This repository contains a model that generates highly aesthetic images of resolution 1024x1024, as well as portrait and landscape aspect ratios.
Originally uploaded to https://huggingface.co/playgroundai/playground-v2.5-1024px-aesthetic
Please note the Playground AI Community License
Description
FAQ
Comments (72)
Hi Ally - thanks for this. I'll be interested to give it a try. Gentle request for future variations: Is there any need for such a long filename? The word "playground" appears three times in it.
Yes, because it's by Playground AI, and they named it "Playground"
You can rename it, I shortened it to pgv2-5, works fine
please forgive me if this is a stupid question but will this work in automatic1111?
Not at the moment but it will eventually, currently there is comfyui support, just update it and use this .json https://gist.github.com/comfyanonymous/6275d1df67f402dc053e3d6991ebe201
Took a while to get the prompting down, but its nice when ya get it... EDIT: After getting a few tips in here from @Monz This is less difficult than before. Simple prompt works.
What is special about the prompting for this, please?
@fauxrealistic I was having problems when I first started using it and I was getting over saturated colors and dark blotches. Worked on the prompts and then the images became clear and don't have issues. I mean it IS SD and all. Prompting is part of it.
Also works best with lower cfg 2-4 in most cases, 30-50 steps and DPM++ 2M Karras is recommended
@Monz Had I followed correctly I may have found that sooner. Thanks for the tip. Check the edit above.
@totes No worries! That's how it's set up on playgrounds site, the model was designed that way for people just to focus on prompting and creativity. It does favour simple prompts, complex is possible however you will find some of your SDXL prompts will have to be restructured or re-worded in some way. I'm still discovering a lot myself.
Is the quality difference between fp16 and fp32 significant? Thanks
literally the same
@Eddie_ oh well! Thanks! 👍
I've searched and searched.
Can somebody please tell me where I can download the ModelSamplingContinuousEDM: edm_playground_v2.5
Thanks in advance.
Great contrast but it does not work with regular sdxl loras... Waiting for a lighting/LCM version :P
I get images with only colored static noice. Do I miss something
Yes, Update ComfyUi ;)
It's currently for ComfyUI only, if you try to use it on automatic1111 it will just make that random noise
@AIArtsChannel thx!
It needs EDM sampling, autodetection for it was added in ComfyUI on 2024-02-28.
@MoonRide I installed the latest version of ComfyUI and the problem is there there. Can I enable EDM sampling myself?
@harolds you have to use a specific node for EDM. you can use this workflow https://gist.github.com/comfyanonymous/6275d1df67f402dc053e3d6991ebe201
It works with Fooocus too.
@codegix the problem is you can't use any of the advance features with it like faceswap etc
any recommendations on how to get good results in the prompting department?
Are Loras useable with this with or without latent imposer?
I just tried using a lora but my image was extremely faded. Perhaps not all loras work well with this? In the comments someone remarked that this doesn't work well with SDXL loras.
You need to retrain the models.
anyone know any img2img workflows for this?
ComphyUI or Automatic1111? I have the workflow saved.
@winstondallen both, didn't know automatic had compatablity
this gives noise results in automatic. I'm doing something wrong?
Automatic 1111 and Forge will (probably) support Playground 2.5 some day later
@OliviaRossi thanks awesome :) this looks great can't wait to try it
@dilectiogames then - use ComfyUI - it works for a week now with this model
Which is better, stable cascade or playground 2.5?
Id say thia by the looks of the images below, I find cascade quite limit in comparason to whats about these days
i can get this model to produce images on its own, but i cant get it to merge with other models in comfyUI, i only get colorful noise when i try to merge models. is it possible to merge this with SDXL models?
I don't believe so. It has a few differences from the regular SDXL Pipeline, especially with the VAE.
How to train Lora
truly awesome, thank you
Incredible
this model could not be used with controlnet？
I am assuming No answer is "No" answer
still not working on forge?
that looks like a nice model bro, id like to give it a run . I use forge only now days. i noticed a few comments below about this on forge. Has that be resolved if so ill ill take it for a spin
A very underrated checkpoint with high potential. Its results are much more interesting than those of the basic SDXL and other models based on it. A large number of tokens are not required to obtain an aesthetic image. The style is more similar to Midjourney4. She, like the basic one, has quality problems (especially the eyes). But I think it would have been fixed quickly if the community had finalized it. It's a pity that almost no one was interested in her. I think this is due to the lack of support from the familiar A1111 and Forge interfaces.
can i run this model in automatic 1111. becuase whenever i try to it just gives gibbereish
only ComfyUI for now
@Nachtbringer thanks
You can try with Fooocus now
@koxpl2 will try thanks
Try in Stability Matrix -> Inference. UI is simple but does the work.
nice
If someone does not know, you can use this model with Fooocus now
in new update , yes
could u please explain how to use it with foooocus? just do the inpainting?
【pros】good @ digital paintings, hands are ok,
【cons】low tokens to be used/missing compared to XL, bad @ mixing styles
Does this work with A1111 by chance? Curious to give this a go!
The issue I've been having is it ignores anything in the advance tab in Fooocus, can't do faceswap or pyracanny for poses, any thoughts?
waiting for v3 OwO
How to upscale images in fooocus, using this model ? If i do not use any upscaler ( don't have one) - upscaling damages image, ruins dress or hair. How to fix it? Which upscaler will work fine in fooocus in that case?
Love this model. Shame that PGai went a different direction, but got to earn a living I guess.
Bummer that it became a kids' playground
@ChristyCanyon nah, it is just an outdated finetune of SDXL 1
Use AMD Amuse AI filters with Playground!
Can I animatediff this with sdxl motion model or hotshotxl?