Good to see you! Try the new model!
(6/13/2005)
Announcement
With this version 2.1, coco-style-Illustrious-XL will no longer be updated.
And finally, we will complete all model updates with the release of two types of coco-style-NoobAI-XL - v8.0(v-pred/epsilon).
Thank you for your patronage.
coco-style-Illustrious-XL-v2.1
This version was developed based on version 2.0.
Changes from the previous version include reducing excessive style effects for a more user-friendly finish. The backlight has also been adjusted to achieve more natural dimming.
The biggest improvement is that it is now possible to generate higher resolution and higher detail, enabling stable rendering down to the fingertips. It offers greater stability than cocostyle-Noob-XL-v6.0 in details such as fingertips.
The resolution remains unchanged from v2.0, with standard support for 1536×1536.
Sample Image Description
Many sample images were generated using high-res (1.5x upscale), ADetailer, and CD Tuner. Of course, high-quality illustrations can be created without these tools.
Additionally, the model supports a wide range of LoRA, character tags, and artist tags.
And if you like it, please download it, write a review and press the Like icon. All of this will be my motivation.
LINK. TensorArt. yodayo. SeaArt. pixai, x.
日本語説明はバージョン情報を読んでください
今回のバージョン2.1をもってcoco-style-Illustrious-XLは更新を終えます。
長らくのご愛顧ありがとうございました。
Please refer to the version information below.
(6/1/2005)
coco-style-Illustrious-XL-v2.0
The base model is Illumium XL (Illustrious) v3.1 (e-perd). I looked for a model that would encompass the new dataset.
I merged my unpublished model (created on IllustriousXL-2.0 base) and style LoRA (2D and illustration types) into this model.
Therefore, when generating images with a resolution of 1536x1536, the model is stable
Also, due to the influence of my style LoRA and the LoRA described below, there is more interpretation of the prompts than in the previous v1.0 model, allowing for freer illustrations and improved color representation.
We have achieved a high quality fusion of realism and illustration, one of our efforts since last year.
Please enjoy the various expressions from extraordinary fantasy scenes to everyday scenes.
Some of the sample images use Hires x 1.5 and ADetailer.
If you have not yet added ADetailer to your extension, clicking on ADetailer will lead you to the creator's github and cloning the repository will install it.
My advice in using this model is that you do not need a LoRA with a strong style system of change. It is already moderately merged. Flavored.
Also, since many of my sample images are high resolution, the clarity may be impaired depending on the monitor you are using. This seems to be a problem on the civita side. If this is the case, please try viewing the images on your smartphone or other device.
How to use this model well
Euler a SGM Uniform /Euler Ancestral CFG++Bet//CFG scale4.5-5.5 / step15-30 / 720x1280 /768x1344, 896×1536, 1080×1920 /clip skip2.
prompts
masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest, 2023, 2024, HDR, 8K, high detailed, Additional Prompts
subtle light, high contrast, detailed background, color depth, chiaroscuro, impasto, smooth shading, glossy surfaces, shiny skin,Also recommended
dak depth, low kye, intricate details, masterful shading, detailed background, chromatic aberration, cinematic composition, blurry foreground, blurry background, moody atmosphere, muted colors, etc.Negative prompts
lowres, worst quality, bad quality, bad ???????, jpeg artifacts, bad finger, missing finger, watermark, artistic error, signature, artist name, displeasing, worst aesthetic, old, early, simple eyes, simple shading, simple drawing, simple face, simple coloring, mismatched pupils, logo, The model may have a taste similar to 2.5 as a feature of the model. In that case, please add 3d, realistic, etc. to the anime style or negative prompt.
Description
A corrected version has been released.
I was going to just replace the contents of BATA.v1, but it's been almost 2 days since it was published, so I'll call it BATA. v2. VAE has been baked in.
The urgent issue of noise is now almost non-existent when using character tags, artist tags, and LoRA. (It is not perfect because it may appear by accident and we cannot test all of them.)
Also, when generating in the vanilla environment (default), the noise disappeared when using DPM++ 2M Karras.
Similarly, BATA. v1 and alpha models also tend to have less noise when using DPM++ 2M Karras when generated in a vanilla environment (default).
修正版をリリースしました。
BATA.ｖ1の中身を差し替えるだけにしようかと思ったのですが、公開から2日近く経ってしまったので、BATA. v2とします。ＶＡＥは焼き付けました.。
喫緊の課題だったノイズに関してはキャラクタータグやアーティストタグ、LoRA使用時はほぼ発生しないと思われます。(偶然出てしまうことや、全てを試せないので完全とは言えない)
また、バニラ環境(デフォルト)で生成する際はDPM++ 2M Karrasを使うとノイズはなくなりました。
また同様にBATA. v1やアルファモデルでもバニラ環境(デフォルト)で生成する際はDPM++ 2M Karras を使うとノイズが少なくなる傾向があります。
1. How to use
1) For reforge, turn off Advanced Model Sampling for reForge.
2) CLIP is 2
3) samplers :eular A / SGM Uniform/ DPM++ 2M Karras.
4) CFG scale4-6. step20-40
5) I did not use various features such as upscaler, face correction, ADetailer, etc. so the detailed settings are your choice.
6) VAE is not required but should be used as the situation requires. (use of lunaXLVAE_luna or sdxl_vae is recommended).
7)Have fun!!
Additional information on usage
Euler a / SGM Uniform. CFG scale4-6. step20-40. / 720x1280. /768x1344, 896×1536/ 1080×1920 /clip skip2.
prompts
masterpiece, best quality, amazing quality, very aesthetic, absurdres, highres, newest, 2023, 2024,Additional Prompts
HDR, 8K, high detailed, subtle, high contrast, detailed background, color depth, Also recommended
dak depth, light, low kye, intricate details, masterful shading, detailed background, chromatic aberration, cinematic composition, blurry foreground, blurry background, moody atmosphere, muted colors, etc.Negative prompts
lowres, worst quality, bad quality, bad ???????, jpeg artifacts, watermark, bad finger, missing finger,