Introduction
Neta Cat Tower is a text-to-image model fine-tuned from NetaYume Lumina.
This model was trained with the goal of enhancing anime style.
No learning was conducted regarding the addition of characters.
Model Components
Diffusion Transformers (DiT): This model
Text Encoder: Pre-trained Gemma-2-2b
AutoEncoder: Pre-trained Flux.1 dev's AE
"all_in_one" is a single model that are combined with DiT, text encoder and autoencoder.
The model uploaded to Civitai is a "all_in_one" DiT model.
(I changed the uploaded model file from all-in-one to DiT to reduce the download file size.)
If you want to get all-in-one model, please download it from my Hugging Face page.
How to Get Started with the Model
Please refer to the Neta Lumina's model card.
You need to use the webui that supports Lumina Image 2.0.
ComfyUI
Forge Neo
Recommended settings
Sampler: res_multistep/ euler_ancestral
Scheduler: linear_quadratic
Steps: >=30
CFG (guidance): 4 – 5.5
Resolution: 1024 × 1024, 768 × 1532, 968 × 1322, or >= 1024
Prompt
Please refer to the Neta Lumina Prompt Book
About character knowledge, please refer to the NetaYume Lumina's Civitai page
Training Information
Please refer my Hugging Face page
Acknowledgments
duongve: Thanks to duongve for sharing awesome model.
Description
FAQ
Comments (13)
guys, what except comfy can work with that?
@compgamer1337267 Currently, only ComfyUI supports Lumina Image 2, but it seems that Forge Neo will also support it in the future. https://github.com/Haoming02/sd-webui-forge-classic/issues/290
Support on Forge-Neo is now live :)
@HaomingGaming Excellent job!
A masshigura fine tune on a non-SDXL model? Now we're cooking!!
Активно жду, когда из Lumina сделают что-то приемлемое. Кажется, эта модель - верный путь к светлому будущему, когда у пользователей будет альтернатива SDXL
Isn't Lumina 2 already usable? It is
@qek In terms of quality, I think it is still inferior at this point compared to models such as illustrious and NoobAI, which have more advanced fine-tuning.
@qek I haven't tested this specific model, but based on the images provided, it looks relatively good and certainly better than what's currently being built on Lumina 2. Overall, even if I get the same quality as shown here, it's still not enough to make me stop using my favorite checkpoints based on Illustrious and NoobAI, even though Lumina allows you to actually write prompts in natural language. Despite the strong claims that NoobAI can handle natural language, that's a blatant lie in my opinion. It works to a small extent, but I still believe the best way to work with SDXL is to use specific concepts the model is familiar with from the annotations in the training data.
It's great to see someone trying to train a new base anime model.
wait this only 4gb ? feels unreal xD
@Seii1 The uploaded model is DiT model only. You need Gemma-2-2b Text Encoder and Flux VAE separately. (Total sum size of 3 models is about 11GB)
@nuko_masshigura yes, already have those, btw thx for making this