✨ Soft Light, Rich Details — A Touch of Elegance! 🌸
Sometimes, beauty is all about the little details. ✨
Delicate hair strands, elegant outfits, beautifully refined textures, and soft, glowing light — all coming together to create illustrations filled with charm and atmosphere.
I especially love how the gentle lighting brings out the subtle colors and glossy highlights, giving each image a soft yet vibrant finish.
From intricate clothing details to graceful expressions, there's something satisfying about watching all those little elements come together.
Soft and delicate, yet rich and expressive — a beautiful balance of simplicity and detail!
Of course, these sample images are just one way to enjoy the model.
Feel free to experiment with your favorite LoRAs, prompts, and generation settings to discover the look that suits your taste.
A little creativity, a little experimentation, and maybe your next favorite masterpiece! 😆✨
🎨 Positive Prompt
masterpiece, sharp anime style, best quality, amazing quality, highres, newest, very aesthetic, absurdres, vivid color, high contrast, clear outline🚫 Negative Prompt
worst quality, bad quality, low quality, low detail, worst detail, bad detail, bad anatomy⚙️ Recommended Generation Settings
SettingValue
Sampler：Euler a / DPM++ 2M / DPM++ SDE
SchedulerKarrasSteps：30–50
Upscaler：RealESRGAN_x4plus_anime_6B
Upscale Factor：2x
Denoising Strength：0.4
💡 A Little Tip
These are my usual generation settings, not strict requirements.
Feel free to mix in your favorite LoRAs and adjust the settings to find your ideal balance of detail, color, and atmosphere.
Your style, your imagination — make every detail shine! ✨
Description
FAQ
Comments (3)
Almost every model from you has baked a VAE that results in faint colors and low contrasts compared to the default sdxl_vae. In my model tests, I often replaced the baked VAE with an externally loaded sdxl_vae to fix this and commented it to let you know about this problem:
https://civitai.red/images/146132853
https://civitai.red/images/145357089
https://civitai.red/images/145136624
https://civitai.red/images/144936889
https://civitai.red/images/145136567
https://civitai.red/images/144952598
https://civitai.red/images/144936886
https://civitai.red/images/144865435
https://civitai.red/images/144936879
https://civitai.red/images/144756675
etc...
This is now the last time that I manually replaced the VAE when testing your models, because maybe faint colors and low contrast are actually your intended style. Otherwise you should bake sdxl_vae in your future models to fix this.
Thank you so much for your continued support and detailed feedback!
To be honest, I've always used an external VAE for my own image generation, so I hadn't paid much attention to the VAE embedded in my merged checkpoints.
Your feedback made me realize this could be affecting how others experience my models.
From now on, I plan to bake the standard SDXL VAE into my future merged models.
I really appreciate the time and effort you've put into testing my models. Thanks again! 😊
@AI_Alvis Great, thank you for fixing this in future versions :)