Mobius: Redefining State-of-the-Art in Debiased Diffusion Models
Mobius, a diffusion model that pushes the boundaries of domain-agnostic debiasing and representation realignment. By employing a brand new constructive deconstruction framework, Mobius achieves unrivaled generalization across a vast array of styles and domains, eliminating the need for expensive pretraining from scratch.
Made By [https://corcel.io/]
Domain-Agnostic Debiasing: A Groundbreaking Approach
Domain-agnostic debiasing is a novel technique pioneered Corcel. This innovative approach aims to remove biases inherent in diffusion models without limiting their ability to generalize across diverse domains. Traditional debiasing methods often focus on specific domains or styles, resulting in models that struggle to adapt to new or unseen contexts. In contrast, domain-agnostic debiasing ensures that the model remains unbiased while maintaining its versatility and adaptability.
The key to domain-agnostic debiasing lies in the constructive deconstruction framework. This framework allows for fine-grained reworking of biases and representations without the need for pretraining from scratch. The technical details of this groundbreaking approach will be discussed in an upcoming research paper, "Constructive Deconstruction: Domain-Agnostic Debiasing of Diffusion Models," which will be made available on the Corcel.io website and through scientific publications.
By applying domain-agnostic debiasing, Mobius sets a new standard for fairness and impartiality in image generation while maintaining its exceptional ability to adapt to a wide range of styles and domains.
Surpassing the State-of-the-Art
Mobius outperforms existing state-of-the-art diffusion models in several key areas:
Unbiased generation: Mobius generates images that are virtually free from the inherent biases commonly found in other diffusion models, setting a new benchmark for fairness and impartiality across all domains.
Exceptional generalization: With its unparalleled ability to adapt to an extensive range of styles and domains, Mobius consistently delivers top-quality results, surpassing the limitations of previous models.
Efficient fine-tuning: The Mobius base model serves as a superior foundation for creating specialized models tailored to specific tasks or domains, requiring significantly less fine-tuning and computational resources compared to other state-of-the-art models.
Recommendations
CFG between 3.5 and 7
3.5 for extreme realism and skin detailing
7 for artstic, anime, surrealism, and so on.
Requires a CLIP skip of -3
Sampler: DPM++ 3M SDE
Scheduler: Karras
Steps: 50
Resolution: 1024x1024
Please also consider using these keep words to improve your prompts: best quality, Masterpiece
Description
FAQ
Comments (13)
The creator chose to remove and repost this model under a different user name, so I am going to more or less reproduce my original comment here:
The lack of specificity and overall hype-isheness of the description made me skeptical. Despite this, I decided to give this model a try and my skepticism was unfortunately validated.
To the extent this model is appreciably different from any other, it feels worse IMO. Despite using the recommended settings, I find that the photorealistic images are worse than many other photorealistic models; and some of the other styles are worse than either artistic and photorealistic models. Notably, the overall image coherence seems worse and the people tend to have a somewhat plastic appearance that is reminiscent of DallE3. Hands are also often worse than other recent SDXL-based models.
The model does seem to be fairly flexible, so I'll give it credit for that. But there are a lot of other flexible models out there that I think do a better job. Even DreamShaper Lightning seems to give consistently better results across a variety of styles I tried, including photographic, cartoon, and oil painting.
As discussed on Reddit, it appears that this model is intended to serve as an advertisement/lead-in for people to use Corcel, which is in turn making money off of a crypto project called bittensor. This explains, in part, why the description for this model is full of bold claims and repetitive promotional language that don't make much sense: "constructive deconstruction," "fairness and impartiality," and four different mentions of how it's better than the "state of the art," among other things.
Even more worrying, when people tried to clarify the claims being made and language being used by the creator, they become outright hostile to people asking questions. I wasn't able to see the creator's full response to my comment on the previously uploaded model before they deleted it, but based on the snippet I could see, it seemed to be characterized by a similar hostility.
For all these reasons, I simply cannot recommend this model and would advise caution when making use of the services advertised.
Set aside all the drama and hate on the reddit community. I really couldn't care less if this is trying to sell anything, I'm evaluating a model released for free...
Testing it (not extensively enough) it gave me amazing results with prompt following. I was using 2 loRas + a complex scene (a character LoRa using specific clothes LoRa running from a monster kraken inside a ship), and maybe I didn't test it enough, but it outperforms all the other realistic models I tried. Most don't generate the monster, or might fuse them with the character. Mobius did somewhat correctly most of the times. Of course it needs more testing, by I'm liking it so far.
I think the understanding of the prompt words by the Mobius exceeds that of other models.
Its not bad and its not good. While it does look different than other models more or less, it needs a lot of work, people are distorted, buildings look good with nice detail, but they also merge in mashed ways.
Steampunk has a nice arrangement with nice lighting, but also has that rushed circle look where its all twisted up close. I get a lot of weird angles that have a snap shot look, and then I have a nice looking still life, with weird objects. Like I wanted books and it gave me book planters? And I wanted bottles but i got books and there were no mention of books. Many of the scenes seem very cluttered, too many objects in the screen. I'm still testing it though, I like it for the still life, and I can see the potential, but its not polished yet.
I am starting to train lora with this model, let's see the results after a few hours.
I don't read reddit, as long as a model's good I don't really care lol. I mean honestly, we're all here trying to sell something, and i've seen others comment outside of this that it's a good model.
Overall a solid model. I would row it with OpenDalle and model1 when it comes to prompt accuracy. It sticks like glue to the prompt what makes creating scenes easy. As you see on the Big Lips Woman I posted in the gallery, it does work with loras trained on the base model - very nice. Sometimes I had deformations on hands or face but......on the other side if the picture comes out nice...it is very nice...so thumbs up for some fresh new prompt creations. Keep up your good work.
Any special VAE I should use? I'm getting quite crunchy results from it. Very sharp and colorful. I've played around with CFG from 3 > 8. Not quite sure what I'm doing wrong... Posted a few images below.
is this the same as i download from huggingface ?
You should consider losing up the license and let people generate on Civitai generator. It's a good model.
The model has an impressive art style and personally I can't stop enjoying it. My best wishes to creator and kudos.
A crazy model, I hope some day we get an uptade
In my opinion a very good and interesting model! Thank you author! Will there be Flux?)))
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.