SDXl 1.0 Is Here! The Best Ever Stable Diffusion Model | Competitor Of Midjourney

so stability here I just released the sdxl 1.0 and it's coming up with lot of improvements and the amount of quality it's producing right now is just insane that we have never ever seen before from any open source a image model so first of all we're gonna look at the number of improvements that they have made and then we will access this tool and try this tool on our own so announcing sdxl 1.0 the stability AI team is proud to release an open model sdxl 1.0 the next iteration in the evolution of text to image generation and they are also claiming that the new version of xdxl has been improved to the world's best open source AI image model right now and I totally agree with that because I have already tried this model uh so if we scroll down a little bit for this is the best image model from stability AI uh so far and this is the graph s t XL is a flagship image model from stability Ai and best open source image generation model we have tested this again a various other models and people prefer images generated by stxl 1.0 and this is the graph they they are showing the comparison between table diffusion models first of all we can see stable definition 2.1 uh stabilitation 1.5 then we can see a massive change in sdxl 0.9 that was also a really good model and now we have stxl 1.0 and this is the difference between their base model and the refine model we scroll down a little bit sdx generate images from high quality in the virtual reality and is the best open source model for photorealism and it is really getting more and more realistic and you can see in this example images that they have shown and you can see this is getting better with the hands and better with the legs and human faces where you can see a woman chasing a dog and this is a really good image in addition sdxu can generate concept art or notoriously difficult for image model such as hands and text it means it's also getting better with the hands and text well sdx is 0.9 was also better at hands but this is much more better then sdxl 0.9 so these are few example images that they have shared sdx will require only a few words to create complex detail and a statically pleasing images so right now we don't have to type the longer prompts so now it's getting more intelligent and better so you can also use shorter prompts to get these kind of amazing images out of it and it's also getting better to differentiate between the words sdxu can understand the difference between concept like the red square or famous place versus a red square or shape it means now it can understand your prompts better and it can also differentiate between the place and the word script that they have shared there in this example you can see sdxl is one of the largest parameters counts of any open source model using a 3.5 billion parameter so the base model and 6.6 billion parameters for the refiner model and this is also getting better with the fine tuning and control and this is the example where you can see this is a simple image of a shoe and now uh it is generated landscape of a mountain uh just based on uh this shoe and you can see and this landscape is also looking like a shoes so there are a couple of ways to use this model so get started with stxl there are several ways to get started with sdxl 1.1 S6 1.0 is live on clip drop so you can use this a model in clip drop from stability AI so these are a couple of the platforms and these are a couple of ways how you can use sdx is 1.0 and you can also use this model in the dream studio so we're gonna try out this model so if we go to the clip drop by stability AI you can see sdx is 1.0 so if you already used sdxl 0.9 in a clip drop by stability I you may be familiar with this interface and now you can see we have few more options so if I click on that you can see we have few more Styles uh like digital art comic book no style enemy photography analog bunch of different styles and now we can also change the aspect ratio and now you can also add the negative prompt in it let's say I'm gonna try a photo of a woman and I'm gonna select no style and I'm gonna select the aspect ratio to landscape and let's say I'm gonna type A negative prompt extra LX extra hands ugly face so let's see what it comes up with foreign look at this uh so if I open this image you can see this is the first image and look at the realism in this image so this is looking like a real photograph and look at the skin texture so I think they made a massive Improvement in terms of skin texture previously it wasn't able to generate this kind of skin texture so this is the next image where you can see so this is a better example where you can understand how realistic this image is and look at the skin texture and I think the best way to test any image model is to test their photorealism because every free AI image generator or any paid image generator can produce fairly good images of a concept art or like cartoon images and anything else but whenever we talk about photorealism every image generator lags in it and I think this could be the biggest competitor for Mid zainik because this is producing so much real images and look at the skin texture so I think this is already achieving the level of photorealism so we're gonna look at some few more examples in a photo for mountain and I'm gonna select the aspect ratio of 21.9 which is a very wide image uh so let's see what it comes up with okay so these are the output images that it generated and look at this how wide this picture is and look at this and look at the results how realistic it is look at the Reflection In The Water look at the realism this is so real man and these are few more examples so this is the second image and this is the third image I think this one is also looking more realistic and look at this uh we can see a series of mountain and we can also see the same reflection in the water and this is really getting better so there is no funkiness going on with the uh Reflection In The Water this is already looking spot on so now we're gonna try hands and text and uh let's see okay so these are the results uh so I give it a prompt uh to generate hands human hands and look at this uh so this is the first image and this is the second image I think this is looking better in all of them uh but still there are some extra fingers in it where you can see uh the left hand is the right hand is looking good but uh you can see there are six fingers in this hand so these are few more results from this so I think this is still struggling with the hands but if we talk about the uh other photos like human faces and everything it's working better so we're gonna try out the text so we're gonna see how does it perform with the text okay so then I type the text office and you can see these are the results from it uh I think this is the first one where you can see the spelling mistake in it and then we have I don't know what's going on with this picture and then we have this image where you can see it's stills struggling with the spellings and text but otherwise I think this is uh this model is very good model if we talk about the in photorealism images and the human faces in Texas this is uh this is really improved model and I think this is the better the best open source image model right now and this could be the competitor of mid Journey so uh that's it from this video and if you guys uh find this video helpful give it a thumbs up and if you guys in a space of AI make sure to subscribe Planet Ai and I'll see you in the next one

Channel: Planet Ai

Views: 769

Keywords: Planet Ai, stable diffusion, sdxl 1.0, sdxl 1.0 release, stable diffusion sdxl 1.0, midjourney ai, midjourney alternative free, midjourney competitors, how to use sdxl 1.0, planetai, midjourney

Id: 3vhGpGBmrvU

Length: 8min 5sec (485 seconds)

Published: Thu Jul 27 2023