mu0n's picture
Update README.md
3745a54
|
raw
history blame
8.22 kB
---
license: other
language:
- ja
library_name: diffusers
pipeline_tag: text-to-image
tags:
- stable-diffusion
- text-to-image
---
<br>
※Translated with www.DeepL.com/Translator<br>
<br>
[license] Modified >>> CreativeML Open RAIL-M<br>
このモデルはマージモデルです(マージの実験モデル)<br>
権利は CreativeML Open RAIL-M に準拠しますが以下の禁止を含みます<br>
※商用利用全般ご遠慮ください ※違法画像生成ご遠慮ください<br>
This model is a merge model (experimental model of merge)<br>
Rights are subject to CreativeML Open RAIL-M but include the following prohibitions<br>
※Please refrain from general commercial use ※Please refrain from creating images<br>
<br>
[NG] Sell images they generate<br>
   画像販売不可<br>
[NG] Run on services that generate images for money<br>
   商用利用不可<br>
[NG] Sell this model or merges using this model<br>
   このモデルの販売不可<br>
[OK] Use the model without crediting the creator<br>
   作者のクレジットなしにこのモデルを使用可<br>
[OK] Share merges using this model<br>
   このモデルを別にマージし再配布可<br>
[OK] Have different permissions when sharing merges<br>
   マージしたものは適宜で権利の選択可<br>
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/UYJw_s0Yy10S6sJhd-GpM.jpeg)
[設定] sampler:DDIM, step:16, clipskip:2(1~3), vae:vae-ft-mse-840000, CFG:5(3~12),<br>
    HiresFix(GeForce users):4x-UltraSharp, Denoising strength:0.3(0~0.7), <br>
    Negative prompt:ExtraArms(LORA), multiarm-4(LORA),<br>
---お願い---<br>
このモデルの使用では以下に関し使用禁止ですご注意ください<br>
1) 暴力的な表現<br>
2) 児童ポルノ<br>
3) 商用利用全般<br>
このモデルは CreativeML Open RAIL-M ライセンスに上記の禁止を追加します<br>
このモデルを使用したことで生じたいかなる問題に対し作者は一切責任を負いません<br>
ご了承の上ご使用ください<br>
Please note that the following are prohibited in the use of this model<br>
1) Violent expressions<br>
2) Child pornography<br>
3) Commercial use in general<br>
This model adds the above prohibitions to the CreativeML Open RAIL-M license.<br>
The author assumes no responsibility for any problems caused by the use of this model.<br>
Please use this model with understanding.<br>
<br>
このモデルは 「El Zipang」 (https://huggingface.co/deadman44) をベースにいくつかの調整で作成しました<br>
マージ使用モデルは、OnlyRealisticV28、MUSE_v1、CandyApple_v1.2、El Zipang、です、各モデルの作者に感謝します<br>
This model is based on "El_Zipang" (https://huggingface.co/deadman44) with some adjustments.<br>
The merged models are OnlyRealisticV28, MUSE_v1, CandyApple_v1.2 and El Zipang, thanks to the authors of each model!<br>
その他(Others) lbwpresets、SuperMerger、Stable Diffusion web UI、、、<br>
Zuntan03 「BIMO method」 (https://github.com/Zuntan03/SdWebUiTutorial/blob/main/_/doc/BimoLoraBlockWeight/BimoLoraBlockWeight.md) <br>
hako-mikan 「SuperMerger」 (https://github.com/hako-mikan/sd-webui-supermerger) <br>
AUTOMATIC1111 「Stable Diffusion web UI」 (https://github.com/AUTOMATIC1111/stable-diffusion-webui) <br>
開発者の方々に感謝します(Thanks to the developers!) <br>
ネガティブプロンプトは少ないほうがキレイな画を出せると思います(Hiresfix 推奨します)<br>
(GeForce 以外の環境の方が、早い、簡単、美しい、モデルを楽しめるように ASAP と名付けました)<br>
(作者も少し前まで AMD の内蔵 GPU で画像生成しモデルマージもしていました)<br>
できるだけ手をきれいに出したいと作成しました作例は xyz プロットをご覧ください<br>
The less negative prompts, the more beautiful the picture will be (Hiresfix recommended).<br>
(I named it ASAP so that non-GeForce users can enjoy fast, easy, and beautiful models.)<br>
(The author also used to generate images and merge models on AMD's internal GPU until a while ago)<br>
I created this model to make the hands as beautiful as possible.<br>
ぜひお試しください (Please try it!) <br>
<br>
<br>
|||TEST 版 sefetensors について|||
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/654314c7c9483f8861a8e296/4ke5Y3D61gY6JlwqY_xxo.jpeg)
さらに早い生成のテストです<br>
[設定] CFG:4, STEP:6, HiresFix-Denoize:0.5(GeForce users) <br>
上の設定で試してみてください(参考プロンプトはファイルの PNG 情報からどうぞ)<br>
Here's a test of even faster generation.<br>
Try the above settings.(For reference prompts, please go to the PNG info in the file)<br>
[説明] このテストは LCM LORA を組み込んだモデルです[LCM(LORA) 0.5]<br>
その他、epi_Noize-(LORA) などをマイナス適用し生成時のノイズを減じてます<br>
※FLAT(LORA) をネガティブに入れると精細化しキレイな画像を得られます<br>
※西洋系モデルを学習元にしたノイズ LORA をマイナス適用すると西洋系の顔立ちを薄められました<br>
※東洋系モデルを学習元にしたノイズ LORA をプラス適用すると東洋系の顔立ちを濃くできました<br>
※FreeU によるチャンネル分けもマージ時の層別適用に効果的と思います<br>
皆さんの「マージ」の参考になれば幸いです<br>
[Description] This test is a model incorporating LCM LORA [LCM(LORA) 0.5].<br>
Other, epi_Noize-(LORA), etc. are applied to reduce noise during generation.[epi_Noize-(LORA) -1.0]<br>
※Negative prompt of FLAT(LORA) results in a more refined and beautiful image.<br>
※Negative noise LORA using Western models as training sources can be applied to reduce Western facial features.<br>
※Noise LORA using an oriental model as a training source can be applied to enhance oriental facial features.<br>
※Channel separation by FREEU is Effective stratified application during merging.<br>
I hope this will be helpful for your "merging"!<br>
<br>
※Translated with www.DeepL.com/Translator<br>
<br>
[LCM 考察] rev020 テストを経て得られたもの<br>
LCM (LORA) 全層適用の高速化は最大 62% になりましたが OUT 層への影響で消えないノイズも残りました<br>
そこで MID 層のみ LCM (LORA) 適用します、高速化は 25% に留まりますがノイズはキレイに消えます<br>
結果的にこれが最適解かなと思います ※LCM 0.5 適用の場合です強度を上げればさらに高速化するはずです<br>
(<<NON LCM STEP:16<<rev 020 LCM ALL STEP:6>>LCM MID Layer STEP:12>>)<br>
[LCM Considerations] What we learned from the rev020 test<br>
LCM (LORA) applied to all layers resulted in a 62% speedup, but some noise remained due to the effect on the OUT layer.<br>
Therefore, LCM (LORA) was applied only to the MID layer. The speedup was 25%, but the noise disappeared cleanly.<br>
As a result, I think this is the best solution.<br>
※This is the case when LCM 0.5 is applied. If the intensity is increased, the speed should be further improved.<br>
(<<NON LCM STEP:16<<rev 020 LCM ALL Layer STEP:6>>LCM MID Layer STEP:12>>)<br>
<br>
[おすすめ] 高効率な LCM 利用法<br>
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す<br>
今回のテスト版では STEP:10 以上の場合に破綻のない画像を高頻度で得られました<br>
※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>
[Recommendation] Efficient LCM usage<br>
When merging LCM(LORA) into a model, set bM1280ch (equivalent to FreeU) to give more leeway for STEP setting.<br>
In this test version, we obtained images with no collapses at a high frequency when STEP:10 or higher was used.<br>
※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>
<br>
<br>