FLUX LoRA Photorealistic Face Enhancement and light blending body
Budget: $250 – $750 USD
I am looking for an expert in training **LoRAs for the Flux model, The goal is to create a LoRA that achieves three distinct things simultaneously:
1. **100% Face Preservation:** The primary requirement is near-perfect identity retention. The face must not "drift" or change features during the generation process.
2. **Natural Human Skin Tones (High Dynamic Range):** Even in difficult lighting scenarios like **Sunset or Golden Hour**, the human body must maintain natural, bright, and vibrant colors. I want to avoid the "muddy" or overly dark skin tones that often occur in sunset AI generations.
3. **Background Flexibility & Color Blending:** The LoRA should focus heavily on background manipulation (compositional variety) while allowing for light color matching between the background and the subject to ensure the subject doesn't look "pasted" on.
### **Key Technical Requirements:**
* **Dataset Preparation:** Ability to curate/tag a high-quality dataset with varied lighting but consistent facial features.
* **Captions:** Expert use of Booru-style or Natural Language tagging to separate the "subject" (identity) from the "lighting/environment."
* **Testing:** You must provide XYZ plots showing the LoRA's performance at different weights across various prompts (Daylight vs. Sunset).
* **Deliverable:** The final `.safetensors` file and a brief guide on the best CFG/Step settings for Flux.
### **Ideal Candidate:**
* Has a portfolio of Flux LoRAs (preferably on Civitai or HuggingFace).
* Deep understanding of **Rank (Dim)** and **Alpha** settings to prevent overcooking the face.
* Experience with **Kohya_ss**, **OneTrainer**, or similar training environments.
1. **100% Face Preservation:** The primary requirement is near-perfect identity retention. The face must not "drift" or change features during the generation process.
2. **Natural Human Skin Tones (High Dynamic Range):** Even in difficult lighting scenarios like **Sunset or Golden Hour**, the human body must maintain natural, bright, and vibrant colors. I want to avoid the "muddy" or overly dark skin tones that often occur in sunset AI generations.
3. **Background Flexibility & Color Blending:** The LoRA should focus heavily on background manipulation (compositional variety) while allowing for light color matching between the background and the subject to ensure the subject doesn't look "pasted" on.
### **Key Technical Requirements:**
* **Dataset Preparation:** Ability to curate/tag a high-quality dataset with varied lighting but consistent facial features.
* **Captions:** Expert use of Booru-style or Natural Language tagging to separate the "subject" (identity) from the "lighting/environment."
* **Testing:** You must provide XYZ plots showing the LoRA's performance at different weights across various prompts (Daylight vs. Sunset).
* **Deliverable:** The final `.safetensors` file and a brief guide on the best CFG/Step settings for Flux.
### **Ideal Candidate:**
* Has a portfolio of Flux LoRAs (preferably on Civitai or HuggingFace).
* Deep understanding of **Rank (Dim)** and **Alpha** settings to prevent overcooking the face.
* Experience with **Kohya_ss**, **OneTrainer**, or similar training environments.