Hikari Noob v-pred版本0.6.5 (ID: 1078719)

NOTE: This model was created for my own use. There is no guarantee of quality.

Hugging Face Repo(1.0.1): https://huggingface.co/RedRayz/hikari_noob_1.0.1

Fine-tuned NoobAI-XL(v-prediction) without merging another base models, and merged SPO(Step-aware Preference Optimization) LoRA

他所のベースモデルを混ぜずに調整し、SPO LoRAをマージしたNoobAI-XL(v-prediction)

NoobAI-XL(v-prediction版)に特定画風を学習させたcheckpointを複数作成しマージしたものです,絵柄はややばらつきがあり、「典型的AIイラスト」感を抑えています,

This model does not have the "AI Masterpiece" style.

Features/特徴

• Improved stability and quality.

• Fixed a problem in which the quality of output was significantly degraded when the number of tokens exceeded 76.

• The base style is not strong and can be restyled by prompts or LoRAs.

• This model does not include any base model other than NoobAI (v-prediction), so it has the equivalent knowledge. You can generate characters that have appeared by August 2024.

• True v-prediction model with ZTSNR, No epsilon models merged

• LoRAs for NoobAI-XL(v-prediction) work fine.

• 安定性と品質を改善

• トークン数が76を超えると出力の品質が著しく低下する問題を修正しました,

• 素の画風は強くないので、プロンプトやLoRAによる画風変更ができます,

• このモデルはNoobAI(v-prediction)以外のベースモデルを一切含まず、それと同等の知識があります,2024年8月までに登場したキャラクターを生成できます,

• NoobAI-XL(v-prediction)用のLoRAがしっかり効きます,

• 真のv-prediction + ZTSNRなモデルです,Epsilonモデルは一切マージしていません,

Requirements/動作要件

• AUTOMATIC1111 WebUI on dev branch /devブランチ上のAUTOMATIC1111 WebUI

• Latest version of ComfyUI /最新版のComfyUI

• Latest version of Forge or reForge /最新版のForgeまたはreForge

Instruction for AUTOMATIC1111 /AUTOMATIC1111の手順

1. Switch branch to dev(Run this command in the root directory of the webui: git checkout -b dev origin/dev or use Github Desktop)

2. Use the model as usual!

(日本語)

1. devブランチに切り替えます(次のコマンドをwebui直下で実行します: git checkout -b dev origin/dev またはGithub Desktopを使う)

2. 通常通りモデルを使用します,

Recommended Prompt/推奨プロンプト

Positive: None(works good without "masterpiece, best quality")

Negative(simple): "lowres"

Negative(long): "worst quality, bad quality, lowres, photoshop \(medium\), abstract"

Anime Style

Positive: "anime screenshot, anime screencap, anime coloring"

Negative: "worst quality, oldest"

Blue Archive Style

Positive: "official art, blue archive, game cg, non-web source, newest"

Negative: "worst quality, bad quality, low quality, lowres, photoshop \(medium\), abstract, oldest"

Recommended settings/推奨設定

Steps: 12-24

Scheduler: Simple or SGM Uniform

Guidance Scale: 2-5(best value is 4)

Note: "Align Your Steps" scheduler produces a very different image from other samplers.

Recommended Samplers/推奨サンプラー

• DPM++ 2M

• DPM++ 2M SDE

• Euler

• Euler a(euler_ancestral)

• DPM++ 3M SDE

tips: SDE samplers will reduce deformation.

Incompatible Samplers/非対応のサンプラー

• Restart(AUTOMATIC1111/Forge)

• UniPC(very unstable, low quality)

• LMS(low quality)

• PLMS(low quality)

• Heun(broken)

• LCM(NaN error)

other samplers are not tested.

Incompatible Schedulers/非対応のスケジューラー

• Uniform(makes artifacts like glaze or nightshade)

• Karras(unstable)

• Beta(makes blurry images)

Recommended settings for Hires.fix/Hires.fixの推奨設定

Upscaler: 4x-UltraSharp(or ESRGAN-based model), Latent(nearest-exact)

Denoising strength: 0.4-0.5(0.65-0.7 for latent)

Recommended settings for DMD2 LoRA

LoRA strength: 0.4

Steps: 6-10

Sampler: Euler/Euler a

Scheduler: Simple

Guidance Scale: 1-1.5

Note: DMD2 will change the style of the model.

Training information

Please visit the Hugging Face repository

Training script

sd-scripts

Notice/おしらせ:

This model is licensed under Fair AI Public License 1.0-SD

If you make modify this model, you must share both your changes and the original license.

You are prohibited from monetizing any close-sourced fine-tuned /merged model, which disallows the public from accessing the model's source code /weights and its usages.

このモデルはFair AI Public License 1.0-SDでライセンスされています,

このモデルを改変する場合、変更点および元のライセンスを明記する必要があります,

モデルのソースコード/ウェイトや使用法などを閲覧できないクローズソースのファインチューン/マージモデルの収益化は禁止されています,

描述:

Based on NoobAI-XL 0.6(v-predicton)

Enabled Zero Terminal SNR(ZTSNR)

训练词语: