详情
推薦
v3.0
Illustrious-XL-v3.0
原創

Illustrious-XL-v3.0

264.3K
2.0K
0
#Illustrious

Overview

We proudly introduce Illustrious XL 3.0, marking a groundbreaking leap with our first stabilized model supporting up to 2048x2048 native generation resolution—a milestone unprecedented in the Illustrious series. The generation resolution flexibility now spans from 256x256 up to 2048x2048, although we recognize that resolutions below 256x256 generally have limited practical value.

Knowledge Cutoff

The dataset for Illustrious XL 3.0 has been meticulously updated to include content through October 2024, reflecting the latest developments and ensuring the model remains contextually relevant.

Key Features

High-Resolution Stability and Versatility

1. Native generation support up to 2048x2048 resolution, offering exceptional detail and clarity.

2. We recommend users test generation at lower resolutions (e.g., 384x384) and then upscale using strength settings (e.g., upscale by a factor of 2.0 at 0.95 strength), demonstrating uniquely interesting results.

Optimized Training Flexibility

1. Training resolution flexibility now allows arbitrary resolutions, with significantly stabilized training losses, provided image sides remain above 256 pixels—a near-strict limit for the SD XL architecture.

2. We acknowledge the computational intensity of generating 2048x2048 images, with typical generation times around 15 seconds per image on RTX4090-equipped devices, and considerably longer (minutes to hours) on less optimized setups. Meanwhile, generation from 384x384 to 768x768 typically completes within seconds.

New Controlling Tokens

1. We introduce a comprehensive suite of new controlling tokens that empower users to precisely manage visual attributes such as:

2. Contrast: 'low contrast', 'medium contrast', 'high contrast', 'very high contrast'

3. Brightness: 'dark', 'normal brightness', 'bright', 'very bright'

4. Sharpness: 'blurry', 'slightly sharp', 'sharp', 'very sharp'

5. Dynamic Colors: 'static colors', 'medium dynamic colors', 'high dynamic colors', 'very dynamic colors'

6. Colorfulness: 'monotonic color', 'medium colorfulness', 'high colorfulness', 'very high colorfulness'

7. Saturation: 'muted colors', 'average colors', 'vibrant colors', 'very vibrant colors'

查看译文

評分與評論

4.7 /5
0 個評分

尚未收到足夠的評分或評論

avatar
OnomaAI
899
4.2K
公告
2025-06-16
发布模型
2025-06-16
更新模型資訊
模型详情
類型
Checkpoint
发布時間
2025-06-16
基础模型
Illustrious
模型參數
sampler:Euler a
negative_prompt:worst quality,low quality, lowers, low details, bad quality, poorly drawn, bad anatomy, multiple views, bad hands, blurry, artist sign, weibo username
训练参數
Epochs:1
Steps:0
许可范围
創作许可范围
在線生圖
進行融合
允许下载
商業许可范围
生成圖片可出售或用於商業目的
允许模型轉售或融合后出售
QR Code
下載SeaArt App
在移動端继續你的AI創作之旅