README.md

---
license: other
license_name: faipl-1.0-sd
license_link: https://freedevproject.org/faipl-1.0-sd/
base_model:
- Laxhar/sdxl_noob
language:
- en
tags:
- stable-diffusion
- sdxl
---
# Hikari Noob v-pred 0.5

![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/630e2d981ef92d4e37a1694e/b9tyKyu2MwbQTQpuqAg2c.jpeg)

Civitai model page: https://civitai.com/models/938672

Fine-tuned NoobAI-XL(v-prediction) and merged SPO LoRA

NoobAI-XL(v-prediction)をファインチューンし、SPOをマージしました。

## Features/特徴
- Improved stability and quality.
- Works with samplers other than Euler.
- Good results with only 10 steps (12 steps or more recommended)
- 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.
- 安定性と品質を改善
- わずか10ステップでよい結果を得られます(ただし12ステップ以上を推奨)
- Zero Terminal SNRの代わりにNoise Offsetを使用することでEuler以外のサンプラーでも利用できるようにしました。
- 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。

## 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の導入手順
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。

### Example Workflow for ComfyUI / ComfyUIサンプルワークフロー
Download it from [here](https://files.catbox.moe/83e2wl.json)

## Prompt Guidelines / プロンプト記法
Almost same as the base model/ベースモデルとおおむね同じ

To improve the quality of background, add `simple background, transparent background` to Negative Prompt.

## Recommended Prompt / 推奨プロンプト
Positive: None/無し(Works good without `masterpiece, best quality` / `masterpiece, best quality`無しでおｋ)

Negative: `worst quality, low quality, bad quality, lowres, jpeg artifacts, unfinished, photoshop \(medium\), abstract` or empty(または無し)


## Recommended Settings / 推奨設定
Steps: 10-24

Sampler: DPM++ 2M(dpmpp_2m)

Scheduler: Simple

Guidance Scale: 3.5-7

### Hires.fix

Hires upscaler: 4x-UltraSharp or Latent(nearest-exact)

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


## Merge recipe(Weighted sum)
I made 6 Illustrious-based models and merged them.

- Stage 0: finetunes v-pred test model with AI-generated images
- Stage 1: finetunes stage 0 model with 300 scenery images from Gelbooru

- Stage 2: Finetune and merge(see below)

*A-F,sd15: finetuned stage1(ReLoRA)

- A * 0.6 + B * 0.4 = tmp1
- tmp1 * 0.6 + C * 0.4 = tmp2
- tmp2 * 0.7 + F * 0.3 = tmp3
- tmp3 * 0.7 + E * 0.3 = tmp4
- tmp4 * 0.5 + D * 0.5 = tmp5
- tmp5 * 0.65 + sd15 * 0.35 = tmp6
- tmp6 + SPO LoRA = Result

## Training scripts:
[sd-scripts](https://github.com/kohya-ss/sd-scripts)

## Notice
This model is licensed under [Fair AI Public License 1.0-SD](https://freedevproject.org/faipl-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.