【SD Frequently Asked Questions】Newbies ask questions

【SD Frequently Asked Questions】Newbies ask questions

March 10th, 2023
View OriginalTranslated by Google

Question 1: What is Stable Diffusion

Stable Diffusion is a text-to-image latent diffusion model created by researchers and engineers from CompVis, Stability AI, and LAION. It is trained using 512x512 images from a subset of the LAION-5B database. Using this model, any image including human faces can be generated, and since there is an open source pre-trained model, we can also run it on our own machine.

Question 2: What is the difference between Stable Diffusion and novelAI?

Stable Diffusion's training data set is LAION-5B, which has 5.8 billion "image-text" data captured from the Internet; a large amount of NovelAI training data comes from Danbooru, "Danbooru" is an image search and sharing website, which mainly provides There are many pictures from Japanese animation, comics and games. NovelAI uses Danbooru's pictures to optimize the model training (fine-tune) on the basis of Stable Diffusion.

In short, NovelAI can almost only generate animation, hand-painted, sketch, and CG-style pictures, but it has a higher output rate on these styles of portraits, and it is a professional player; Stable Diffusion is not good at generating animation in comparison Pictures, but can fully master photography, oil painting, watercolor, concept art and other styles, more like an all-rounder.

Question 3: How to use Stable Diffusion?

Just download, install and configure the corresponding environment, and you can run it. For details, see the local deployment documentation.

Question 4: Does Stable Diffusion have any requirements for hardware configuration?

The hardware requirements are quite high, basically a 3-series or above graphics card is required, and the video memory needs to be at least 8GB. The SD model also occupies a lot of hard disk space, at least 20-30GB.

Question 5: The hardware does not meet the requirements, how to use it?

You can choose to rent a computing power cloud server, which is very cheap and billed on demand. For example, you can use Google's colab to deploy Stable Diffusion for drawing.

Question 6: Is there any charge for Stable Diffusion?

Stable Diffusion is free, just install and deploy on your own machine to run

Question 7: What models are there for Stable Diffusion?

There are so many models, it is recommended to go to station C to search for the model you want.

If you have other questions, please leave a message for this note, and it will continue to be added:)


no dataCoffee time! Feel free to comment