Accelerating SDXL 3x Faster with DeepCache and OneDiff

Make SDXL run 3.5x faster on RTX 3090 and 3x faster on A100.

DeepCache was launched last week, which is called a novel training-free and almost lossless paradigm that accelerates diffusion models from the perspective of the model architecture.

Now OneDiff introduces a new ComfyUI node named ModuleDeepCacheSpeedup (which is a compiled DeepCache Module), enabling SDXL iteration speed 3.5x faster on RTX 3090 and 3x faster on A100.

Here is the example: https://github.com/Oneflow-Inc/onediff/pull/426

Run

ComfyUI node name：ModuleDeepCacheSpeedup
You can refer to this URL on using the node：https://github.com/Oneflow-Inc/onediff/tree/main/onediff_comfy_nodes#installation-guide

Example workflow

Depending

1. The latest main branch of OneDiff: https://github.com/Oneflow-Inc/onediff/tree/main
2. The latest OneFlow community edition:

cuda 11.8:

python3 -m pip install --pre oneflow -f 
https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu118

cuda12.1:

python3 -m pip install --pre oneflow -f
https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu121

cuda12.2:

python3 -m pip install --pre oneflow -f
https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu122

Welcome to join OneDiff Discord group to discuss related questions.