Welcome to LimitlessVision! I am SilasAI6609. I am the creator of JadeDefiant on Mage.Space. After months of requests to make a public model, I finally caved in. This model is very similar in character to JadeDefiant, but it has it's own unique flavor. LimitlessVision will be an ongoing project in tandem with an upcoming SDXL release as well. It has been uniquely trained to perform well at low, med, and even high res generations! It removes the BIAS on sexual content! Yup, if you want a scantily clad female, male, or somewhere between, it is all baked in. The trade off is limited capabilities in anime and furry styles.
Please feel free to use, abuse, and enjoy my work!
Description
Added over 5,000 images in training to improve general quality of life. Gen higher 768+ is much more stable. More style variation (still prefers photorealism). Better plumbing for males and females. As always, this is a continued effort. Even now, I am setting up another number of datasets to continue training. Thank you all for trying out LimitleVision. It has been a pleasure speaking with many of you, and look forward to more amazing input from the community!
FAQ
Comments (8)
Ruh oh, this doesn't import right for whatever reason. Shows the blank default thumbnail and only lists it as "Model (A0F9F9B5B2)" -
A0F9F9B5B2 was the name of the beta. It is the same model though. Here is a link to it from Huggingface, it may work better for some reason.
https://huggingface.co/SilasModder/LimitlessVision03/resolve/main/LimitlessVisionV003.fp16.safetensors?download=true
@SilasAI6609 Gracias, amigo!
@SilasAI6609 I think I'm just gonna have to put a thumb and title in manually. For some reason once Mage imports a model as a model number and blank image - it always will for that same model regardless of the source. No worries, I'll get by with A0F9F9B5B2!
@Picky2011670 correct, any imported model, may it be checkpoint or LoRA, Mage will rename it. It may be the hash name, not sure. I use import to test models while my GPU is tied up training, and I have never has it carry the true name.
@SilasAI6609 Yep! Works great so far - just the model name is off. Haha, I don't even know what kind of power my gpu has as I'm on a mac. Gasp! A mac!
With version 3, it seems that when you request an image of two individuals of different ethnicities (for example, when prompting “1 Asian guy and 1 white guy”), the outcome is either a single person or two people. However, the faces of these individuals appear to be a MIX of the two ethnicities, rather than one person fully embodying one ethnicity and the other person fully embodying the other.
In contrast, with version 2, to my memory, the model was more likely to create either two people of the same ethnicity or one person of one ethnicity, but never a mixed face.
Aside from this, I haven’t noticed much difference in version 3 so far, but I want to say that overall, this model is excellent.
Thank you for the input. V3 is a quality of life update more than anything. It is more formulated to gen 768×1024 and variants near those resolutions. With any training, there is a chance to have odd effects in the neural network. I will play with it and see if it is more of an issue with the training, or it just adjusted how it wants to be prompted.