Update README.md
Browse files
README.md
CHANGED
@@ -85,7 +85,7 @@ rev01A、rev010A、TEST-rev020A、それぞれ非公開とします(ComfyUI 等
|
|
85 |
<br>
|
86 |
rev010A を修正し、新たに、rev011A、rev011B-LCM、として公開します<br>
|
87 |
<br>
|
88 |
-
|
89 |
Correction of "Unet corruption" found in previous model data.<br>
|
90 |
rev01A, rev010A, and TEST-rev020A are now private (could not be generated by ComfyUI, etc.).<br>
|
91 |
※I will inform you that the above 3 models are "merge ineligible".<br>
|
@@ -94,7 +94,7 @@ rev01A Revised rev010A is modified, rev011A and rev011B-LCM are newly released.<
|
|
94 |
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/7wI8xJM4izX3R7dZCNS9c.jpeg)
|
95 |
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/yQT90GSxTeKXyc8LuMHYW.jpeg)
|
96 |
<br>
|
97 |
-
[Easy settings] //rev011A// STEP:16~ CFG:5 //rev011B-LCM// STEP:12~ CFG:5
|
98 |
<br>
|
99 |
[おすすめ] 私的 LCM 活用<br>
|
100 |
LCM(LORA) をモデルにマージする場合 bM1280ch (FreeU 相当) にすると STEP 設定に余裕が生まれました<br>
|
@@ -103,4 +103,37 @@ LCM(LORA) をモデルにマージする場合 bM1280ch (FreeU 相当) にする
|
|
103 |
When merging LCM(LORA) into a model, set bM1280ch (equivalent to FreeU) to give more leeway for STEP setting.<br>
|
104 |
※bM1280ch = 0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0<br>
|
105 |
<br>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
106 |
<br>
|
|
|
85 |
<br>
|
86 |
rev010A を修正し、新たに、rev011A、rev011B-LCM、として公開します<br>
|
87 |
<br>
|
88 |
+
|Fixes| Merge Model Corruption and Fixes<br>
|
89 |
Correction of "Unet corruption" found in previous model data.<br>
|
90 |
rev01A, rev010A, and TEST-rev020A are now private (could not be generated by ComfyUI, etc.).<br>
|
91 |
※I will inform you that the above 3 models are "merge ineligible".<br>
|
|
|
94 |
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/7wI8xJM4izX3R7dZCNS9c.jpeg)
|
95 |
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/yQT90GSxTeKXyc8LuMHYW.jpeg)
|
96 |
<br>
|
97 |
+
~[Easy settings] //rev011A// STEP:16~ CFG:5 //rev011B-LCM// STEP:12~ CFG:5 あたりがオススメです~<br>
|
98 |
<br>
|
99 |
[おすすめ] 私的 LCM 活用<br>
|
100 |
LCM(LORA) をモデルにマージする場合 bM1280ch (FreeU 相当) にすると STEP 設定に余裕が生まれました<br>
|
|
|
103 |
When merging LCM(LORA) into a model, set bM1280ch (equivalent to FreeU) to give more leeway for STEP setting.<br>
|
104 |
※bM1280ch = 0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0<br>
|
105 |
<br>
|
106 |
+
<br>
|
107 |
+
|LCM について (追記)| >LCM の効果的な使用方法について<<br>
|
108 |
+
モデル 「ASAP - As_soon_eastAsian_PEEPS - rev011C (53ae6f7c53) LCM1-5+[05R1280G].fp16.safetensors [4be77e33f2]」<br>
|
109 |
+
上のモデルにて STEP、Sampler、CFG、を比較してみると良好な設定が見えてきました(上記モデルも共有します)<br>
|
110 |
+
<br>
|
111 |
+
|About LCM (addition)| > How to use LCM effectively < <br>
|
112 |
+
Model"ASAP - As_soon_eastAsian_PEEPS - rev011C (53ae6f7c53) LCM1-5+[05R1280G].fp16.safetensors [4be77e33f2]"<br>
|
113 |
+
Comparing STEP, Sampler, CFG, on the model above reveals a good setup (the model above is also shared)<br>
|
114 |
+
<br>
|
115 |
+
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/cTGDruQHEFhui6jr0RF3Y.jpeg)
|
116 |
+
<br>
|
117 |
+
[Easy settings] //LCM 適用// STEP:4~ CFG:1~3 Sampler:Euler a、DPM++ 2S a Karras、 オススメします
|
118 |
+
<br>
|
119 |
+
Samplerは、Euler a、DPM++ 2S a Karras、この2つを選択するのが最良かなと思います<br>
|
120 |
+
このモデルへの LCM 適用は、以下の層別マージを「0.5」で行っています<br>
|
121 |
+
※R1280Gch:0,1,1,1,1,1,0,1,0,0,0,0,0,1,0,0,0,0,0,1,0,1,1,1,1,1 (実質/bm320、bm640 [FreeU 相当])<br>
|
122 |
+
これだけで STEP 4 くらいでキレイな生成画像を得られるようになります<br>
|
123 |
+
Sampler is Euler a, DPM++ 2S a Karras, I think these two are the best choices!<br>
|
124 |
+
LCM application to this model has the following stratified merges at "0.5"<br>
|
125 |
+
R1280Gch:0,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1 (real/bm320, bm640 [FreeU equivalent])<br>
|
126 |
+
This is all it takes to get a beautiful generated image in about STEP 4!<br>
|
127 |
+
<br>
|
128 |
+
LCM (LORA) は、使用するモデルにより強度を下げて適用するとよいです<br>
|
129 |
+
実際にいくつかのモデルでは強度(重み)を「1」にすると崩れた画像ばかりになります<br>
|
130 |
+
今回の試験では「0.5」を上記のように適用し生成時間を短くできました<br>
|
131 |
+
特に「GeForce 以外の環境の方」は LCM 効果を実感できるはずです<br>
|
132 |
+
LCM (LORA) should be applied with reduced intensity depending on the model used.<br>
|
133 |
+
In fact, for some models, setting the strength (weight) to "1" results in only broken images.<br>
|
134 |
+
In this test, "0.5" was applied as shown above to shorten the generation time.<br>
|
135 |
+
In particular "Non-GeForce users" will be able to feel the LCM effect.<br>
|
136 |
+
<br>
|
137 |
+
あなたのお気に入りのモデルにも LCM を使用してみてください<br>
|
138 |
+
You can also use LCM on your favorite model. Try LCM on your favorite models!<br>
|
139 |
<br>
|