💐Favorite now: StableDiffusion recommended real portrait model

💐Favorite now: StableDiffusion recommended real portrait model

May 19th, 2023
View OriginalTranslated by Google

The second issue of model sharing is coming. This issue is a series of real-life models , all of which are collected for personal use!

Chillout MIX

Features: universal mask, Asian face

The famous ChilloutMix, the model body is trained based on Stable Diffusion 1.5 version, on this basis, the fusion of two important models is added: Basilmix and Dreamlike, the former is a semi-realistic "2.5D style" after drawing realistic textures and Asian faces The latter is a derivative model, which is optimized using photo set data to make the generated results more 3D realistic.

In addition, the author also added some small Embeddings to the model to enhance the performance of the AI model. The scenes drawn by ChilloutMix are also excellent, and the effects do not belong to DreamShaper. In addition, the model has a matching LoRA: koreadoll\taiwandoll\japanese, the effect of superimposed drawing is remarkable, but it is easy to produce NSFW pictures-. -

ChilloutMix is also widely used for model fusion, especially the two-dimensional model, such as Aniflatmix - Anime Flat Color Style Mix made by Oedo. It is worth noting that this model is controversial because it is used to refine a real-life model. The author also stated in civitai that this model is prohibited in principle from being used for training style model training based on portraits of stars and public figures.

use :


The author recommends parameters

  • Sampler: DPM++ 2M Karras
  • Steps: 25
  • CFG Scale: 7
  • Hires. Fix: On
  • Upscaler: R-ESRGAN 4x
  • Hire Steps: 0
  • Denoising Strength: 0.7
  • Width*Height: 768*1024
  • Negative embedding: badhandv4 , bad-picture-chill-75v


Features: Asian face shape, can draw Asian men

The full name of BRA: Beautiful Realistic Asians, the current author @PleaseBanKai has released BRAV3/BRAV1/BRAV4 and the latest paid version of BRAV5, V3 is mainly a female front face image, and the processing of portrait background is relatively weak, V1 adds posture, but full body processing Relatively average, big head effect is remarkable😂

The author of V4 has added male photo training, and it is more likely to appear female pictures when the prompt words are relatively weak and do not emphasize men. At present, the best effect is V5 (after all, it is a paid version). The effect of processing female half-length images is the most prominent, and the facial details have also increased a lot. The most important thing is that it can handle group portraits. Enable hire.fix and reverse words, basically Multiple faces can be handled very well, and the image output is relatively stable.

use :


Recommend a set of Negative: NSFW, text, watermark, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck


Features : European face, can draw real objects

The full name of the model is Limitless Originality Free from Interference, which is characterized by no special face arrangement; improved line details; and improved timely understanding. Currently there are three major versions, V1, V2, and V2.1, with vae and pre versions added in the middle. V2 added more camera angles; V2.1 added detail effects, the face effect was reverted to a version, and a redraw model was added. After using LOFI, the overall picture is grayish, and VAE needs to be added. The portraits are mainly European girls. V2 supports various camera shooting methods, and can also be used for color correction.

use :

Description : Recommended parameter settings Steps: 20, Sampler: Euler a, CFG scale: 7, Size: 256x384, Denoising strength: 0.75, Hires upscale: 2, Hires steps: 40, Hires upscaler: Latent (bicubic antialiased) Negative Embedding: DeepNegative

Cyber Realistic

Features : European and American faces, film lighting texture

Versatile and realistic model, the author has incorporated several custom elements to add a layer of uniqueness to its output. One of the main strengths of the model is that it can efficiently handle text inversion and LORA, providing accurate and detailed output. In addition, the model can produce better picture effects by inputting very few prompt words. The disadvantage is that the faces generated by too simple prompt words are similar, which can be changed by adjusting the number of seeds or steps. It is worth reminding that this model needs to cooperate with VAE. See below for specific usage instructions.

use :

Description :

The author recommends using

MajicMIX realistic

Features : light and shadow effects, support for both male and female face shapes

There are three versions of this model: V2\V3\V4. The second version has improved the weight and light and shadow effects of specific components. It integrates a variety of models, which can generate beautiful faces and effectively deal with dark parts.

Long-distance faces require inpaint to achieve the best results. Special reminder , don't use face restoration! The method of face repair is to inpaint the face: inpaint-->only masked-->set to 512x512-->Denoising strength: 0.2~0.5.

The third version improved the body model and continued to strengthen the light and shadow. The disadvantage is the failed color problem. The fourth version is re-integrated. If you want a darker image increase: in the dark, deep shadow, low key, etc.


illustrate: Recommended parameters

  • Sampler: DPM++ 2M Karras (bug-fixed) or DPM++ SDE Karras
  • Steps: 20~40
  • Hires upscaler: R-ESRGAN 4x+ or 4x-UltraSharp
  • Hires upscale: 2
  • Hire steps: 15
  • Denoising strength: 0.2~0.5
  • CFG scale: 6-8
  • clip skip: 2
  • Negative Embedding: ng_deepnegative_v1_75t , badhandv4
  • Recommended positive prompts: Best quality, masterpiece, ultra high res, (photorealistic:1.4), 1girl

Majic MIX lux

Features : Real-life oil painting texture, 2.5 cut 3D, suitable for drawing cards

It feels like a fusion model of Chilloutmix and ReV Animated. The characters add a sense of oil painting. According to the emphasis of the prompt words, they can switch between 2.5D and real. According to the author, this model combines MajicMIX fantasy on the basis of MajicMIX realistic, which is both realistic and fantasy. The author recommends a set of prompt words for drawing cards: official art, unity 8k wallpaper, ultra detailed, beautiful and aesthetic, masterpiece, best quality, (zentangle, mandala, tangle, entangle), 1girl, extremely detailed, dynamic angle, cowboyshot , the most beautiful form of chaos, elegant, a brutalist designed, vivid colours, romanticism, by james jean, roby dwi antono, ross tran, francis bacon, michal mraz, adrian ghenie, petra cortright, gerhard richter, takato yamamoto , ashley wood , atmospheric, ecstasy of musical notes, streaming musical notes visible


illustrate: Recommended Use

  • Negative Embedding: ng_deepnegative_v1_75t , badhandv4
  • Recommended positive prompts: official art, unity 8k wallpaper, ultra detailed, beautiful and aesthetic, beautiful, masterpiece, best quality, (zentangle, mandala, tangle, entangle)
  • Inpaint the face: inpaint-->only masked-->set to 512x512-->Denoising strength: 0.2~0.5

Do you have any real-life models that you like to use? Welcome to comment and share!

Happy Alchemy 🎉