majicMIX realistic Maiju Realistic_v7
更新2025-04-11 21:25发布时间2025-04-11 21:25
963
2
34
majicMIX realistic Maiju Realistic_v7 - 1
头像
User_xt6597
类型
Checkpoint
基础模型
SD 1.5
发布时间
2025-04-11 21:25
文件签名
75f0357e0184cd270c62f1495a4b186e67269a184eed48ba9a48567711a314a5
majicMIX realistic 麦橘写实_v7.safetensors
1.99 GB

Community repost: https://modelscope.cn/models/MusePublic/majicMIX_realistic

Recommended: CFG: 7 | 35 steps DPM++ 2M Karras, DPM2 Karras, DPM++ adaptive, DDIM Normal

  • Acceleration LoRA weight: 0.8-1 DDIM sgm_uniform CFG: 1 5 Steps LCM SGM Uniform CFG: 1.5-2 Steps: 12-16, max 18 DPM++ adaptive CFG: 1.5-2 12-16 steps

The smaller the face in the image, the higher the step count should be. For full body shots, faces may distort; perform face restoration using DPM++ adaptive. LCM works best for upper body shots.

Recommended Settings: Sampler: Euler a, Euler, DPM++ 2M Karras (bug fixed) or DPM++ SDE Karras Steps: 2040 Hires Upscaler: R-ESRGAN 4x+ or 4x-UltraSharp Upscale by: 2 Hires steps: 15 Denoising strength: 0.20.5 CFG scale: 6-8 Clip skip: 2 Base Recipe: KanPiroMix + XSMix + ChikMix:

ASIAN ALERT! Recommended Settings: Sampler: Euler a, Euler, restart Steps: 2040 Hires Upscaler: ESRGAN 4x or 4x-UltraSharp or 8x_NMKD-Superscale_150000_G Upscale by: 2+ Hires steps: 15+ Hires denoising strength: 0.050.5 Clip skip: 2 If you need face restoration, please use After Detailer instead of "Restore faces": https://github.com/Bing-su/adetailer I usually enable Dynamic Thresholding to better control CFG values; 1~20 are all worth trying: https://github.com/mcmonkeyprojects/sd-dynamic-thresholding If you need to add filter effects or noise, please use BMAB: https://github.com/portu-sim/sd-webui-bmab Euler is recommended as the sampler.

Take my advice: do NOT enable "Restore faces"! If you need to fix faces, please use After Detailer: https://github.com/Bing-su/adetailer I usually enable Dynamic Thresholding to better control CFG values; feel free to try anything from 1~20.

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding Sorry that I used LoRA Block Weight in previous sample images, which caused confusion and made it difficult for everyone to reproduce my results. So in this new version of sample images, I didn't use any LoRAs.

If you want to learn about LoRA Block Weight, refer to: GitHub - hako-mikan/sd-webui-lora-block-weight Merged with multiple models, it can generate beautiful faces and handle dark areas effectively. Distant faces require inpainting to achieve the best results. You can also use After Detailer.

Recommended Positive Prompts: Best quality, masterpiece, ultra high res, (photorealistic:1.4), 1girl

If you want darker images, please add: in the dark, deep shadow, low key, etc.

Negative Prompts: Use ng_deepnegative_v1_75t and badhandv4 as negative prompts.

I used the bug-fixed DPM++ 2M Karras,

作品

暂无作品