Generate consistent faces from different angles workflow for ComfyUI
Showcase:
Setup Instructions:
Required model files - Flux Dev - Flux Dev Fill - LoRA: flux1-depth-dev-lora; chinfixer-2000; skin-texture-style-v5 (or alternatives) - Ultralytics: Eyes.pt or Eyeful_v2-Paired.pt
How it works - Upload a reference 3x3 grid image (https://imgur.com/a/ZEZp82J ) - Set flux1-depth_dev_lora strength (default 0.75). - Set FluxGuidance (default 4.0). - Set Sampler and Scheduler (default euler/beta). - Set Steps (default 35). - Set Seed (random or fixed). - Select an upscale model and set upscale_by 3 (default model 4x_RealWebPhoto_v4). - Modify the Prompt to describe in detail the subject appearance (skin tone, physique, face shape, hair and eyes). - In the Step 2: Set the strength of the Skin LoRA (default 0.75) - Set the dishonesty_factor for detail, lower negative numbers mean higher detail (default -0.05 - use this with care, it may change the image a lot!) - Start generating the images.
The workflow will upscale by 3 an image and then generate 9 images (1024x1024) from it, this mean it has to process FaceDetailer, EyesDetailer, apply Skin Lora and apply Chin-fix Lora for 9 images. This is why the whole generation will take very long. It is normal.
Generate consistent faces from different angles workflow for ComfyUI
Showcase:
Setup Instructions:
Required model files - Flux Dev - Flux Dev Fill - LoRA: flux1-depth-dev-lora; chinfixer-2000; skin-texture-style-v5 (or alternatives) - Ultralytics: Eyes.pt or Eyeful_v2-Paired.pt
How it works - Upload a reference 3x3 grid image (https://imgur.com/a/ZEZp82J ) - Set flux1-depth_dev_lora strength (default 0.75). - Set FluxGuidance (default 4.0). - Set Sampler and Scheduler (default euler/beta). - Set Steps (default 35). - Set Seed (random or fixed). - Select an upscale model and set upscale_by 3 (default model 4x_RealWebPhoto_v4). - Modify the Prompt to describe in detail the subject appearance (skin tone, physique, face shape, hair and eyes). - In the Step 2: Set the strength of the Skin LoRA (default 0.75) - Set the dishonesty_factor for detail, lower negative numbers mean higher detail (default -0.05 - use this with care, it may change the image a lot!) - Start generating the images.
The workflow will upscale by 3 an image and then generate 9 images (1024x1024) from it, this mean it has to process FaceDetailer, EyesDetailer, apply Skin Lora and apply Chin-fix Lora for 9 images. This is why the whole generation will take very long. It is normal.
Generate consistent faces from different angles workflow for ComfyUI
Showcase:
Setup Instructions:
Required model files - Flux Dev - Flux Dev Fill - LoRA: flux1-depth-dev-lora; chinfixer-2000; skin-texture-style-v5 (or alternatives) - Ultralytics: Eyes.pt or Eyeful_v2-Paired.pt
How it works - Upload a reference 3x3 grid image (https://imgur.com/a/ZEZp82J ) - Set flux1-depth_dev_lora strength (default 0.75). - Set FluxGuidance (default 4.0). - Set Sampler and Scheduler (default euler/beta). - Set Steps (default 35). - Set Seed (random or fixed). - Select an upscale model and set upscale_by 3 (default model 4x_RealWebPhoto_v4). - Modify the Prompt to describe in detail the subject appearance (skin tone, physique, face shape, hair and eyes). - In the Step 2: Set the strength of the Skin LoRA (default 0.75) - Set the dishonesty_factor for detail, lower negative numbers mean higher detail (default -0.05 - use this with care, it may change the image a lot!) - Start generating the images.
The workflow will upscale by 3 an image and then generate 9 images (1024x1024) from it, this mean it has to process FaceDetailer, EyesDetailer, apply Skin Lora and apply Chin-fix Lora for 9 images. This is why the whole generation will take very long. It is normal.
Generate consistent faces from different angles workflow for ComfyUI
Showcase:
Setup Instructions:
Required model files - Flux Dev - Flux Dev Fill - LoRA: flux1-depth-dev-lora; chinfixer-2000; skin-texture-style-v5 (or alternatives) - Ultralytics: Eyes.pt or Eyeful_v2-Paired.pt
How it works - Upload a reference 3x3 grid image (https://imgur.com/a/ZEZp82J ) - Set flux1-depth_dev_lora strength (default 0.75). - Set FluxGuidance (default 4.0). - Set Sampler and Scheduler (default euler/beta). - Set Steps (default 35). - Set Seed (random or fixed). - Select an upscale model and set upscale_by 3 (default model 4x_RealWebPhoto_v4). - Modify the Prompt to describe in detail the subject appearance (skin tone, physique, face shape, hair and eyes). - In the Step 2: Set the strength of the Skin LoRA (default 0.75) - Set the dishonesty_factor for detail, lower negative numbers mean higher detail (default -0.05 - use this with care, it may change the image a lot!) - Start generating the images.
The workflow will upscale by 3 an image and then generate 9 images (1024x1024) from it, this mean it has to process FaceDetailer, EyesDetailer, apply Skin Lora and apply Chin-fix Lora for 9 images. This is why the whole generation will take very long. It is normal.
![[Image: 29030.jpg]](https://i.postimg.cc/3xt7n5c0/29030.jpg)
![[Image: rainbow-diamond.gif]](https://i.postimg.cc/QCBM79jv/rainbow-diamond.gif)
