ML Engineer for Video Model

We need an ML Engineer to fine-tune a video model using LoRA and deploy it as an API on RunPod or Hugging Face. The work includes training the model, optimizing performance, and setting up a reliable deployment pipeline. This is a small, part-time project for someone who can deliver a functional video model API efficiently. I'm fine-tuning an open-source video generation model using LoRA training, and I need an engineer to handle the technical setup, training execution, validation, and API deployment on RunPod/ huggingface. I prepare all the data and own the creative direction. Your job is to make the training run correctly, confirm the results are usable, and turn the finished model into a working API endpoint I can call from my application. This is an execution and infrastructure role, not a research or data-prep role. What you'll do Set up the training environment on RunPod — GPU pods, storage for checkpoints, and a working LoRA training pipeline. Run the training jobs on the datasets I provide, with the correct configuration for this model family. Validate the output — generate test results, confirm the trained LoRAs perform as expected, and flag when a run needs to be redone rather than calling it done prematurely. Deploy as an API — serve the base model plus the trained LoRAs as a documented HTTP endpoint (RunPod Serverless preferred) that I can integrate into my own project. Required skills Hands-on experience training LoRAs for diffusion models — video diffusion (Wan 2.1/2.2) strongly preferred; image (SDXL/Flux) only if paired with real video work. Confident with RunPod — Pods, network volumes, and Serverless. Experience with a diffusion training toolchain (ComfyUI, AI Toolkit, DiffSynth, or Musubi). Deploying a generative/image-or-video model behind an API — Docker, a serving handler, GPU memory and cold-start handling. Python, comfortable working independently and reporting progress clearly. Nice to have Prior work serving models on RunPod Serverless specifically. A public portfolio (HuggingFace, Civitai, GitHub, or sample reels) showing trained LoRAs or deployed gen-AI endpoints. Deliverables (must be handed over to me) The trained LoRA weight files (.safetensors). The training configuration used. The serving/handler code and Dockerfile. A short runbook so the setup can be re-run and re-deployed later. All deliverables go into my repository and storage — not a service only you control. How we'll work Fixed-price, milestone-based: Setup + proof — training environment stood up on RunPod, one LoRA trained from a data sample, test results delivered. Full training — remaining LoRAs trained and validated. API deployment — endpoint live, documented, with a working test call. I'd like to start with a small paid trial (Milestone 1) before committing to the full project. Please answer in your proposal Have you trained a LoRA for a video diffusion model on RunPod? Briefly describe the project and the GPU/pod setup you used. How would you expose a trained diffusion model as an API on RunPod, and how do you handle GPU cold starts? How do you decide a LoRA is finished training versus needing another run? Share one link to a LoRA you trained or a generative-AI API you deployed. Budget: [your range] · Type: Fixed-price, milestone-based · Start: ASAP