REED_Illustration_IL_SDXL
🐲🐴 Illustrious Style Merged Recommended Settings
Main Settings
Steps: 20–30 (recommended: 30 for balanced detail and speed)
CFG Scale: 5-7
Sampler: Euler a or DPM++ 2M (both recommended; use DPM++ 2M for smoother gradients and consistent rendering)
VAE: Integrated (no external VAE needed)
Resolution: Larger than 1024x1024 for high detail
Hires Upscale: 1.5
Hires Steps: 20
Hires Upscaler: R-ESRGAN 4x+ Anime6B
Denoising Strength: 0.4 (mid value to balance detail and smoothness)
✅ Positive Prompt
masterpiece,best quality,amazing quality,🚫 Negative Prompt
bad quality,worst quality,worst detail,sketch,censor,⚠️ Safety Rating Tags
general, sensitive, nsfw, explicit
Add “nsfw” to the negative prompt when filtering inappropriate content.
✨ Additional Tips for Quality Enhancement
LoRA compatibility (for 2D anime style):
((anime)), clean lines, bright colors, soft shading, stylized features, chibi, large eyes, cute, dynamic pose, large expressive mouth, high contrast lighting, character design
Pose / Composition suggestions (optional):
full body, standing, facing front, natural pose, dynamic silhouette, backlight, side light, symmetrical pose, 3/4 view, centered subject
💡 Usage Notes
Lower CFG Scale to 6.5–7 if the merged style is too strong.
Increase denoising strength to 0.45–0.5 if details are too sharp or broken.
To combine Illustrious’s metallic aesthetics with Pony’s cuteness, add:
metallic accents, glossy finish, cute design
DPM++ 2M provides smoother results with better lighting consistency, while Euler a is faster for dynamic anime styles
Description
FAQ
Comments (7)
What has changed from v1.0IL to v2.0?
Also, what is the difference between REED_NSFW_illustrious and REED_MIX_illustrious?
REED_NSFW_illustrious has a more stable quality than the Mix version and is more aligned with my personal style of work. The Mix version can create a wider variety of styles based on positive and negative prompts, making it more diverse but also more prone to errors.
Regarding the differences between REED_Illustration_IL_SDXL v1.0 and v2.0, I first addressed the issue of single-hand finger merging that frequently occurred in the previous version. During the testing phase, I found that in three rounds of testing, with 50 test images per round, only 0-3 bad images appeared, which I can say is a version I am personally very satisfied with.
@ReedVaughn
Thank you. I'm currently using REED_NSFW_illustrious and REED_Illustration_IL_SDXL v1.0, but I wasn't sure what the differences were, so this was very helpful. REED_Illustration_IL_SDXL v1.0 is a particularly great model, so I'm excited to hear that the accuracy will be further improved with the upgrade.
@hagurin_rin456 Unfortunately, in order to improve the accuracy of version 2.0, I had to reduce the proportion of stylistic content. If you like the style of version 1.0, I will provide a separate development path to maintain its unique style.
@ReedVaughn Does this mean that the prompt accuracy is lower in 2.0?
I'm not that good at using the prompts, so if the accuracy of the generated illustrations has improved, that's enough for me.
@hagurin_rin456 Maybe v2.0 is more suitable for you, give it a try.