Add LoRA to Combine with DeepCache for Acceleration (#511)

This PR is done:

- [x] Add workflow.
After testing multiple trained LoRAs, apply the most relatively
appropriate DeepCache hyperparameters.
- [x] Add README.



![image](https://github.com/siliconflow/onediff/assets/54010254/e71e6428-8cf1-4b21-b607-ce141f4712a8)

For the Aether Cloud style of LoRA, it can be downloaded from
https://civitai.com/models/141029/aether-cloud-lora-for-sdxl.

onediff_comfy_nodes/README.md

@@ -135,6 +135,8 @@ Here are the example of applying DeepCache to SD and SVD models.
 
 [Module DeepCache SpeedUp on SVD](workflows/svd-deepcache.png)
 
+[Module DeepCache SpeedUp on LoRA](workflows/lora_deepcache/README.md)
+
 
 
 

onediff_comfy_nodes/workflows/lora_deepcache/README.md

@@ -0,0 +1,22 @@
+## LoRA to Combine with DeepCache for Acceleration
+
+We combined [OneDiff DeepCache](https://github.com/siliconflow/onediff/tree/main/onediff_comfy_nodes#deepcache) and LoRA, achieving more than a 2.7x acceleration for shapes of 1024*1024, workflow reference:
+![](lora-deepcache.png)
+
+For the Aether Cloud style of LoRA, it can be downloaded from https://civitai.com/models/141029/aether-cloud-lora-for-sdxl.
+
+
+#### Performance of Community Edition:
+
+Updated on Jan 11, 2024. Device: A100 GPU
+
+|                                      | LoRA | LoRA + DeepCache | Percentage improvement |
+| -------------------------------------------------------------- | --------------- | ------------------ | ---------------------- |
+| 512 * 1024 | 16.30 it/s        | 39.57 it/s           | 142.45%                 |
+| 1024 * 1024              | 9.10 it/s        | 25.19 it/s           | 176.04%                 |
+
+However, due to the characteristics of DeepCache, combining it with LoRA introduces some differences in the output:
+![](compare.png)
+
+
+Note: If your system or business requires more personalized support, please check: https://github.com/siliconflow/onediff?tab=readme-ov-file#onediff-enterprise-edition