MidnightRunner commited on
Commit
c5fb512
·
verified ·
1 Parent(s): 31fe5bb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -2
README.md CHANGED
@@ -42,9 +42,16 @@ MIDNIGHT_NAI-XL_vPredV1 is a specialized fine-tuning of the NoobAI-XL (NAI-XL) m
42
 
43
  ## Usage Recommendations
44
 
45
- - **Dynamic CFG Plugin:** To prevent oversaturation or dull images, it's recommended to use the Dynamic CFG & CFG Rescale plugin with a configuration setting around 0.2.
46
- - **Sampling Methods:** While compatible with most sampling methods, V-prediction models do not support the Karras series. For optimal results, use Euler or DDIM sampling methods.
 
47
 
 
 
 
 
 
 
48
  ## Custom Workflow: MIDNIGHT1111_Chasm
49
 
50
  For an optimized generation process, consider using the custom workflow [`MIDNIGHT1111_Chasm 2025-02-02.json`](https://huggingface.co/MidnightRunner/MIDNIGHT_NAI-XL_vPredV1/blob/main/MIDNIGHT1111_Chasm%202025-02-02.json). This workflow is tailored to leverage the strengths of the MIDNIGHT_NAI-XL_vPredV1 model, providing a streamlined and efficient image generation pipeline.
 
42
 
43
  ## Usage Recommendations
44
 
45
+ **Sampling Methods:**
46
+ MIDNIGHT_NAI-XL_vPredV1 is optimized specifically for **Euler (normal)**.
47
+ Other samplers may not provide stable results, and **V-prediction models do not support the Karras series**.
48
 
49
+ **Model Sampling Method:**
50
+ For best performance, use **ModelSamplingDiscrete** with **V-prediction** and **ZsNR set to true**.
51
+
52
+ **CFG Scaling:**
53
+ Dynamic CFG Plugin is bypassed as backup in future need, **adjust CFG scaling manually** within a range of **5-6** for best balance.
54
+
55
  ## Custom Workflow: MIDNIGHT1111_Chasm
56
 
57
  For an optimized generation process, consider using the custom workflow [`MIDNIGHT1111_Chasm 2025-02-02.json`](https://huggingface.co/MidnightRunner/MIDNIGHT_NAI-XL_vPredV1/blob/main/MIDNIGHT1111_Chasm%202025-02-02.json). This workflow is tailored to leverage the strengths of the MIDNIGHT_NAI-XL_vPredV1 model, providing a streamlined and efficient image generation pipeline.