大数据显示童书市场仍火爆：80后成消费主力军

Question

This question came to mind after reading other questions about the need of having separate tags for different models under the same technology.

Wikipedia defines Stable Diffusion as:

Stable Diffusion is a deep learning, text-to-image model released in 2022 based on diffusion techniques. It is primarily used to generate detailed images conditioned on text descriptions, though it can also be applied to other tasks such as inpainting, outpainting, and generating image-to-image translations guided by a text prompt.

So, technically, stable diffusion is a specific model that is based on latent diffusion techniques.

Stable diffusion also provides the ability to further fine-tune the initial model via embeddings, hypernetworks and Dreambooth models. As a result, many finetuned "models" came to be, with some sites specializing in their archival and distribution (see Civit.AI for example).

This means that you can generate images using the base stablediffusion checkpoint... or you can use a finetuned custom checkpoint (for example Anything v3 for anime related pictures, Dungeon&Diffusion for D&D related ones or Funko Diffusion for pictures of Funko style fugures). The problem is that users will still talk about "Stable Diffusion" in both scenarios.

The tag definition for [stable-diffusion] currently says:

In the context of GenAI, stable diffusion refers to a type of generative model that is used to generate images from text prompts. Stable diffusion models are trained on a massive dataset of images and text, and they can be used to generate realistic images of a wide variety of objects and scenes. "Stable Diffusion" is a latent text-to-image diffusion model that researchers at CompVis, Stability AI, and LAION have developed.

In this context, the first passage seems to consider "stable diffusion" of a supergroup of all models - finetuned or not - based on stable diffusion. The second line seems to refer to the specific checkpoint made by Stability AI.

How should the tag be used?

A basic use scenario could for example be:

• Questions that use just the base checkpoint should be tagged with [stable-diffusion]
• question that use a custom checkpoint that is different enough to make generic suggestions useless should be tagged [stable-diffusion] and [name-of-the-specific-checkpoint]

Would that be correct?

Answer 1

There are several approaches.

1. Rely on the tag wiki
2. General tag plus "child tags"
   • stable-diffusion + stable-diffusion-algorithm
3. General tag combined with other general tags
   • stable-diffusion + algorithm

While for a long time on a couple of sister sites, I had preferred the 2nd approach, after learning the opinion from other community members, nowadays I lean toward using the 3rd approach.

---

• Approach 1. Very few people pay attention to the tag wikis. This might require a lot of curation work.
• Approach 2. This will make be beneficial when making custom filters, and simple searches, but This might make it hard to find similitude across similar LLM and tools.
• Approach 3. This looks to require less curation work and might facilitate serendipity but might "polute" the watch notifications and might require more complex custom filter rules and searches.