Recover detail or bring video games characters to life!   / discord  .
Designed for you to learn how to use comfy effectively and make use of essential nodes so that you can debug it on your own.

Help keep this channel brutally honest: https://kofi.com/stephantual

Workflow/models/etc (free, evidently) at:
https://comfyworkflows.com/workflows/...

▬ TIMESTAMPS ▬▬▬▬▬▬▬▬▬▬▬▬
00:00 Introduction
01:00 Studying Krea & Magnific features
04:00 Getting started and basic worfklow
09:00 Organizing the workflow and extensions
10:03 First result and observations
13:35 Adding Moondream for automatic prompting
16:42 Adding upscalers including CCSR
18:57 Saving and watermarking for convenience
20:30 Adding an update via model option
22:10 Cleaning up the workflow
23:40 Quite a lot of work left!
24:30 A new approach and organizing a large workflow
25:50 The model pipeline
27:00 ClipSkip, Selfattention Guidance (SAG)
28:22 FreeU_v2, DeepShrink, AutoCFG
30:20 Workflow structure & ContextBig from RgThree
32:05 Placing the upscaler first (and last)
32:54 Tiling system used in Magnific.ai
33:36 Adding color matching
34:23 Adding the facedetailler
39:05 Trying a large face and working on a per image basis
40:30 Adding an IPadapter to transfer style
45:40 Adding controlnets and aux preprocessor explanation
50:10 Calculating 'resolution' for the aux preprocessors
52:28 Tiled diffusion: why is it useful?
54:18 What could go wrong? Debugging
56:46 Examples of varied pictures
01:01:00 Conclusions and SUFIR release

▬ SOCIALS/CONTACT/HIRE ▬▬▬▬▬▬▬▬▬▬▬▬
Discord:   / discord  
All Socials: https://linktr.ee/stephantual
Hire Actual Aliens: https://www.ursium.ai/