Text2Image with Inpaint, Face Detailer, Upscaler and optional Text2Video

Credit to: shikasensei (Ultra Upscale 1.5) & ControlAltAI for Face Detailing explanation on Youtube.

These are workflows for beginners to generate high quality Text to Image, and optional Text to Video / Image to Video if using SD 1.5.

SD 1.5 version I have not worked on for a little and will get back to soon.

SDXL version I'm working on with ComfyUI merging 2 SDXL model checkpoints with various loras for general fun at the moment.

Again, I'm just a beginner but having a lot of fun learning. So this should be relatively beginner friendly.

As always, the upscaler takes a while on either version - please adjust the settings if you don't want it upscaled as much. The information on how to do so is included by shikasensei's notation in their upscaler, edited to use the merged nodes names I have in my workflow with the SDXL version as well at the moment.

Many options for additional loras to be loaded, has been set for easy testing. Please always check the loras listing for proper use of strengths, as they can vary quite a bit with dramatic changes to final picture.

A lot of notes included about operation, and still will generate NSFW content. Use similar model checkpoints as always, as you can really wonk stuff up by going anime x realism etc.
***********************************************************************************************
SDXL NOTES:
-**_Currently just running with very little prompt for testing. I've been able to get some decent results with quite a different amount of merged SDXL checkpoints using dpmp 2m sde w/ karras. Can always change samplers and scheduler but it will change outcome a lot as well.

-I realize the embeddings in negative prompt don't work but they are for when I'm attempting to merge 1.5 with SDXL which works surprisingly well sometimes. More steps and cfg may be required, but can get some nice results. Have gone all the way up to 150 steps x 20 cfg with decent results on some merged checkpoints._**

-I've cut out the additional model load at the face editing to save on resources as using SDXL and still loading another checkpoint in the upscaler to finalize into whatever style of model you like best. VAE loads are optional throughout if you don't like one of the checkpoint models included, or require one with one. There is optional LoraLoaderModelOnly nodes in case you need a renderer lora or whatever else. Lora stacks are for the pose / details etc loras.

You can install most of the missing nodes with comfyui manager, through the install missing nodes / install custom nodes option. Same thing goes for models. The styler only shows up if you search for it manually, it is the SDXL comfyui styler.

****PLEASE CHECK THE CHECKPOINTS, LORA, AND VAE THAT ARE REQUIRED WITHIN EACH WORKFLOW OR USE WHICHEVER YOU THINK ARE BEST, THESE WORKFLOWS ARE TESTS TO MERGE EITHER SDXL OR SD 1.5 MODELS IN COMFYUI.****