YumeFlux XL (Illustrious)
If you enjoy the model, please add YumeFlux XL to your favorites using the button in the upper-right corner of the page and leave a like. I would also be grateful for any Yellow Buzz gifts! š
Feel free to share your creations on the site. Iām always happy to see what you make and how the model performs with different characters, prompts, and visual ideas.
Hello! Iām excited to introduce YumeFlux XL (Illustrious), my new Stable Diffusion XL model based on the Illustrious architecture.
YumeFlux XL is designed to generate high-quality anime and illustration-style images with expressive characters, refined linework, vibrant colors, soft lighting, and a dreamy visual atmosphere. The model works especially well with detailed character portraits, elegant outfits, dynamic poses, atmospheric backgrounds, and cinematic compositions.
The model understands both simple and detailed prompts. It pays close attention to facial expressions, eyes, hair strands, clothing details, anatomy, lighting, and overall composition. YumeFlux XL can create anything from clean character illustrations to complex scenes with rich environments and dramatic framing.
A suitable VAE is already baked in, providing clear colors, balanced contrast, and crisp details.
I hope you enjoy exploring the creative possibilities of YumeFlux XL!
Recommended Settings
Steps: 20ā35
Sampler: Euler a
Scheduler: Karras
CFG Scale: 2ā6
Resolution: SDXL-compatible resolutions
Clip Skip: 2, if required by your interface
Try different CFG values for each prompt. Lower values can produce softer and more natural images, while higher values may follow the prompt more strictly.
Please avoid using too many quality-enhancing tags or an excessively long negative prompt. The model generally works best with clear and concise prompts.
Positive Prompt
anime style, amazing quality, soft light, detailed clothing,Negative Prompt
bad quality, worst quality, worst detail, sketch, censor, extra fingers, watermark, (eye pupils artifacts), badhandPlease use the model responsibly and follow the Civitai content guidelines.
Description
FAQ
Comments (3)
I would be glad to receive your feedback!
Interesting, I'll give it a try. Is there a list of trained characters, or did you base it on Danbooru?
Based on Danbooru tags, the model was not additionally trained on characters, only a better understanding of lighting, clothing, and more correct poses when specifying them (no longer is it necessary to specify poses in super detail if you do not use ControlNet)











