Rouwei 0.8

🚀 RouWei-0.8: Advanced Text-to-Image Model

RouWei-0.8 is a cutting-edge text-to-image model that undergoes in-depth retraining of Illustrious. It offers outstanding prompt adherence, rich knowledge, and state-of-the-art performance, leveraging a well-curated dataset.


🚀 Quick Start

Model Information

Key Features

• Rich Knowledge: It has fresh and extensive knowledge about characters, concepts, styles, and cultural aspects.
• Excellent Prompt Adherence: Achieves the best prompt adherence among SDXL anime models at the time of release.
• Problem Solving: Solves common problems such as tag bleeding and biases.
• Wide Style Support: Offers excellent aesthetics and knowledge across over 50,000 artist styles.
• High Flexibility: Provides high flexibility and variety without sacrificing stability.
• Watermark-Free: Thanks to a clean dataset, there are no annoying watermarks for popular styles.
• Vibrant Colors: Features vibrant colors and smooth gradients.
• Pure Training: Trained purely from Illustrious v0.1 without third - party checkpoints.

The dataset cut-off is the end of April 2025. More detailed description on Civitai

✨ Features and Prompting

Important Change

⚠️ Important Note

When prompting artist styles, especially when mixing several, their tags MUST BE in a separate CLIP chunk. Add BREAK after it (for A1111 and derivatives), use conditioning concat node (for Comfy) or at least put them in the very end. Otherwise, significant degradation of results is likely.

The model can work with both short booru tag-based and long complex natural text prompts. The best results are achieved by combining tags and natural text phrases. Classic danbooru-style comma-separated tags without underscores are used.

Basic Settings

~1..1.5 megapixel for txt2img, any AR with resolution multiple of 64 (1024x1024, 1152x, 1216x832,...). Euler_a, CFG 4..8 for epsilon/3..5 for vpred, 20..28 steps. LCM/PCM/DMD untested, cfg++ samplers work fine, some schedulers not working. Highresfix: x1.5 latent + denoise 0.6 or any gan + denoise 0.3..0.55.

💡 Usage Tip

Please note that the vpred version requires a lower CFG value.

Examples can be found in the repo, more on civitai.

Quality Tags

There are only 4 quality tags:

• Positive: masterpiece, best quality
• Negative: low quality, worst quality

All except low quality in the negative tags can be omitted. Meta tags like lowres have been removed.

Negative Prompt

worst quality, low quality, watermark

💡 Usage Tip

For best results, keep the negative prompt as clean as possible. Spamming popular sequences will not improve results but may lead to unwanted effects, biases, and poor quality.

Artist Styles

The model knows over 35k artist styles. List, grids with example on Mega. Used with by , will not work properly without it.

General Styles

2.5d, anime screencap, bold line, sketch, cgi, digital painting, flat colors, smooth shading, minimalistic, ink style, oil style, pastel style

Natural Text

You can use natural text in combination with booru tags. Version 0.8 has an advanced understanding of natural text prompts, providing state-of-the-art performance among SDXL anime models. However, using only tags is also fine as the understanding of tag combinations is improved.

Brightness/Colors/Contrast

You can use extra meta tags to control brightness, colors, and contrast: low brightness, high brightness, low saturation, high saturation, low gamma, high gamma, sharp colors, soft colors, hdr, sdr

Vpred Version

The Vpred version for RouWei-0.8 will come soon.

📦 Installation

No specific installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

Basic Usage

The model can be used with short booru tag-based or long complex natural text prompts. For example, using classic danbooru-style comma-separated tags without underscores:

# Example of a short tag-based prompt
tag1, tag2, tag3

Advanced Usage

Combining tags with natural text phrases can achieve the best results. For example:

# Example of a combined prompt
tag1, tag2, beautiful scenery, in the style of a famous artist BREAK

📚 Documentation

Base Model and Float Version

You can use FP32 version for more accurate merging or to get benefits from using text encoders in fp32 mode with Comfy. If you want to use RouWei in merges, extract or finetune it without the last aesthetic polishing, you can use the BASE VERSION of RouWei: FP16 FP32

Discord Server

join

Safety

The model tends to generate NSFW images for corresponding prompts. Consider adding extra filtering. Outputs may be inaccurate and provocative and must not be used as a reference.

License

Same as illustrious. Please check the original page for limitations. Feel free to use it in your merges, finetunes, etc., but please leave a link.

Thanks

A number of anonymous persons, Bakariso, dga, Fi., ello, K., LOL2024, NeuroSenko, rred, Soviet Cat, Sv1., T., TekeshiX and other fellow brothers that helped.

Donations

• BTC: bc1qwv83ggq8rvv07uk6dv4njs0j3yygj3aax4wg6c
• ETH/USDT(e): 0x04C8a749F49aE8a56CB84cF0C99CD9E92eDB17db
• XMR: 47F7JAyKP8tMBtzwxpoZsUVB8wzg2VrbtDKBice9FAS1FikbHEXXPof4PAb42CQ5ch8p8Hs4RvJuzPHDtaVSdQzD6ZbA5TZ