Merge of different SDXL photo models and loras.
No refiner or upscaler was used on any of the images.
If you want a good looking Midjourney-like image with a short prompt and no negative prompt set clip skip to 2, but if you want more control over the image and you're using a negative prompt then you can set clip skip to 1.
Recommended settings for 1.0:
Sampler: DPM++ 2M SDE Karras or Euler
CFG: 6-10
Steps: 30 or more
Recommended settings for 1.5:
Sampler: DPM++ 2M SDE Karras or DPM++ SDE Karras
CFG: 6-10
Steps: 30 or more
Most of the prompts are originally from Midjourney 🤣
Description
Added some more models and loras in the merge, including a tiny bit of refiner model to add detail.
Better NSFW, photography and some details than 1.0
FAQ
Comments (4)
amazing. what width/heigh was this trained on?
MBBXL has the 'No sharing merges' permission displayed. The first version did not have that selected but it seems the creator changed their mind or realized it wasn't selected when they updated it - just a heads up 👍
🤔 I just realized I've been using SDXL with clip skip 2 by default in ComfyUI. I usually get good images with my merge. If you're getting bad images, maybe you should try it with clip skip 2
API keeps returning info about "v2.0", is it planned or private?