prithivMLmods commited on
Commit
4868ae7
·
verified ·
1 Parent(s): 82d7ddb

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +64 -0
README.md CHANGED
@@ -31,6 +31,70 @@ Photo-Restore-i2i is an adapter for black-forest-lab's FLUX.1-Kontext-dev, desig
31
  > [!note]
32
  [photo content], restore and enhance the image by repairing any damage, scratches, or fading. Colorize the photo naturally while preserving authentic textures and details, maintaining a realistic and historically accurate look.
33
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
34
  ## Trigger words
35
 
36
  You should use `[photo content]` to trigger the image generation.
 
31
  > [!note]
32
  [photo content], restore and enhance the image by repairing any damage, scratches, or fading. Colorize the photo naturally while preserving authentic textures and details, maintaining a realistic and historically accurate look.
33
 
34
+ ---
35
+
36
+ ## Sample Inference
37
+
38
+ | ex1 | ex2 |
39
+ |------|-------|
40
+ | ![Left Screenshot](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/g--KoNLNm45CzOByLkokN.png) | ![Right Screenshot](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/zBL4ZAapuRgrfQCu0-zga.png) |
41
+
42
+ ---
43
+
44
+ ## Parameter Settings
45
+
46
+ | Setting | Value |
47
+ | ------------------------ | ------------------------ |
48
+ | Module Type | Adapter |
49
+ | Base Model | FLUX.1 Kontext Dev - fp8 |
50
+ | Trigger Words | [photo content], restore and enhance the image by repairing any damage, scratches, or fading. Colorize the photo naturally while preserving authentic textures and details, maintaining a realistic and historically accurate look. |
51
+ | Image Processing Repeats | 50 |
52
+ | Epochs | 28 |
53
+ | Save Every N Epochs | 1 |
54
+
55
+ Labeling: DeepCaption-VLA-7B(natural language & English)
56
+
57
+ Total Images Used for Training : 50 Image Pairs (25 Start, 25 End)
58
+
59
+ Synthetic Result Node generated by NanoBanana from Google (Image Result Sets Dataset)
60
+
61
+
62
+ ## Training Parameters
63
+
64
+ | Setting | Value |
65
+ | --------------------------- | --------- |
66
+ | Seed | - |
67
+ | Clip Skip | - |
68
+ | Text Encoder LR | 0.00001 |
69
+ | UNet LR | 0.00005 |
70
+ | LR Scheduler | constant |
71
+ | Optimizer | AdamW8bit |
72
+ | Network Dimension | 64 |
73
+ | Network Alpha | 32 |
74
+ | Gradient Accumulation Steps | - |
75
+
76
+ ## Label Parameters
77
+
78
+ | Setting | Value |
79
+ | --------------- | ----- |
80
+ | Shuffle Caption | - |
81
+ | Keep N Tokens | - |
82
+
83
+ ## Advanced Parameters
84
+
85
+ | Setting | Value |
86
+ | ------------------------- | ----- |
87
+ | Noise Offset | 0.03 |
88
+ | Multires Noise Discount | 0.1 |
89
+ | Multires Noise Iterations | 10 |
90
+ | Conv Dimension | - |
91
+ | Conv Alpha | - |
92
+ | Batch Size | - |
93
+ | Steps | 4100 |
94
+ | Sampler | euler |
95
+
96
+ ---
97
+
98
  ## Trigger words
99
 
100
  You should use `[photo content]` to trigger the image generation.