Verified: a month ago
SafeTensor
Type | |
Stats | 508 339 |
Reviews | (121) |
Published | Jul 18, 2024 |
Base Model | |
Training | Steps: 5,000 Epochs: 5 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | HTDSBS 3 boxes 4 boxes 6 boxes 9 boxes Colored final box Photo in final box Mass drawing Stick drawing method Red colored added line |
Hash | AutoV2 CD919CF77A |
185
1.5k
V2 fixes some problems and help getting more consistent and relevant images. Next step will be choosing relevantly the numérique of steps. Here, most trzined images had 6 steps or 9 steps.
Use it at weight 0.8 to 1.2 or experiment anything else (and tell me if it works, so that I could share). I use Euler A or DDIM for this one.
I'm proud to propose this first step aimed to generate step by step drawing guides.
For v2: triggers: "howtodraw", "x steps", and then, the things you want to describe, "greyscale" or "colored final touch" if you want, then the style and so on.
For v1: The main triggers are "drawing of a (subject) in (x) steps", "sketch tutorial " , "step by step guide".
It is the first version, it won't be sufficiently accurate because it has been trained with 3 to 10 steps drawing sheets and I think it's too complicated to have a so good model with it. I made it as a proof of concept.
Discussion
Finally, everyone can make useless drawing tutorials now, with a bunch of circles and pretentious guidelines!
a very interesting lora, sometimes it works, sometimes it does not. it's wired, i used the sample img's prompt as input, in most of situations, i couldn't get what i wanted.
