DPM2 Ancestral

DPM2 Ancestral is a diffusion model sampling algorithm that is based on the DPM2 solver. It is an ancestral sampler, which means that it uses information from previous sampling steps to generate the current step. This can lead to more accurate and detailed results, but it can also be slower than other sampling algorithms.

DPM2 Ancestral is typically used to generate high-quality images from text prompts. It can also be used to generate other types of creative content, such as music and code.

Here is a brief overview of the DPM2 Ancestral algorithm:

  1. Start with a noisy image.
  2. Repeatedly denoise the image using the DPM2 solver.
  3. At each step, add noise to the denoised image using an ancestral noise schedule.
  4. Continue until the desired level of detail is reached.

The ancestral noise schedule is what makes DPM2 Ancestral unique. It is designed to allow the model to learn from previous sampling steps and generate more accurate and detailed results.