flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
Back
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0
flux2 klein base 9b YX - Realistic Facial Texture Details V2.0

flux2 klein base 9b YX - Realistic Facial Texture Details V2.0

132 Views0 Likes5 Favorites
Characterization

YXW

YXW

Characterization

Model Information

Active
Model Type:
LoRA
Basic Model:
Flux2-Klein-9B
Resource Name:
models/loras/flux2 klein 9b Y-v2.0-8_000002600.safetensors
MD5:
34c4c6cae2a1bd7f92ecf4f84e474070

V2.0
For Checkpoint: flux2 klein base 9b (Do not use flux2 klein 9B/4B; using 9B/4B models has a certain probability of making characters appear older)

Steps: 8-20 (When ineffective, first increase the LoRA strength; if still ineffective, add steps, default is 8 steps)

cfg: 1 (Do not add acceleration LoRA or other LoRA, otherwise it will affect the model weights)

Model Weight: 0.3-1 (0.75 recommended by default)

The LoRA weight and steps for each image are inconsistent, because the depth of facial texture, lighting, and angle of every image are different, so there are no overly uniform parameter recommendations, only approximate values can be provided. Change the specific values according to your own needs.

This is the second version (Original Image Version — the skin of generated characters will be closer to the original image's skin, note!! There will be a slight reddish tint, the previous version was the Yellow Skin Version). It fixes many issues with facial redness, but some images still suffer from severe redness, so I changed a sampler, which can more effectively reduce redness. Version 1 address: https://www.RunningHub.cn/model/public/2069468964757004290

New instructions added: This LoRA is mainly used to add facial texture details, but with one prerequisite: the original image is AI-generated (real photos sometimes work, sometimes have severe redness; anyway, real photos are quite unstable), and it is best to be an image around 1k with a clearly visible face (profile views, high angles, and low angles are all fine). The photos generated by gpt image 2 lacked sufficient facial texture details, so I trained this LoRA to enhance facial textures. This LoRA provides a native enhancement for images generated by gpt image 2. It only works effectively for images below 2k!! Input images above 2k are destined to have inconspicuous effects, which is a limitation of open-source models. To my surprise, when I input a 4k image, the facial texture unexpectedly decreased, but it also smoothed out the facial textures, giving me a feeling of being more natural. This can also be considered a disguised repair of some messy texture details, nevertheless, I still do not recommend images above 2k.

Workflow: https://www.RunningHub.cn/post/2072296887910232066
AI Application Address: https://www.RunningHub.cn/ai-detail/2073320263101538306
AI applications can be used directly

V2.0
For Checkpoint: flux2 klein base 9b (Do not use flux2 klein 9B/4B; using 9B/4B models has a certain probability of making characters appear older)

Steps: 8-20 (When ineffective, first increase the LoRA strength; if still ineffective, add steps, default is 8 steps)

cfg: 1 (Do not add acceleration LoRA or other LoRA, otherwise it will affect the model weights)

Model Weight: 0.3-1 (0.75 recommended by default)

The LoRA weight and steps for each image are inconsistent, because the depth of facial texture, lighting, and angle of every image are different, so there are no overly uniform parameter recommendations, only approximate values can be provided. Change the specific values according to your own needs.

This is the second version (Original Image Version — the skin of generated characters will be closer to the original image's skin, note!! There will be a slight reddish tint, the previous version was the Yellow Skin Version). It fixes many issues with facial redness, but some images still suffer from severe redness, so I changed a sampler, which can more effectively reduce redness. Version 1 address: https://www.RunningHub.cn/model/public/2069468964757004290

New instructions added: This LoRA is mainly used to add facial texture details, but with one prerequisite: the original image is AI-generated (real photos sometimes work, sometimes have severe redness; anyway, real photos are quite unstable), and it is best to be an image around 1k with a clearly visible face (profile views, high angles, and low angles are all fine). The photos generated by gpt image 2 lacked sufficient facial texture details, so I trained this LoRA to enhance facial textures. This LoRA provides a native enhancement for images generated by gpt image 2. It only works effectively for images below 2k!! Input images above 2k are destined to have inconspicuous effects, which is a limitation of open-source models. To my surprise, when I input a 4k image, the facial texture unexpectedly decreased, but it also smoothed out the facial textures, giving me a feeling of being more natural. This can also be considered a disguised repair of some messy texture details, nevertheless, I still do not recommend images above 2k.

Workflow: https://www.RunningHub.cn/post/2072296887910232066
AI Application Address: https://www.RunningHub.cn/ai-detail/2073320263101538306
AI applications can be used directly