Is Playground v2 A Step Up From SDXL?

Get ready for an exciting update from Playground! In this video, we're thrilled to announce the release of Playground V2, a new and improved version of Playground v1 built on a similar architecture to SDXL. With the ability to run it locally, this open-source model has been tested and preferred by users two and a half times more than SDXL. Today I'm going to show you how to run it locally on ComfyUI. Don't forget to hit that subscribe button to stay updated!
Check it out on the official Playground channel • Breaking News! Playgro...
Read about Playground v2 blog.playgroundai.com/playgro...
Download Diffusers model huggingface.co/playgroundai/p...
Download Safetensor Compatible with any SD platform. huggingface.co/playgroundai/p...
⏲Time Stamps
0:00 What is Playground v2?
0:45 ComfyUI workflow and best settings
2:13 How to install Playground v2 Diffusers in ComfyUI
4:20 Checkpoint for A1111 and other SD platforms
5:11 Current limitations
5:37 Sample images
*Disclaimer Affiliate Links Below*
📸 Gear I use
Sony A7C CAN amzn.to/3spWX8C
Sony 35mm 1.8 CAN amzn.to/36Apekr
OBS (FREE) obsproject.com/
Editor Davinci Resolve www.blackmagicdesign.com/ca/p...
Audio: Rode Podmic amzn.to/35sSnxv
🎵 Epidemic Sound
🔦 Find us on:
Discord: / discord
Instagram: / monzonmedia
Facebook: / monzonmedia

[@MonzonMedia]

There is now a safetensor checkpoint compatible with any SD platform (6.94GB) download it from HF huggingface.co/playgroundai/playground-v2-1024px-aesthetic/tree/main

[@MonzonMedia]

If you get to run it locally, let me know your thoughts! Playground v2.1 is coming soon!

[@MyAmazingUsername]

I've tried Playground v2 now. It's extremely good for making heavily stylized, contrast-heavy, color-rich photographs. It really nails that style. You've done an amazing job. :D

[@MonzonMedia]

Great to hear and thanks for the feedback. Of course it can be improved in many ways which as mentioned we are already working on. Just the beginning my friend!

[@MyAmazingUsername]

@@MonzonMedia Fantastic to hear! :D

[@weebgrinder]

I've only been using PGV2 for a few days now and there's definitely a what appears to be a pretty significant jump in quality as far as what I'm getting through it versus SDXL. SD though still has an edge for certain things it seems to me like oil paintings for example. Enjoying this model.

[@MonzonMedia]

Great to hear it, PGv2.1 is very close to being released and will be much better than this one!

[@hleet]

WOW ! another new toy to play with ! :D It's so cool and ... a bit exhausting, I don't fully understand the IPAdapater and FreeU that I have another new thing to learn (again !). But thanks for the news ! Talking about news, there's that new MagicAnimate thing that can make danse one picture that you feed in.

[@MonzonMedia]

It's crazy how quickly things move, hard to keep up. I'll have to check out that MagicAnimate, been meaning too. As for the IP-adpater, I've been working on some content on that too.

[@hleet]

@@MonzonMedia yeah. Nerdy rodent (youtube) talk ajout it and even try on comfyui (not perfect but kind of work). IPAdapter is hard to start with, I got errors because when you download via comfyui manager, the model can be corrupted (happened to me), I downloaded manually and "worked" ... still trying to figure out this thing, glad that you will make a video on it :)

[@dennisherold7887]

"Monzon Media consistently delivers top-notch content on Stable Diffusion and Ai! 🚀 The depth of insight and clarity in each video is truly commendable. I appreciate how Monzon breaks down complex concepts, making them accessible and engaging for viewers. The production quality is fantastic, and the passion for the subject matter shines through. Keep up the terrific work, Monzon Media! Your videos are an invaluable resource for anyone diving into the world of Stable Diffusion and Ai. Looking forward to more enlightening content! 👏👍

[@MonzonMedia]

Thanks so much for the encouragement and kind words. Means a lot to me. 🙌👍

[@mindor2670]

Thanks... 👍

[@MonzonMedia]

You're welcome! 👍

[@MyAmazingUsername]

The results look beautiful. If I understood the description, it's an exact copy of SDXL's architecture, but the weights are trained from scratch instead of using their weights? Does that mean I can train LoRAs for it the same way I would do for SDXL's weights?

[@MonzonMedia]

Yes you've read that correctly, I wouldn't say exactly 100% but SDXL was definitely the main inspiration. Yes weights trained from scratch. As for Lora's I'll have to check on that for you but I don't see why it would be any different.

[@MyAmazingUsername]

@@MonzonMedia Thanks for clearing that up. It sounds really good. :)

[@MonzonMedia]

No worries, honestly it's great for the community as a whole. Our hopes is that this opens up new doors.

[@MonzonMedia]

@@petec737 built on similar architecture but trained from scratch. Besides isn't it better for the community?

[@MonzonMedia]

Not trying to debate my man just having a discussion, it's all good. Firstly we are talking about base models so civitai models of course will be better. The point of Playground V2 being based on the same architecture is so that training, models, loras etc would be the same process, except a slight change in the code. If it were to be brand new architecture no one could use it to train anything or use for anything else. I do appreciate your opinion as I like to see others perspectives.

[@76abbath]

I'm going to test it! Please tell me how to get straight lines in comfyui. Thank you for the video!

[@MonzonMedia]

Yes please do and let me know what you think. For the straight lines in the menu, top right click on the gear for settings, at the bottom link render mode, select straight and you're good to go!

[@timooothy1234]

So in comparison it is more brighter in colours, might use it to start a fashion page😅 ( Are there hand deformity???)

[@MonzonMedia]

Definitely brighter in general but I can tell you V2.1 will have better dynamic range and better multi aspect ratios. As for the hands I'd say it's comparable to SDXL. I think you'll also see PGv2 does background really well.

[@timooothy1234]

@@MonzonMedia Can you please include a feature for constant characters

[@Yxxx6]

What's the difference using comfyui vs fooocus? It all seems the same 🤔 is there any improvement between performance?

[@MonzonMedia]

Generally speaking Fooocus is a lot easier to use and straight forward where ComfyUI is node based but very powerful and versatile.

[@streamtabulous]

but not more than civitai sdxl models those are amazing, ai playground is great its the go to i send people too without hardware to run sd locally. some i tell to use firefox and Android in desktop mode to use the webui

[@MonzonMedia]

Keep in mind those are fine tuned models, this is a base model. Hopefully the community embraces it and creates some custom models from it. 👍🏼

[@streamtabulous]

@@MonzonMedia very true

[@b.radical]

Does it fail to render 80% of the time due to censorship? (This is why I canceled my membership to Playground)

[@MonzonMedia]

If you run it locally you should be fine. 80% is exaggerated man, come on. 😬

[@b.radical]

@@MonzonMedia It didn't used to be so bad, but lately I just find it unusable.

[@aguyandhisguitars435]

Can this be ran inside of fooocus?

[@MonzonMedia]

Yes of course, just download the file to your checkpoints folder and you are good to go.

[@JieTie]

This or Juggernaut XL ?? :)

[@MonzonMedia]

But this is a base model, juggernaut is merged model 😊

[@JieTie]

@@MonzonMedia thank you for clarification and sorry for such a newbie question, I'm new in this generative by AI graphic field. So correct question should be SDXL vs Playground ;) I'm gonna download today and play with it. Offtopic about prompting: when I generate with e.g. Fooocus, when I promped with only few key words I noticed in console that app generates random words to 'fill' whole text area. Is there a way to turn it off, or should I just came up with as many prompts as possible just to try to save consistency?

[@MonzonMedia]

No worries bud, I'm here to answer any questions if I can. Random words? Are you talking about the built in styles? You can turn those off and just prompt from scratch. Not sure if that's what you are talking about though. But for me personally. I try to keep my prompts simple as possible. I do have some long ones but they are only 77 tokens long. Anything more it get's ignored. Simple is always best, build on it more if needed.

[@JieTie]

@@MonzonMedia I also try to write short, straight to the point prompts :). But when I type e.g. "alien looking at the sky, night, forrest, stars" and try to generate image the I see my prompts followed by randomes like "space, monster, science fiction, high contrast etc."

[@MyAmazingUsername]

​@@JieTieThat's prompt enhancement. The app you use is trying to add more detail to the result. It is very useful for people who aren't good prompt engineers.

[@blender_wiki]

IMMO is not better that any custom checkpoint, like all the model he as strength and weakness. In my personal taste and usage i find is too biased in a "mediocre" general aesthetic: good to build a fast website or ahor for IG/TK, bad as an artistc tool.

[@MonzonMedia]

Appreciate the honest feedback. Keep in mind it’s not a “custom” SDXL model. It’s a base model trained from scratch. Built on similar architecture as SDXL so that existing tools can train it for Lora’s, checkpoints etc with minimal change in the process. v2.1 will be coming soon to improve it and polish it up.