Sampler

A sampler in text-to-image diffusion is a technique for generating images from text prompts using a diffusion model. Diffusion models are a type of generative model that work by gradually adding noise to an image until it is completely obscured. The sampler then reverses this process, gradually removing noise from the image until it is clear and recognizable.

• PLMS
• DDIM
• Heun
• Euler
• Euler Ancestral
• DPM2
• DPM2 Ancestral
• LMS
• DPM Solver (Stability AI)
• DPM++ 2m (Karras)
• DPM++ 2m SDE (Karras)
• DPM++ SDE (Karras)
• DDPM
• DEIS
• UniPC TU

There are a number of different samplers that can be used in text-to-image diffusion. Some of the most common samplers include:

• DDPM sampler: The DDPM sampler is a simple but effective sampler that is often used as a baseline. It works by randomly sampling noise from the diffusion process and then removing it from the image.
• PNDM sampler: The PNDM sampler is a more advanced sampler that uses a neural network to guide the sampling process. This can lead to higher quality images, but it is also more computationally expensive.
• Score-SDE sampler: The Score-SDE sampler is a state-of-the-art sampler that uses a neural network to learn the gradient of the diffusion process. This can lead to very high quality images, but it is also the most computationally expensive sampler.

The choice of sampler will depend on the specific needs of the application. For example, if the application requires real-time image generation, then a simpler sampler such as the DDPM sampler may be preferred. However, if the application requires the highest possible image quality, then a more advanced sampler such as the Score-SDE sampler may be preferred.