This workflow creates an image using SDXL then refines it with FLUX and Face Detailer.
As the workflow evolved it became more broad allowing you to use Pony + SDXL + FLUX or really anything you can load as a checkpoint or UNET model, you can set it up anyway you see fit and use it in whatever way works for you. I've tried to keep everything fairly organized and interchangeable but if anything is confusing or not working feel free to leave a comment.
you can find the upscalers and luts here: https://freshluts.com/
https://openmodeldb.info/?sort=date-desc&q=skin
Skin Upscalers: https://openmodeldb.info/models/1x-SkinContrast-High-SuperUltraCompact
https://openmodeldb.info/models/1x-SkinContrast-SuperUltraCompact
Anti-Aliasing Upscaler: https://openmodeldb.info/models/1x-Anti-Aliasing
General Upscaler: https://openmodeldb.info/models/2x-classicalSR-DF2K-s64w8-SwinIR-M
My personal favorite lut: https://freshluts.com/luts/34
the only absolutely necessary upscalers are the skin and anti-aliasing one's, feel free to use whatever for the general Upscaler and LUT.
if you get this error "'NoneType' object has no attribute 'tokenize'". make sure to enable the lora stack, it has a toggle at the top. you can also bypass it by right clicking and selecting bypass if you don't want to use lora's.
Description
This is mostly the same workflow but I added two extra groups for Pony/SDXL to be used on either end of the process.
FAQ
Comments (8)
1) I keep getting this error with the negative prompt:
CLIPTextEncode
'NoneType' object has no attribute 'tokenize'
A search seem to suggest it's the something to do with the checkpoint but that's with SD3, not SDXL and I've tried it with several models, including RealVisXL and an EpicRealismXL one.
Nothing else has changed. Any suggestions please?
2) What should I put in the image loader? No image was supplied or available for download.
Try to bypass the two Lora nodes and just experiment with any image, I believe it provides a guide for the Perlin power fractal noise. Very cool workflow.
The 'NoneType' error is coming from the lora stack not being enabled, at the top you should see a toggle to switch it on or off.
For the image loader you can put anything you want in it, i recommend resizing whatever image you use to somewhere between 768x1280 to 960x1600 or reverse that for a landscape image. you can also use your generations in the load image node.
There is also an empty latent image node above the latent blend node if you don't want to use the img2img/perlin noise
don't forget to set the mask as well for the image, you can do this by right clicking the node and in the drop down it will be towards the bottom.
@soupman1 Thanks, I will try all that.
@jingalee Cheers, will give that a go.
thanks for the new version.
Np