Enchanting AI Persona Development - 10/02/2026 00:55 EST
Budget: $250 – $750 USD
I am seeking a highly skilled AI technician to develop high-fidelity LoRA models for FLUX.1 [dev] and Wan 2.1. The goal is to achieve [insert goal: e.g., high character likeness / a specific artistic style] that remains consistent across both high-resolution images and video generation.
Required Technical Expertise
Model Experience: Proven experience training LoRAs for FLUX.1 (Dev/Schnell) and Wan 2.1 (specifically the 1.3B or 14B models).
Tooling: Proficiency with Ostris AI-Toolkit, Musubi-tuner, or Kohya_ss.
Dataset Management: Ability to curate and caption high-quality datasets (images for Flux, video clips for Wan). Experience with Danbooru/WD14 tagging or natural language captioning (LLM-based).
Temporal Consistency: For Wan 2.1, you must be able to demonstrate that the LoRA maintains identity/style without significant "flicker" or warping during motion.
Scope of Work
Dataset Audit: i need to create a dataset
Training & Optimization: Execute multiple training runs, adjusting learning rates and rank/alpha settings for the best balance of flexibility and likeness.
Testing & Validation: Provide a sample sheet showing the LoRA performing across different prompts and aspect ratios.
Final Delivery: Provide the .safetensors files along with the recommended "Trigger Words" and optimized ComfyUI workflows (if applicable).
Preferred Qualifications
Active profile on Civitai or Hugging Face with public LoRA examples.
Deep understanding of ComfyUI nodes for both Flux and Wan.
Ability to troubleshoot VRAM constraints and training crashes.
Required Technical Expertise
Model Experience: Proven experience training LoRAs for FLUX.1 (Dev/Schnell) and Wan 2.1 (specifically the 1.3B or 14B models).
Tooling: Proficiency with Ostris AI-Toolkit, Musubi-tuner, or Kohya_ss.
Dataset Management: Ability to curate and caption high-quality datasets (images for Flux, video clips for Wan). Experience with Danbooru/WD14 tagging or natural language captioning (LLM-based).
Temporal Consistency: For Wan 2.1, you must be able to demonstrate that the LoRA maintains identity/style without significant "flicker" or warping during motion.
Scope of Work
Dataset Audit: i need to create a dataset
Training & Optimization: Execute multiple training runs, adjusting learning rates and rank/alpha settings for the best balance of flexibility and likeness.
Testing & Validation: Provide a sample sheet showing the LoRA performing across different prompts and aspect ratios.
Final Delivery: Provide the .safetensors files along with the recommended "Trigger Words" and optimized ComfyUI workflows (if applicable).
Preferred Qualifications
Active profile on Civitai or Hugging Face with public LoRA examples.
Deep understanding of ComfyUI nodes for both Flux and Wan.
Ability to troubleshoot VRAM constraints and training crashes.