🎀 MaideNShinE ✨

This is an epsilon model created through meticulously merging models. Illustrious 2.0 was used as a base, I merged it with NoobAI eps using the Dare-Ties method and used add difference to add RouWei eps subtracting illustrious 0.1. Clips were merged separately using Dare clip merge. Followed by merging other models with various methods.

This model's base style lets you make the most of artist tags and forged with the less is more principle. Being based on Illustrious 2.0 the model understands natural language as well but if you want to go that route I recommend using a mix of natural and classic Danbooru tags style prompting.

Recommended Settings:

ComfyUI is recommended.
Sampler: DPM++2M, DPM++_2M_SDE (and its CFG_PP variant), DPM++SDE, Euler, Euler A (and its CFG_PP variant), Deis
Scheluder: Karras, Normal, Beta, SGM-Uniform, AYS
VAE is baked, but you can use one that you prefer.

Steps: 20~40, Clip Skip: 2, CFG Scale: 3-7
Try using RescaleCFG set to 0.5 with CFG above 5.5. and with CFG_PP samplers use CFG Scale: 1-3.

Resolutions : I prefer 1MP : 896 x 1152, 1216 x 832, 832 x 1216, 1344 x 768, 768 x 1344, 1536 x 640, 640 x 1536 higher resolutions are also supported, courtesy of Illustrious 2.0: 1536x1536, 1728x1344, 1344x1728, 1824x1248, 1248x1824, 2016x1152, 1152x2016. Using higher generations can yield fantastic results but it was a little unstable for my preference.

Using Hi-Res Fix and Face Detailer is recommended.

For Hi-Res and Ultimate SD Upscale, You can use anything, I like NMKD YandereNeo, Lollypop, and Animesharp. For simple upscaling I use IllustrationJaNai-V1-DAT2

Check the image previews, comfyui workflows are included in the image metadata, click and drag them to the comfy. WebUI users should use the PNG INFO tab to drag in the images.

Prompt Recommendations:

Positive Prompts: masterpiece, best quality, newest, absurdres, highres, intricate details, ultra-detailed, volumetric lighting

Negative Prompts: worst quality, low quality, jpeg artifacts, watermark, missing fingers, extra digits, distorted anatomy, text overlay, simple background, old, early

You don't need embeddings, but for me, it feels weird not using one, it's a habit from the old 1.5 days, I tested a few with this model and liked smooth pos and neg embeddings the best.

Rules & Usage Guidelines:

The license is the same as with the ones used as the base of this model. Since this model includes NoobAI , Illustrious, and RouWei in its merge, I recommend you read their usage restrictions.

I'm going to keep it simple and give you a TLDR. Do not sell or monetize this model in any way. You're welcome to share your creations anywhere however you are fully responsible for what you generate and share, model authors won't be held accountable for anything you create using this model. The same rules will apply to models that use this model in their merge.