Help me make better models!:https://ko-fi.com/yottaart
Settings:
Sampling method: DPM++2M Karras or DDIM
Sampling steps: 20-30
Upscaler: R-ESRGAN 4x+ Anime6B or Latent (nearest-exact)
Hires steps: 10-15
Denoising strength: 0.4-0.6
CFG Scale: 6-10
Vae: Baked-in
Description
Reupload of Yotta Mix v1.0
FAQ
Comments (6)
Hello my friend, can you create this beautiful style
Hello there! I could certainly try it, but it'll take some time as I use Collab for training!
these results are great but the negative prompts in your examples stress me out. Why so long? There's sooo many duplicated words, it's hard to sift through it all and figure out what negative embeddings you're using too.
And these models have vector limits, so past a certain point those negative prompts aren't doing much anymore. Not saying they're not doing anything, but the way they're laid out now it's difficult to actually control anything.
I actually didn't know that! Thanks for letting me know, i just use these prompts since i started on pixai, a friend of mine recommended, but i really need to clean them up.
yeah your model is actually robust enough that you don't need any
@Yottax what Bullseyetroll said! So many people are still using wild negative prompts to try and negate issues that came along with the early models from last year, but most of the newer models have no need as they're trained to avoid those things already. Like disfigured faces. Some newer models are even getting better at hands without any negative prompting.
I'd suggest even starting there with your models to see what weaknesses actually exist and then build your negative prompt from there :)