Not-For-All-Audiences
Sentinel7 commited on
Commit
0e11f0c
·
verified ·
1 Parent(s): d764901

Upload 2856467/3266628/README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. 2856467/3266628/README.md +20 -2
2856467/3266628/README.md CHANGED
@@ -11,7 +11,7 @@ tags:
11
  - woman
12
  ---
13
 
14
- # [MMH3] Mystic XXX - v4.0 (FL2VA & REF2VA) - 3266628
15
 
16
  **Model Type**: LORA
17
 
@@ -46,6 +46,24 @@ tags:
46
 
47
  ## Description
48
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49
  New ref2va file to download (experimental, no audio trained)
50
  ============================================================
51
 
@@ -67,4 +85,4 @@ Model: [CivitAI Model Page](https://civitai.com/models/2856467?modelVersionId=32
67
 
68
  Archive: [CivArchive Page](https://civarchive.com/models/2856467?modelVersionId=3266628)
69
 
70
- <!-- Version: 20260831_upload -->
 
11
  - woman
12
  ---
13
 
14
+ # [MMH3] Mystic XXX - v4.0 - 3266628
15
 
16
  **Model Type**: LORA
17
 
 
46
 
47
  ## Description
48
 
49
+ V5 - REF2VA
50
+ ===========
51
+
52
+ This version was refined from the FL2VA using only MLP from 10-49 to reduce interference with other loras and references, it has the same audio as FL2VA and was retrained to gain reference capabilities that FL2VA don't have, for this a dataset with references and goals is used. It was made with Musubi Trainer.
53
+
54
+ V5 - FL2VA
55
+ ==========
56
+
57
+ This version was trained with audio using the full layers spectrum, including both MLP and attention layers, with Musubi Trainer.
58
+
59
+ Compared to V4, it delivers more natural skin rendering and lighting, improved movement, more realistic results, and better audio quality. The dataset was increased by approximately 50%, and the training resolution was raised to capture more detail. Anatomy is also slightly improved, with better prompt adherence in some cases.
60
+
61
+ I tested it against V4 across more than 100 videos, and this version produced better results in approximately 70% of them. The validation results also improved, with a particularly significant improvement in T2VA performance.
62
+
63
+ V5 is based on FL2VA and uses all layers, resulting in a larger LoRA with a greater potential for interference with other LoRAs, particularly those targeting overlapping layers. If you encounter compatibility issues, let me know. A version trained without attention and identity layers may be a viable alternative, although removing these layers would likely reduce the overall quality.
64
+
65
+ ---
66
+
67
  New ref2va file to download (experimental, no audio trained)
68
  ============================================================
69
 
 
85
 
86
  Archive: [CivArchive Page](https://civarchive.com/models/2856467?modelVersionId=3266628)
87
 
88
+ <!-- Version: 20261020_upload -->