New Playground v2.5 version.
Finally HGHD_PLAYGROUND_XL meets PLAYGROUND V2.5 though they are unrelated except by name they compliment each other well. This is the best version of the Base HGHD XL model yet. it does exceptionally well with text from prompting if you do it logically. It does an amazing job with realism and photo realistic outputs, it is very artistic and the output quality is unbelievable. none oth the test images are altered in any way. they are all exactly as they came out of generation. no edits no upscaling, nothing but the model my LoRAs and the prompting used to create the images. Give It a try and please like rate review and past your creations to the page. also tips don't hurt. if you like this model please consider showing the creator some buzz love!
< NEW CASCADE VERSION>
IMPORTANT READ
There are 2 models needed HGHD_PLAYGROUND_CASCADE LITE B and HGHD_CASCADE LITE C you must have both of these to use STABLE CASCADE though you can likely mix and match as long as they are b and c versions. I created both so they could work together for optimum outputs in the PLAYGROUND style. You will also need the cascade clip and the Cascade 'A' VAE Model the last 2 are just stable cascade versions though I may see about a HGHD version when things improve. I do have the full size versions but they are quite large files and will only do those if the lite versions do well. My testing was basically to see that it works LOL. so try it out. It does take getting used to using but the words work well once you find the right seed and fixed seed will be consistent. It does work in different samplers and schedulers but it definitely takes some trial and error to find your sweet spots. I foung the cfg and steps will vary depending on which you use. Seems fine with long and short prompts. Have fun and please plst to the galleries to show how you were able to use it!
I have noticed cascade doesn't like weighting--tends to break the model.
XL version of Original HGHD_PLAYGROUND model. I took some time to upgrade the basic PLAYGROUND model to be better out of the box. I did a couple merges of XL models in the same style as the original. though all are SDXL based. So A blend of artistic and realistic trending on realism. I find it works best between 2 and 5 cfg and 12 and 35 steps though you can push those as you like. It works with XL LoRAS and embeddings so you can tweek as you see fit. I use a wide range of samplers and schedulers but some will definitely do better than others. I like the KARRAS and EXPONENTIAL schedulers and the eular A and DPMPP_3M_sde_gpu and huenpp samplers give some cool outputs But Please experiment with what you like. And I would love to see and hear how it works for you . All feed back is appreciated and I take it seriously so I can improve going forward. Thanks and enjoy
* New Lightening version!
I merged The base PLAYGROUND_XL with SDXL Lightening 4step base. WOW is it fast lol. It is a bit challenging and takes some experimenting but in testing cfg from 1-2.5 were best with the lower end doing best. I would recommend starting a 4 steps. and if lacking detail or clarity increase as needed. Steps from 4 to 8 are fastest and extremely good when run in the Euler sampler with the sgm uniform scheduler. You can use other samplers and schedulers. If you do you may find settings you like more. if you increase cfg and steps above recommended you will get very different results and some just don't run, but it can do the higher counts just slows it down as if you were not using a "LIGHTENING" model lol. again you should try different settings and combinations to find what is working best for the prompt etc. It works well with LoRAS and embeds of various weights. I hope you have as much fun as I did XD.
