Flesh4Fantasy

My first merged model. I've been merging checkpoints for fun for the past three months without any intention of publishing one. But I think it's so good that I have to share it.

I think it has a very good balance of color, contrast and brightness as well as sharp rendering even in one shot without a refiner. But using a refiner is a good idea, however you shouldn't use face and eye detailer since the refiner (using this model) does a fine job.

This model works very well in both Forge and Comfy, however in Comfy "Clip Skip 1" doesn't work as with many other models, but it works fine in Forge (SDXL's default is actually Clip Skip 2). So use Clip Skip 2 by default in Comfy. No need for a DMD2 4step lora.

Prompts recommendations:

Don't go overboard with weights, especially in Forge. I don't recommend using words like "masterpiece, breathtaking, 4K, award winning ... they are not necessary at all.

The default settings, in my opinion, should be:

In comfy:

* LCM / Exponetial or Karras --- Steps 11 --- CFG 1 or 1.1 --- Clip Skip 2 --- + use this model again in the refiner nodes with 0.3 or less for denoise (no face/eye detailer).

In Forge:

* Same as above but for best results you should activate the HiresFix tab using ESRGAN_4x and 0.3 or less denoise, and upscale by 1.4 or 1.5

* As well you should activate the Refiner tab and select this model and "Switch at" 0.8 or more.

My preferred resolution is 1088 x 1344 with this model. Even though one mega-pixel is what SDXL is trained on, I think it's a good compromise to get a sharper result from the start at the cost of an occasional rendering error.