Ai-tensa commited on
Commit
5a5294d
1 Parent(s): ecfd338

upload models

Browse files
FlexWaifu1.3.1.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b8c719863591c5441e93a7d8f5263b81d1deaaa289713df953bd42d751be0a4
3
+ size 4265145965
LoRA/IR_1girl1boy_1.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2bb59f1af5145f8f96a969966821afbd5a670c86dbbc592b4834528c27aee01e
3
+ size 604095017
LoRA/IR_1girl1boy_2.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a2bfd63a4a58fef4b42ffeac6e7a0a993b8f6b0ec2fdee82f3633a032a9cf9e
3
+ size 604095017
LoRA/IR_1girl1boy_3.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:84fb199ed4a3c461b97c9e7194cddce0371a34c83566f23035c46420e876b725
3
+ size 604095017
README.md CHANGED
@@ -15,26 +15,47 @@ FlexWaifu is a fine-tuned model from Waifu Diffusion 1.3 for producing high reso
15
 
16
  ## Model Description
17
 
18
- This model was created by merging two original LoRAs of [testLoRAs](https://huggingface.co/Ai-tensa/testLoRAs) into WD1.3.
 
 
 
 
 
 
19
 
20
- | Model Name | Recipe |
21
- | ---------- | ------------------------------------------ |
22
- | FlexWaifu | WD1.3 + 2.0 * hires_test_a + smooth_test_a |
23
 
24
  It is just a merged model.
25
  While this model is likely to produce good generation at medium resolution, consider using LoRAs of [testLoRAs](https://huggingface.co/Ai-tensa/testLoRAs) if it does not produce well.
26
 
27
- ## Flex Waifu Rainbow
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
 
29
  This model is further fine-tuned from FlexWaifu with ~17k nijijourneyv5 tagged images of various authors published on the Internet.
30
  It is merged from six dim 8 LoRAs made in various settings, and FWRLoRA is the merged LoRA (dim 48).
31
  Most LoRAs were fine-tuned with Aspect Ratio Backetting with a maximum resolution of 1152x768 images, but some are up to 768x768 or 512x768.
32
  Image captions are made by BLIP and ~12k images also used WD1.4-tagger.
33
 
34
- | Model Name | Recipe |
35
- | ---------------- | ------------------- |
36
- | FlexWaifuRainbow | FlexWaifu + FWRLoRA |
37
-
38
  ### Usage
39
 
40
  The format of the caption suggests that a short natural language sentence followed by a comma-separated tags is the most natural way to describe the prompt.
@@ -42,6 +63,8 @@ Using more tags that are well-estimated by the tagger in the trained images may
42
  "tags.json" lists the tags estimated for over 200 of the 12k images.
43
  Tag semantics may be inappropriate for automatic tagging, so please emphasize appropriately.
44
 
 
 
45
  ## License
46
 
47
  This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
@@ -56,7 +79,28 @@ The CreativeML OpenRAIL License specifies:
56
 
57
  These Models build on the two excellent works: SD1.4, developed by [CompVis Researchers](https://ommer-lab.com/), and WD1.3, developed by [Anthony Mercurio](https://github.com/harubaru), [Salt](https://github.com/sALTaccount/), and [Cafe](https://twitter.com/cafeai_labs).
58
 
59
- ## Examples (Flex Waifu Rainbow)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
 
61
  **Prompt 1**
62
 
 
15
 
16
  ## Model Description
17
 
18
+ | Model Name | Recipe |
19
+ | ---------------- | --------------------------------------------------------------------------------- |
20
+ | FlexWaifu | WD1.3 + 2.0 * hires_test_a + smooth_test_a |
21
+ | FlexWaifu v1.3.1 | FlexWaifu + 20.0 * IR_1girl1boy_1 + 16.0 * IR_1girl1boy_2 + 16.0 * IR_1girl1boy_3 |
22
+ | FlexWaifuRainbow | FlexWaifu + FWRLoRA |
23
+
24
+ ### FlexWaifu
25
 
26
+ This model was created by merging two original LoRAs of [testLoRAs](https://huggingface.co/Ai-tensa/testLoRAs) into WD1.3.
 
 
27
 
28
  It is just a merged model.
29
  While this model is likely to produce good generation at medium resolution, consider using LoRAs of [testLoRAs](https://huggingface.co/Ai-tensa/testLoRAs) if it does not produce well.
30
 
31
+ #### v1.3.1
32
+
33
+ The model is fine-tuned with self-generated images with a single word prompt "1girl" or "1boy" and generates well without much prompting.
34
+
35
+ Twin LoRA has reduced the percentage of bad output without changing the style much.
36
+ The images for the three Twin LoRAs were generated by FlexWaifu or a model merging Twin LoRA into it.
37
+ 3-5k images are used per Twin LORA, no duplicates.
38
+ The reward value of ImageReward was used to classify the images.
39
+
40
+ ##### Twin LoRA
41
+
42
+ When fine-tuning, we create two LoRAs, a good LoRA and a bad LoRA, and take the difference between them in order to suppress adverse effects and achieve the desired effect.
43
+
44
+ **method**
45
+
46
+ 1. Create a set of images with the same prompt.
47
+ 2. Divide the image set into two equal parts, good and bad, according to certain evaluation criteria.
48
+ 3. Create a LoRA for each data set.
49
+ 4. Subtract the bad LoRA from the good LoRA. (Note: the dimension is twice the original LoRA)
50
+ 5. Apply differential LoRA in any weight.
51
+
52
+ ### Flex Waifu Rainbow
53
 
54
  This model is further fine-tuned from FlexWaifu with ~17k nijijourneyv5 tagged images of various authors published on the Internet.
55
  It is merged from six dim 8 LoRAs made in various settings, and FWRLoRA is the merged LoRA (dim 48).
56
  Most LoRAs were fine-tuned with Aspect Ratio Backetting with a maximum resolution of 1152x768 images, but some are up to 768x768 or 512x768.
57
  Image captions are made by BLIP and ~12k images also used WD1.4-tagger.
58
 
 
 
 
 
59
  ### Usage
60
 
61
  The format of the caption suggests that a short natural language sentence followed by a comma-separated tags is the most natural way to describe the prompt.
 
63
  "tags.json" lists the tags estimated for over 200 of the 12k images.
64
  Tag semantics may be inappropriate for automatic tagging, so please emphasize appropriately.
65
 
66
+ CLIP Skip 1 is recommended.
67
+
68
  ## License
69
 
70
  This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
 
79
 
80
  These Models build on the two excellent works: SD1.4, developed by [CompVis Researchers](https://ommer-lab.com/), and WD1.3, developed by [Anthony Mercurio](https://github.com/harubaru), [Salt](https://github.com/sALTaccount/), and [Cafe](https://twitter.com/cafeai_labs).
81
 
82
+ ## Examples
83
+
84
+ **CLIP Skip 1 is recommended.**
85
+
86
+ ### Flex Waifu v1.3.1
87
+
88
+ **Prompt 1**
89
+
90
+ - with Negative Prompt
91
+
92
+ ![](images/grid-0050-1443377636.png)
93
+
94
+ - without Negative Prompt
95
+
96
+ ![](images/grid-0051-3282638012.png)
97
+
98
+ ```
99
+ solo, 1girl, white_background, full_body, twintails, braid, white_background, bangs, frills, closed_mouth, brown_hair, jewelry, blush, standing, dress, food, strawberry, (blueberry:1.1), (cake:1.1), sweets, brown_eyes, hair_ornament, skirt, bow
100
+ Steps: 30, Sampler: UniPC, CFG scale: 7, Seed: 1443377636, Size: 768x768, Model hash: 7b8c719863, Model: FlexWaifu_FlexWaifu1.3.1, Denoising strength: 0.6, Version: v1.2.1, Hires upscale: 1.5, Hires steps: 18, Hires upscaler: Latent
101
+ ```
102
+
103
+ ### Flex Waifu Rainbow
104
 
105
  **Prompt 1**
106
 
images/grid-0050-1443377636.png ADDED

Git LFS Details

  • SHA256: 381536cb97bdf1ba8a16fd09477bb1cedf83671ebd749ab6f71fc3da5f67ee62
  • Pointer size: 133 Bytes
  • Size of remote file: 34.8 MB
images/grid-0051-3282638012.png ADDED

Git LFS Details

  • SHA256: afa51a31ffcd87bdaf55cf30306c38e283f55422340804203ba8b5dea9dc39f5
  • Pointer size: 133 Bytes
  • Size of remote file: 33.5 MB