High res (+ NSFW and controlnet canny workflow) (Patreon)
Content
Trying a new workflow using the controlnet canny model, more info below if you're interested.
------
Initially I didn't try using it because I thought I would need to make a really good sketch or use 3D programs I'm not very good at, then I tried using some other generations I've made as input and got good results.
I'll explain more in detail how I use it:
The idea is pretty simple, first generate images focusing only on the pose and angle, then use that as input for controlnet canny so that you can add keywords for the character and other parts of the image without interfering with the composition.
The main reason to do this is if you're trying to make something specific and want way more control. This solves that in many cases.
Here's an example:
I made image N.1 with txt2img by only describing the act (without including many character, quality or location keywords so that it was easy to get), then I used it as input for controlnet canny to generate image N.2 (where I added the keywords for the character). Image N.3 was an extra step, where I used N.2 as input. Then I finished making it with my usual img2img/inpaint/Photoshop workflow...
------
A few more things:
- I might still use the scribble model to get the first image.
- I always avoided Lora models for poses because they usually change the style too much or decrease the quality, but now I might consider trying them if I need some extra help getting the pose (just for the initial image though).
- Over time, I'll try other controlnet models to see what else I'm missing.
- I might have to make a new workflow video some day, since it changed so much since my last one.