I'm not really sure how to start this one...
The Problem
Recently (literally while I was in the middle of making practice generations this afternoon) Civitai rolled out more updates to their site. They introduced new resolutions and aspect ratios in the generation, which is great and something that I've been looking forward to.
They also got rid of the "draft" option for Illustrious models.
For those who don't use the on site generator, the "draft" option produced a minimum of four low quality images for a low buzz cost (typically between 6-8 buzz). This feature was intended to help with experimentation and testing out new prompts or models. Something I found while playing around with a few different models was that, when using img2img or hires fix, I could use those drafts as the foundation for a final image that matched my aesthetic leanings. It felt like the sort of hand-drawn style I would use if my hands would let me draw (IRL TLDR: nerve damage). Part of the reason I've stuck with Illustrious as most of the community has moved into Anima and Krea is because 1.) I can't generate locally, and 2.) Those models don't offer "draft" in the on-site generator.
How To: EmmiAI
Let me demonstrate what I'm talking about. Starting with a prompt (I'm not copying one here; my prompts are hideous but you can find them if you want lol), I generator several drafts, iterating on them until I get the result I wanted or one I like:
You can tell it's a draft because it's a little muddy and the facial features aren't fully defined, but it is a great foundation for an image. Think of the WIP images digital artists will use as the base layer to guide their final version. Note also the lighting and shadow effects. I like how subtle those are. They are intentionally made to be like that, and I want to keep them without them becoming too intense.
To produce the final image, I then use hires fix to produce a, well, Hi-Res final version. Again, my process is inspired by how I used to make art by hand before said hands betrayed me:
The final image is the same, but better. Higher quality, more refined lines, and it keeps what I liked from the draft. I've been making images on this account this way the whole time. It's madness but it works for me.
What It Looks Like Now
So now draft is gone. The foundational layer of my process has been stripped away. This is on me for assuming it would stay, especially since I am relying on tools that are not mine that I'm using in a likely non-standard way. I've tried reducing the steps to a minimum to mimic what draft does, but...
That ain't it. That's not what I want to make. I'm not sure what to do next. I could make a style LoRA for one of the more popular models, but those are buzz-expensive and just punts the learning process down the road. I'm sure I can find a way to recreate my look, but until then I'm dead in the water.
I'm trying not to be dramatic. I've done that enough. But this might be too tough a nut to crack.