Sentinel7 commited on
Commit
d6470ff
·
verified ·
1 Parent(s): d6349bc

Upload 2695779/3032800/README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. 2695779/3032800/README.md +20 -9
2695779/3032800/README.md CHANGED
@@ -36,32 +36,43 @@ tags:
36
 
37
  A double handjob blowbang.
38
 
39
- Include the keyword **bl0wb4ng** in your prompt and wan will handle the rest.
 
 
 
 
 
 
 
 
 
40
 
41
  Usage
42
  -----
43
 
44
- * **Prompting**: keep it simple.
45
  * **Shift**: keep it low. **7** is OK with most of schedulers and evenly split steps.
46
  * **Cut**: the cut happens at the 9th frame.
47
  * Lower **Strengths** (high-noise):
48
 
49
  + ~0.5: makes a progressive transition instead of a cut.
50
- + ~0.7: 2 men instead of multiple.
51
 
52
  Misc
53
  ----
54
 
55
- All the showcase videos have their **workflow** embedded.
56
-
57
- Tested only with **lightx2v** LoRAs enabled (1030 on high-noise and 1022 on low-noise).
58
 
59
- Tested only in **ComfyUI**.
 
 
60
 
61
  Training
62
  --------
63
 
64
- 67.5 epochs, 72 5-seconds clips (4860 steps).
 
65
 
66
  ---
67
 
@@ -71,4 +82,4 @@ Model: [CivitAI Model Page](https://civitai.com/models/2695779?modelVersionId=30
71
 
72
  Archive: [CivArchive Page](https://civarchive.com/models/2695779?modelVersionId=3032800)
73
 
74
- <!-- Version: 20260615_upload -->
 
36
 
37
  A double handjob blowbang.
38
 
39
+ Include the keyword **bl0wb4ng** in your prompt and wan will *handle* the rest.
40
+
41
+ v1.1
42
+ ====
43
+
44
+ Tweaked a few knobs and pushed the training a bit further.
45
+
46
+ 🟢 Less hallucinations
47
+ 🟢 Better hand skills
48
+ 🔴 Slightly reduced facial accuracy
49
 
50
  Usage
51
  -----
52
 
53
+ * **Prompting**: keep it simple; avoid NSFW vocabulary when possible.
54
  * **Shift**: keep it low. **7** is OK with most of schedulers and evenly split steps.
55
  * **Cut**: the cut happens at the 9th frame.
56
  * Lower **Strengths** (high-noise):
57
 
58
  + ~0.5: makes a progressive transition instead of a cut.
59
+ + ~0.7: less crowded scene.
60
 
61
  Misc
62
  ----
63
 
64
+ All the showcase videos have their **workflow** embedded.
65
+ Tested only:
 
66
 
67
+ * with **lightx2v** LoRAs enabled (1030 or 1022 on high-noise and 1022 on low-noise).
68
+ * in **ComfyUI**.
69
+ * on the following **base wan** models: Q4, Q5, Q6 and nvfp4.
70
 
71
  Training
72
  --------
73
 
74
+ * v1.0: 67.5 epochs, 72 5-seconds clips (4860 steps).
75
+ * v1.1: 80.0 epochs, 72 5-seconds clips (5760 steps).
76
 
77
  ---
78
 
 
82
 
83
  Archive: [CivArchive Page](https://civarchive.com/models/2695779?modelVersionId=3032800)
84
 
85
+ <!-- Version: 20260617_upload -->