WASMoebius
A Jean Giraud (Moebius) Faux Style Textual Inversion
Usage:
Ending your prompt with the following segments is probably the best method. Name may differ for different versions.
art by WASMoebiusart by [(WASMoebius):0.5](Less effect)art by (WASMoebius)(More Effect)art by [((WASMoebius)):1.5](Extreme Ultra Mode)
Example Positive Prompts:
A portrait of a beautiful woman, art by WASMoebiusa sprawling desert with mesa plateaus in the background, art by WASMoebius
Example Negative Prompts:
(((3d, cgi, digital painting, smooth shading, vibrant), disfigured, bad art, deformed,extra limbs,close up,b&w, weird colors, blurry, duplicate, morbid, mutilated, [out of frame], extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, ugly, blurry, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, out of frame, ugly))(((3d, cgi, digital painting, smooth shading, vibrant)))
Notes:
It's a trip. Lol Jean has one hell of a style. Capturing line and color style is one thing, but his world is from all sorts of perspective, so cohesion is hit or miss. Far away people will likely need some sort of img2img on their faces due to the ink line size interfering with details, least from what I've seen.
The colors are far more flat, and thus accurate to Moebius in Stable Diffusion 1.5, but composition is pretty bad in vanilla 1.5 versus other models. So, I chose a custom model for sample images here, but all the models I have are all meant for photo-realism, so I had to negative weight a lot to get the depth toned down.
Description
FAQ
Comments (20)
My dream come true ! This will be like creating a graphic novel story in the art style from those 1980/1990s magazines of "Heavy Metal" and "Epic" comics! Does anyone still remember the film "Heavy Metal"?
Thank You very much ! 10 stars!
Oh yeah! Was planning a TI based on it. I actually tried on on Heavy Metal 2000 and it was OK. But the weird shading and 3D elements I think polluted the style too much for release imo.
Ex: https://prnt.sc/ig83OxiC75fT and https://prnt.sc/kiTZygPKp6HD
@WAS Those artsy images also look pretty cool too.
finally, someone tried the Moebius style in the correct way! thanks a lot bro! I'm a fan of this artist! could you do other artists? can i give you any suggestions?
Who do you have in mind?
These artists have always seemed interesting to me:
Marija Tiurine, Romain trystram, David Sosella, Igor Wolsky, Ben Bauchau, Butcher Billy, Charles Burns, Willian Santiago, Iain McArthur
Making one of those would be great!
By the way, I would like to try to create my own, how could I start in stable diffusion training? any videos on youtube that can help me? thanks @WAS !
@jesuelmdesign383 I honestly am not sure about tutorials, I just played around with settings until I nailed something that was relatively fast for my 4090.
I manually crop my images, downsize them to 512x512, then use AI upscaling, after that, I downscale back to 512x512. This helps with slight bad compression (though if too bad, just scrap the image altogether), and especially helps with game images as it's basically super resolution and gets rid of jaggies for the most part.
I use the Train tab in WebUI, and preprocess the images. Usually just check flip images, and BLIP Captioning. Then I go through and edit the captioning where it's bad or didn't understand the image.
Theeeen, I open up all the caption files in Notepad++, and use regex Search and Replace mode, then I search for "$" (no quotes) which is the end of the file. Then I use the "Replace in all open files" replace method with something like ", Art by Jean Giraud, pastel colors, line art, watercolor, highly detailed, muted colors". That will add that text to the end of every caption file. Then just Close All files, and save to all.
I train with a batch size of 4, with accumulation steps of 2, and a learning rate of 0.5 for 5,000 to 10,000 steps (my latest sweet spot). For 5k steps it only takes about 2 hours on 4090. For my latest TI's, I have been training with max vectors for the TI creation part, so 75 vectors per token for the embedding. I think this was done with 12 or 16 vectors per token.
---
I will take a look at these artists. I've added them to my request list.
@WAS thanks for your answer! It's really detailed! I still have to learn some things, I'm new to this! I'm researching on youtube for some tutorials! So far the community is growing and I'm getting a lot of information like yours!
I'm glad you liked some of my artists' proposals! I'll be following you to see how you manage to synthesize models for those artists!
@WAS hey bro look at this artist style! https://lilidesbellons.com/
@jesuelmdesign383 Oh that's interesting. Reminds me of the color tones and such of classic artists like Moebius stuff, but in a modern illustration style.
@jesuelmdesign383 Would you want a TI of that style?
@WAS I really liked what you could achieve with moebius! I would love to see what you achieve with a style like this! a TI would be great!
@jesuelmdesign383 Thanks, yeah I'll give it a shot. Seems plenty of resources there.
@WAS Yeah! that's why I thought of you! there is a lot of sample content! I like these type of artists because they have a somewhat more vector art style... something like that
@jesuelmdesign383 Training is underway...
@WAS you are the best bro! this will be very interesting! I admit that I downloaded all your Embeddings! The results so far are amazing! I love mixing them!
@jesuelmdesign383 Thanks a lot! Here is the TI: https://civitai.com/models/19323/was-lili-des-bellons
@WAS oh bro!!! it looks amazing!!! great job for real! I love being able to work with these artists styles who are little known but have a very unique and different felling! the results are great!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.