|
|
--- |
|
|
license: other |
|
|
license_name: newbie-nc-1.0 |
|
|
license_link: LICENSE.md |
|
|
language: |
|
|
- en |
|
|
pipeline_tag: text-to-image |
|
|
tags: |
|
|
- next-dit |
|
|
- text-to-image |
|
|
- transformer |
|
|
- diffusion |
|
|
- NewBie |
|
|
--- |
|
|
 |
|
|
This is the Hugging Face model repository where NewBieAi Lab stores each epoch of the NewBie image v0.1-exp model. |
|
|
|
|
|
We have achieved some initial promising experimental results, but it is still not perfect. |
|
|
|
|
|
If you would like to test it, please contact the NewBieAi administrator. We may grant you access and testing permission after evaluating your situation. |
|
|
|
|
|
Basic Information: |
|
|
|
|
|
NewBie image v0.1-exp is a new arch model that use the Lumina-image2.0 arch as its research target. |
|
|
|
|
|
Anime Type (we chose a neta lu2 model trained on lumina arch to further research this type of task) |
|
|
|
|
|
>Text Encoder: |
|
|
|
|
|
Google/Gemma3-4b-it |
|
|
|
|
|
Jina Ai/Jina Clip v2 |
|
|
|
|
|
>Networks: |
|
|
|
|
|
Next-DIT 3.5b (modified from 26 layers to 36 layers) |
|
|
|
|
|
>VAE: |
|
|
|
|
|
Flux 1 Dev-VAE |
|
|
|
|
|
---------------------< |
|
|
|
|
|
Pretrain Dataset: full dan + 1m e621 |
|
|
|
|
|
Use 8*h200 train for three months (17500 h200 hour) |
|
|
|
|
|
Restructure text data use XML format |
|
|
|
|
|
The model is currently in the train phase (100% complete). |
|
|
|
|
|
Overall progress is approximately 95%. |
|
|
|
|
|
The model is expected to be open sourced on Hugging Face on December 7, 2025. |
|
|
|
|
|
example： |
|
|
 |
