Preface
When we use Stable Diffusion to draw, ordinary users often suffer from the problem of insufficient video memory and slow drawing output because the computer's video memory configuration is too low. SD-WebUI is not as convenient as ComfyUI and Fooocus in terms of memory optimization, but there are some solutions to make up for the memory problem of SD-WebUI, namely LCM and FP8.
LCM Tutorial
Introduction
LCM is a new sampler for Stable Diffusion models. Stable Diffusion is a generative artificial intelligence model mainly used for text-to-image conversion, that is, generating corresponding images based on given text prompts. The sampler is an important component in this model, responsible for drawing samples from the probability distribution to generate the final output.
LCM stands for "Lora Consistency Model", where Lora is a new sampling method and Consistency Models emphasize maintaining consistency during the generation process. This model claims to significantly improve image output speed, for example, it can generate 4 images in a very short time (such as 3 seconds), and may be 10 times faster than the previous generation model. In addition, LCM requires the addition of specific Lora parameters when used under the WebUI, the specific choice depends on the large model used.
Typically, this process requires a large number of iterative steps (Steps), which may result in a longer time required to generate an image. As a new type of sampler model, LCM aims to significantly reduce the number of iterations required while maintaining the quality of the generated images, thereby increasing the speed of image rendering.
It is described that the LCM model