IMPORTANT: this LoRA will affect your style
This LoRA is based on a Raven from 3 comics made by Shadaman (hence the name Räven), because i liked the style she was drew in them and i wanted to make LoRA with limited material available.
Since Shad drew Raven mostly naked in these comics, getting her in proper clothes may be a little hard. Cape,cloak and leotard are pretty easy to get, but red gems don't appear as constantly. The biggest problem is belt, since it is made only with these gems and puttin "belt" in prompts, ends up as regular belt on character. the only solution i could find is to inpaint regular belt with denoise strength 0.5-0.7
Tigger: ShadRaven
Weight: 0.8-1
Basic prompt: 1girl,solo,raven \(dc\),grey skin, red forehead jewel,purple hair
Clothes prompt: black leotard,blue cape, blue cloak, belt,gem
CFG/Steps: 8/30
Sampler: Euler A
Description
FAQ
Comments (9)
Thanks for this. It works very well with Yogirimix but seems to not do quite as well on other checkpoints. You ever decide on a V2 it would be great to get more flexibility. Still though, I'm very much enjoying it.
Which checkpoints did you used ?
I tried Galena and Meinahentai. Though it seems when I put in the full negative from your sample images I had an easier time getting closer to the style.
@Trefolder I checked how it works with meinahentai, i gave Lora weight 1.1 and prompts "grey skin","red forehead jewel" strength of 1.5 and 1.2, the results were much better. I usually work with Yogiri and i make LoRas for myself, so it works for me, but generally, before i release Lora i check how it will work with other checkpoints. It doesn't generate the style as well with other models, mostly the coloring is the same, but it's still useful as lora for Raven, that's why i decided to release it. Idk how to make loras more flexible, but if i learn it i will probably revisit Räven
@Alan_Rex Many thanks, I made those changes and it came out a lot better. A very good Lora, and I look forward to more of them from you.
I was unable to recreate any of the preview images using the generation parameters provided, was controlNET used for these?
will help youNope. I usually upscale from 512x768 to 1024x1536, maybe this
@Alan_Rex ???
@jackov5162872 yeah I was never able to figure it out, even using the exact generation parameters for the images embedded in them. I have every textual inversion and lora referenced in them, but can't get any good results.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.