Maintaining a stable diffusion model is very resource-burning. Please consider to support me via Ko-fi.
AingDiffusion and AingEXP will ALWAYS BE FREE.
===
AingEXP is part of AingDiffusion's EXPerimental model. For stable model, please visit AingDiffusion.
Because this is an experimental model, I will post every "interesting" changes here often. (a.k.a Ganbatte boku no GPU).
I will also post every major milestone as a new AingDiffusion version, so please also stay tuned there.
This model is not as stable as the main AingDiffusion model, and the art style will often changes. I hope you understand.
Guide to generate good images with this model
Use any negative textual inversion (e.g. EasyNegative), they will help you a lot.
Recommended samplers are "Euler a and DPM++ SDE Karras" for AingDiffusion v7.1 and up.
Hi-res fix is a must if you want to generate high-quality and high-resolution images. For the upscaler, I highly recommend 4x-UltraMix Balanced or 4x-AnimeSharp.
Set Clip skip to 2, ENSD (eta noise seed delta) to 31337 [to replicate image] and eta (noise multiplier) for ancestral samplers to 0.667.
Description
This version was merged using the ComfyUI merger.
It is the result of around 14 hours of learning ComfyUI.
The merge recipe is similar to AingDiffusion v8.0 (with different weights, CLIP, etc.), but instead of using hako-mikan's SuperMerger, it utilizes ComfyUI for merging. This allows me to adjust settings that I usually can't using SuperMerger.
This version is highly experimental and has been tested within ComfyUI. It will also be tested with AUTO1111 after its release.
FAQ
Comments (8)
"It is the result of around 14 hours of learning ComfyUI." - it looks like ComfyUI is difficult for people without knowledge of programming.
I'm a computer science student and it is still confusing for me until now.
@kayfahaarukku maybe you have a lot of extensions, so it's difficult to find right nodes. ComfyUI developer have a repository on github with workflow examples. All examples using only default ComfyUI nodes.
Comfy has the most atrocious and absolute garbage interface, but it works and is easily twice as fast in generations as automatic1111 is for me. I don't have a lot of the performance issues with it as I do A1111. I still hate using it though.
@Kaladae I think, you have no useful extensions for it. It's much better than A1111 garbage. You can even upload extension that replace node interface into something like A1111's interface.
@Shio_N Do you know what it's called? Auto1111 takes forever but at least I can find things with minimal effort. And comfy GUI is hot trash but it always at least works, unlike updates that break everything with the other. IDK, I'm a hardware tech, I solder and de-solder, I disassemble and rebuild, I hate playing with software.
I learned ComfyUI within about 1 day, but perhaps thats likely thanks to using Unreal Engine which has node type stuff too (called blueprints)
@Jatts_Art That has a lot to do with it. I used UE node modding once and hated it, but that's also because I've been text based/tree structure modding since early 2000's and never used blueprint style until maybe a year ago. If I took the time it wouldn't be so bad, but I still much prefer other UIs because of specific addons. With the Forge updates rolling into A1111 everything is much better performance-wise so I have little reason to look elsewhere right now. Comfy can be pretty pretty powerful for those willing to take the time to intimately understand it, but I have little time for extra things right now.