Instructions to use F16/z-image-turbo-sda with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use F16/z-image-turbo-sda with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("F16/z-image-turbo-sda", device_map="auto") - Notebooks
- Google Colab
- Kaggle
sda for krea2?
#8
by cushycrux - opened
It would be good like if you could create a sda lora for krea2.
Might keep poking at diversity research—D-OPSD, danceOPD, etc. look promising. Too many ideas, not enough compute, so it’ll be one thing at a time. No promises on results, but these problems bring me joy. So I’ll keep at it.
I use SDA in all my ZImage pipelines. An SDA generation phase + re-generate with Z Image Turbo at 0.5 to 0.7 denoise (depending on task) to fix errors. (Sometimes with a latent upscale step.) It is truly indispensable. A Krea2 SDA lora would dramatically increase the model's usefulness, as it depends on prompt variety for now.