ComfyUI & Character LoRA Tutor: Setup, Training & Video-to-Video Workflow (ThinkDiffusion & RunPod)

I am looking for an experienced AI video engineer to guide me 1-on-1 through the complete pipeline of building a custom character LoRA and using it in Video-to-Video (V2V) motion-matching workflows. We will use ThinkDiffusion for dataset pruning and LoRA training (taking advantage of its pre-loaded interfaces without massive local model downloads) and RunPod for uncensored, private V2V video rendering. This is strictly an instructional and setup project. You will be teaching me via live 1-on-1 screenshare sessions (Zoom / Google Meet) and providing working .json workflow files so I can run everything independently afterward. Scope of Work Milestone 1: Dataset Curation & LoRA Training on ThinkDiffusion * Review my existing stash of 100+ raw photos and help me select/prune down to the top 15–25 optimal images (balancing headshots, upper body, full body, and varied expressions). * Show me best practices for 1024x1024 cropping and text captioning (.txt files) with custom trigger words. * Guide me through configuring and launching a character LoRA training run on ThinkDiffusion (using Kohya_ss or AI-Toolkit). * Evaluate output checkpoints to ensure the face looks exact without overtraining or freezing motion capabilities. Milestone 2: RunPod Private GPU Setup * Guide me through deploying a persistent RunPod template with attached storage so my files persist across pod restarts. * Help me set up ComfyUI on RunPod with required custom nodes for motion transfer (DWPose, ControlNet, Wan 2.1 Fun-Control modules). * Show me where to upload our finished character .safetensors file into the RunPod directory. Milestone 3: V2V Motion Matching Walkthrough ($50) * Provide a clean, fully functional ComfyUI Video-to-Video JSON workflow graph. * Walk me step-by-step through setting up a driver video, extracting pose motion, applying my custom LoRA, and rendering the final uncensored video. * Teach me basic troubleshooting for frame consistency, face restoration/detailers, and motion artifacts. Requirements * Proven track record training character LoRAs (SDXL, Flux, or custom base models) using Kohya_ss / AI-Toolkit. * Hands-on experience with ComfyUI V2V workflows (DWPose, ControlNet, AnimateDiff, or Wan 2.1). * Direct experience operating both ThinkDiffusion and RunPod environments. * Fluent English and ability to teach clearly over 1-on-1 live screenshare. Contract Offer Details * Type: Fixed Price ($100 Total across 3 Milestones) * Milestones: * Dataset Selection, Captioning & ThinkDiffusion LoRA Training * RunPod Environment Setup & Node Configuration * ComfyUI Motion-Matching V2V Masterclass Application Requirement: Portfolio Verification To be considered for this project, you must attach or link 2–3 visual proof samples in your proposal: * Character LoRA Samples: High-resolution images showing a custom character LoRA you trained (include close-up, upper-body, and full-body examples to prove face/body consistency). * Video-to-Video (V2V) Sample: A short video clip (or GIF) showcasing a motion-matching/V2V output generated using ComfyUI (DWPose, ControlNet, or Wan 2.1). Note: Proposals submitted without verified visual/video work samples will not be considered.