"Can I use Stable Diffusion to decensor NSFW artworks?"

That was a question that came from the depths of my soul.

DeepCreamPy has been discontinued for years, and while it still works to some extent, I really hope that the rapidly advancing generative models today can become a better alternative.

However, the community doesn't seem particularly concerned about this usage. All I could find were scattered discussions and vague instructions.

So what is the answer?

In short: YES.

Furthermore, powered by modern, powerful SDXL models, we can achieve better results and fine-tuning compared to our earlier DeepCreamPy.

Usage & Comparisons: https://github.com/Schwi-AI/DiffusionCreamPy

If you got better solutions for decensoring or tips for inpainting, please let me know!

I understand that, for Stable Diffusion users, inpainting is such a fundamental and commonly used technique.

And decensoring and inpainting are essentially the same process; the only difference lies in the visibility and confidence level of the context.

But for someone like me, just a few days ago, it was entirely new and unfamiliar magic. Since there are hundreds of diffusion models with various inpainting solutions, selecting the right one for your needs is not a walk in the park.

That’s the reason why I am here and make these workflows.

Bar Removal Workflow

Mosaic Removal Workflow

Basic Inpaint Workflow