Please, check About this version. This version has a more subdued color palette than v2.0, closer to that of the anime screenshot.
The reason we insist on eps prediction instead of vpaerd is that a small change in cgf value can result in a very different finished product. But we will need to move to Vpaerd at some point.
(12/15/2024). coco-Illustrious-NoobXL-Style-v3
If you like my work, please write a review and press the heart icon. All of them will be my motivation.
1. How to use
1) For reforge, turn off Advanced Model Sampling for reForge.
2) CLIP is 2
3) samplers :eular A / SGM Uniform
4) CFG scale2-7. step20-40
5) I did not use various features such as upscaler, face correction, ADetailer, etc. so the detailed settings are your choice.
6) No VAE required.
7)Have fun!!
(12/1/2024). NoobAI-XL vpred0.65s+vpred0.6marge/日本語説明は下に↓
I have repeatedly merged vpred 0.65s and vpred 0.6 into my model and the result is the best model I have ever had. The model itself is epsilon, but I think the result is close to it, because I don't like the illustration output by NoobAI-XL's vpaerd (personal value), and I think the difference between vpard and epsilon is just a difference in prediction, and personal preference is different.
I have improved my understanding of prompts and can now express a wide range of dark and light areas, SFW, NSFW, and even maniacal expressions. Posing has also been improved. This has increased the number of expressions that are possible without using LoRA. I think that backgrounds can also be expressed in detail. Of course, it works well with LoRA. Please use your favorite LoRA in combination with the new version and submit it to us!
Also, I am not a very good advertiser, so if you download this model and like it, please spread the word to many people.
If you like my work, please write a review and press the heart icon. All of them will be my motivation.
1. How to use
1) For reforge, turn off Advanced Model Sampling for reForge.
2) CLIP is 2
3) samplers :eular A / SGM Uniform
4) CFG scale2-5. step20-40
5) I did not use various features such as upscaler, face correction, ADetailer, etc. so the detailed settings are your choice.
6) No VAE required.
7)Have fun!!
prompts
positive prompts: (masterpiece, best quality), amazing quality, very aesthetic, absurdres, highres, newest, HDR, 8K, high detail RAW color art, intricate details, high contrast, photo-referenced, detailed background,
Negative prompts
lowres, worst quality, bad quality, bad anatomy, jpeg artifacts, old, oldest, watermark, anime screenshot, anime coloring, nostrils, nose,
このモデルは、vpred0.65sとvpred0.6を私のモデルに繰り返しマージを行いました。その結果今までで最高のモデルが出来たと感じてます。私がNooｂAI-XLのvpaerdで出力されたイラストがあまり好きではないので、モデル自体はepsilonですがそれに近い仕上がりになってると思います(個人の価値観)またVpard もepsilonも予測の違いなだけで好みは個人で別れると思います。Vpaerd予測の方が優れているとも聞きますが個人的はepsilonが好き。
プロンプトの理解力が向上し暗い場所や明る場所、SFWやNSFW、マニアックな表現まで幅広く表現で来ました。更にポージングも改善されました。これによって、LoRAを使わなくても可能な表現が増えました。背景も詳細に表現出来ると思います。勿論Loraとの相性もいいです。是非お気に入りのLoRAと組み合わせて投稿してください！
あと私はあまり宣伝が上手ではないので、もしこのモデルをダウンロードして気に入ったのなら色んな人にこのモデルを広めてほしいです。
(11/19/2024). v1.5-vpred0.6 marge/日本語説明はは下に
Model Concept
This model is a merge of cco-Illustrious-NoobXL-Style and NoobAI-XL (v-pred-0.6-version), so it is not completely v-pred. It will be practically eps.
It may not be a perfect reproduction, but it is probably an accurate representation of the anatomy and a powerful representation of the high quality of the image. The same goes for artist and character tags.I merged the v-prediction model because it seemed unwieldy to me. The number of merging was 5 times, subtraction and addition.
I think it is smoother and less noisy than v-prediction. However, personal opinion.
How to use
No LoRA or artist tag or extensions are used in the preview image.
If you are using reforge, turn off Advanced Model Sampling for reForge. You will get terrible noise. Other than that, it's basically the same; I don't know about ComfyUI and AUTOMATIC1111, as I haven't used them.
Recommendation
Quality Prompts
(masterpiece, best quality), amazing quality, very aesthetic, absurdres, highres, newest, HDR, 8K, high detail RAW color art, intricate details,
Negative prompts
lowres, worst quality, bad quality, bad anatomy, jpeg artifacts, old, oldest, watermark,
sampler scheduler settings.
Euler a or Euler Ancestral CFG++/ SGM Uniform or BETA
CFG scale3-5. step20-50
All models used in the merge are copyrighted and licensed to the creator.
(11/20/2024).
Corrected incorrect notations and added some information
このモデルはcco-Illustrious-NoobXL-StyleとNoobAI-XL（v-pred-0.6-version）をマージしたものなので、完全なv-predではありません。実質的にはepsになります。
完全な再現ではないかもしれませんが、おそらく解剖学的な表現は正確であり、画像のクオリティの高さを力強く表現していると思います。アーティストやキャラクターのタグも同様です。
プレビュー画像はLoRA、アーティストタグ、拡張機能は使用していません。
使用方法
forgeを使用している場合は、Advanced Model Samplingをオフにしてください。ひどいノイズが出ます。それ以外は基本的に同じです。ComfyUIとAUTOMATIC1111は使っていないので分かりません。
製作者の方に深く感謝です。
(11/17/2024). v1.0
The concept aimed for high quality without LoRA, artist tags or character names.
This model is based on Illustrious. I merged several models based on NoobAI-XL, Obsession and Illustrious with my unpublished base models to get the best fit. Credit for all models used in the merging belongs to their creators. Thanks to all the creators.
Please refer to the preview of images created without ADetailer or upscaling.
Recommended prompts
(masterpiece, best quality), amazing quality, very aesthetic, absurdres, highres, newest, HDR, 8K, high detail RAW color art, intricate details, high contrast, detailed background, beautiful detailed, beautiful face, blight face, large pupil, eyelashes, brilliant skin, shiny skin, perfect hand,
Negative prompt
(lowres, bad quality:1.2), bad hands, missing finger, watermark, username, logo,
Incorrect use of recommended settings may lead to unexpected results.
Try adding photo background, photo-referenced, to the prompt to change the flavour. Adding anime screenshot, anime coloring, to the negative prompt could be a new discovery.
Recommendation setting
Euler a or DPM++ 2M Karras, Step 30-50, cfg 5-7, 768x1344, 864x1536
Recommendations are basic settings or prompts. Other tags can be added/removed to find the best.
Hope you will have fun and enjoy
LoRA、アーティストタグ、キャラクター名を使わず、高品質を目指したコンセプト。
追記：公開したプレビュー画像に未公開のモデルで作成した物が数点混じっており後ほど一部差し替えます。
このモデルは Illustrious に基づいています。NoobAI-XL、Obsession、Illustrious に基づくいくつかのモデルを、未公開のベース モデルと統合して、最適な適合を実現しました。
マージに使用されたすべてのモデルの著作権は作成者に帰属します。
すべてのクリエイターに感謝します。
ADetailerやアップスケーリングなしで作成された画像をプレビューしていますので参考にしてみてください。
プロンプトや推奨設定は上記確認してください。プロンプトに photo background, photo-referencedを追加したり、ネガティブプロンプトにanime screenshot, anime coloring,を追加するのも新しい発見になるかもしれません。雰囲気を変えてみるのも楽しいですよ。
推奨は基本的な設定またはプロンプトです。他のタグを追加/削除できますのでベストを探してみてください。
それでは楽しんでくださいね！
