this time, and as you can see, this is the

images made by the 3.0, so pretty much on the same level of atrocity compared to the 3.5, when you have some extra legs, extra limbs, where basically the human anatomy is just not present and not respected at all, and it's really just difficult to understand which one is which. So, if a few months ago, people complained about the 3.0 and how bad it was for human anatomy and refused to use it because of that, I don't see a lot of reason why people wouldn't complain this time around because, I mean, this is really the same types of generation. And, funnily enough, civil diffusion Excel actually does add a little bit better compared to all of those models, at least when it comes to basic human anatomy. So, I mean, yeah, this is really not good. So, basically, yes, the 3.5 medium is better than stable diffusion Excel overall when it comes to the quality of image generation because it's more precise, it follows the prompt better, it has a lot of

styles built into the model, but the problem is that the 3.0 used to be the same types of model as well. It was also a model that could follow the prompt pretty well. It was also a model that was more powerful than the St diffusion XL, that had a lot of styles and could generate very beautiful pictures. But this model was forsaken by the community for its bad human anatomy and the horrible commercial license. Whereas the 3.5 medium still has the same anatomy issues, but this time has a much better license. So now I'm wondering if the community will use this model or not. This is really not up to me to answer. This is a question that I'm asking you. And basically, my personal opinion on this model is that, well, just like what I said on the 3.0 last time that I made a video about it, I said that it was a very powerful model that could become much, much better if people were to fine-tune and train new checkpoints so that the final models become better and better and

better just like the ones that we have right now based on the stable diffusion Excel. And if that model is more powerful than Cil diffusion Excel, imagine what kind of model we could make if the community decides to train it with the same level of enthusiasm as they did with st diffusion Excel. But once again, this is really up to the community, not me. So basically, I'm going to conclude this video with the same conclusion that I did in my previous video, meaning that for now, I'm not going to make any new videos about the 3.5 models since I don't know if the community are going to use them or not. I'm still going to stick now with the flux model because it is a much better, much more powerful model. And the flux model can generate some absolutely beautiful, amazing images with absolutely amazing human anatomy from the base model only. This is still right now my favorite model. And since apparently right now, as I'm making this video, we have an actual real fine-tune model of the flux Dev model, maybe the

flux model has more future than I thought initially. So yeah, I mean again, wait and see. That's some really good news. So yeah, there you go. This has been the 3.5 medium. I'm not quite sure how the community will really react to this release, at least as of right now. We'll see in the next few weeks if people find better ways of using this, or maybe people will completely forget the 3.5 and will stay on the flux instead. For now, I really have no idea, so definitely let me know in the comments down below what you think of this new release, what you think of this new model. Are you going to use it? Do you think the community will use it and replace the stable diffusion Excel with it, or do you think it's going to have the same backlash as the 3.0 where the model will be forgotten in about a week? So definitely let me know in the comments down below. Don't forget to subscribe and follow me on Patreon to get notified for

my next video and definitely try this mod yourself right now so that you can form your own opinion and don't forget to have some fun. And there we are, folks. Thank you guys so much for watching. Don't forget to subscribe and smash the like button for the YouTube algorithm. Thank you also so much to my supporters for supporting my videos. You guys are absolutely awesome. You people are the reason why I'm able to make these videos, so thank you so much, and I'll see you guys next time. Bye-bye.

NEW UNCENSORED SD 3.5 MEDIUM AI Model Is HERE BUT...

Welcome to Aitrepreneur, I make content about AI (Artificial Intelligence), Machine Learning and new technology.

Thank you so much for watching and don't forget to subscribe! Thanks!



Aitrepreneur

5 라지 모델 출시 일주일 만에 3.

0 미디엄 및 스테이블 디퓨전 XL 모델을 능가하는 성능을 목표로 출시되었습니다.

5 Large와 비슷한 이미지 품질을 제공하지만 더 빠르게 작동하므로 GPU가 약한 사용자에게 좋은 옵션입니다.

또한 이 모델은 업데이트된 자동 설치 프로그램 또는 수동 설정을 통해 설치 옵션을 사용할 수 있어 교육 및 미세 조정이 더 쉬워졌습니다.

5 중형 모델은 이미지 생성을 위한 다양한 스타일과 아티스트를 제공하여 멋진 결과물을 만들어냅니다.

5 대형 모델에 비해 해상도 제한이 2메가픽셀로 더 높으며 무노이즈 처리되어 더욱 다양한 콘텐츠를 제작할 수 있습니다.

5 미디엄은 이미지 품질에서 안정적인 확산 Excel을 지속적으로 능가하므로 이미지 생성에 탁월한 선택입니다.

0 버전은 모두 이상하고 낮은 품질의 이미지를 생성하며, 3.

0 버전이 때때로 더 나은 결과를 생성합니다.

5 미디엄은 1메가픽셀 이상의 고해상도 이미지를 생성하는 데 어려움을 겪는 반면, 3.

0 버전은 고해상도 이미지를 쉽게 생성할 수 있습니다.

0 버전에 비해 이미지 품질이 크게 향상되지 않았습니다.

5 모델은 인체 해부학에 어려움을 겪고 있으며, 팔다리가 추가되어 혼란을 야기하고 인체 형태를 존중하지 못합니다.

5 중간 모델은 더 나은 이미지 생성 품질을 제공하며, 더 정확하고 빠르며, VRAM을 덜 필요로 하므로 GPU가 약한 사용자에게 이상적입니다.

5 미디엄은 결함에도 불구하고 프롬프트를 따르는 데 효과적이며 특정 요청에 대해 정확한 결과를 제공합니다.

5 미디엄 모델: 강력하지만 결함이 있는 옵션

5 중형 모델은 25억 개의 파라미터를 가지고 있어 3.

5 대형 모델보다는 작지만 Stable Diffusion XL보다는 더 강력합니다.

5 중형 모델은 그 잠재력에도 불구하고 이전 모델과 동일한 인체 해부학 문제를 공유합니다.

커뮤니티에서 이 모델의 성공 여부는 불확실하지만, 3.

5 대형 모델과 유사한 설치 가이드와 워크플로우를 제공하여 쉽게 전환할 수 있습니다.

이 발표자는 시청자들에게 새로운 3.

5 모델을 직접 사용해보고 의견을 공유할 것을 권장합니다.

그들은 출시에 대한 커뮤니티의 반응과 플럭스 모델을 대체할 가능성에 대해 불확실성을 표현합니다.

발표자는 더 강력하고 멋진 이미지를 생성할 수 있는 것으로 입증된 플럭스 모델에 당분간 집중할 계획입니다.