Conversation

Notices

Embed this notice
konstruct (konstruct@woof.tech)'s status on Monday, 01-Dec-2025 02:13:27 JST konstruct
- Matthew Lyon
@mattly the mathematical process for genAI to create images is called diffusion. To my knowledge, all image generation models are diffusion models. It's basically a mix of adding and removing complex noise until it reaches a certain result. Some models use tricks like inline generation to generate part of the image at a time, similar to rendering quadrants of a 3D scene on a CPU, instead of doing it all at once to yield more coherent results. When you ask chatGPT to generate an image, your prompt can be processed by an LLM to make it work better with the diffusion model's dataset of labelled imagery. It's also the method for video generation, although other transformer models are used to improve object permanence and other flaws on recent models like Sora and Veo 3. But yeah, you have LLMs for text and Diffusion models for imagery.
Highly recommend the wikipedia article and its linked articles to get a good technical overview : https://en.wikipedia.org/wiki/Diffusion_model
In conversation about 9 months ago from woof.tech permalink
Attachments
1. Domain not in remote thumbnail source whitelist: upload.wikimedia.org
  
  Diffusion model
  
  In machine learning, diffusion models, also known as diffusion-based generative models or score-based generative models, are a class of latent variable generative models. A diffusion model consists of two major components: the forward diffusion process, and the reverse sampling process. The goal of diffusion models is to learn a diffusion process for a given dataset, such that the process can generate new elements that are distributed similarly as the original dataset. A diffusion model models data as generated by a diffusion process, whereby a new datum performs a random walk with drift through the space of all possible data. A trained diffusion model can be sampled in many ways, with different efficiency and quality. There are various equivalent formalisms, including Markov chains, denoising diffusion probabilistic models, noise conditioned score networks, and stochastic differential equations. They are typically trained using variational inference. The model responsible for denoising is typically called its "backbone". The backbone may be of any kind, but they are typically U-nets or transformers. As of 2024, diffusion models are mainly used...
- Embed this notice
  Paul Campbell (paulgc@mastodon.social)'s status on Monday, 01-Dec-2025 02:35:45 JST Paul Campbell
  - Matthew Lyon
  @mattly LLM is a slightly weird term but I think it usually means a transformer based language model.
  I think in the most part the image generation models are diffusion models (https://en.wikipedia.org/wiki/Diffusion_model), so perhaps it's diffusion generation model or just image generation model?
  In conversation about 9 months ago permalink
  Attachments
  1. Domain not in remote thumbnail source whitelist: upload.wikimedia.org
    
    Diffusion model
    
    In machine learning, diffusion models, also known as diffusion-based generative models or score-based generative models, are a class of latent variable generative models. A diffusion model consists of two major components: the forward diffusion process, and the reverse sampling process. The goal of diffusion models is to learn a diffusion process for a given dataset, such that the process can generate new elements that are distributed similarly as the original dataset. A diffusion model models data as generated by a diffusion process, whereby a new datum performs a random walk with drift through the space of all possible data. A trained diffusion model can be sampled in many ways, with different efficiency and quality. There are various equivalent formalisms, including Markov chains, denoising diffusion probabilistic models, noise conditioned score networks, and stochastic differential equations. They are typically trained using variational inference. The model responsible for denoising is typically called its "backbone". The backbone may be of any kind, but they are typically U-nets or transformers. As of 2024, diffusion models are mainly used...
- Embed this notice
  konstruct (konstruct@woof.tech)'s status on Monday, 01-Dec-2025 04:11:39 JST konstruct
  - Matthew Lyon
  @mattly what is that need if I may ask ?
  
  In conversation about 9 months ago permalink

Public

Conversation

Notices

Feeds