AI Training & Workflow Engineer — ComfyUI / LoRA & Fine-Tuning (Video&Image)

AI Training & Workflow Engineer — ComfyUI / LoRA Trainer Optimization & Fine-Tuning (Image & Video) — 100% Remote

Training-pipeline optimization and workflow optimization are the two cores of this role. On one side, you will optimize and improve our LoRA training setup itself — the trainer, the configs, the pipeline — to get better quality, faster speed, and lower cost. On the other, you will design and continuously optimize our ComfyUI generation workflows. This is an engineering role focused on improving HOW we train and generate — not a data-collection or run-the-training operator role.

Models & Tools You'll Work With
• ComfyUI — primary workflow environment
• LTX 2.3 — video generation
• Wan 2.2 — video generation / image-to-video
• Qwen-Image-Edit — image editing / instruction-based editing
• Z-Image — Alibaba Tongyi 6B model (Turbo / Base / Edit variants; including LoRA training via DiffSynth-Studio and ComfyUI integration)
Training toolkits: AI-Toolkit, Musubi Tuner (and similar). Familiarity with the broader open-source landscape — new model releases, quantization, ComfyUI custom nodes — is a strong plus, since this space moves fast.
• Actively following the latest open-source image or video generation model releases,

Requirements:
• Staying on top of the open-source community — actively following the latest open-source image or video generation model releases, evaluating them as soon as they drop, and building ComfyUI workflows around them quickly.
• Deep, hands-on experience optimizing LoRA training — not just running trainers, but tuning configs and hyperparameters and improving training quality, speed, and cost (you must share concrete before/after examples or sample outputs).
• Expert-level knowledge of AI-Toolkit and Musubi Tuner — able to configure, tune, and get the most out of these trainers.
• Experience with full fine-tuning of base models, not just LoRAs — including training config and evaluation.
• Proven ability to build and optimize complex ComfyUI workflows (not just using prebuilt ones).
• Deep understanding of diffusion models and the mechanics of training (learning rate, rank, regularization, overfitting control, schedulers).
• Experience optimizing training pipelines for speed, cost, and quality.
• Experience with video generation / motion workflows (LTX 2.3, Wan 2.2, ).
• Python scripting for automation, batch training jobs, or custom nodes.
• Good English communication (this is a remote role — clear written communication is essential).
• Organized, detail-oriented, and self-disciplined — comfortable working independently in a remote setup.


Primary responsibilities:
• LoRA trainer optimization — improving and tuning the LoRA training setup itself: training configs, hyperparameters (rank, learning rate, scheduler), trainer settings, and pipeline efficiency, to maximize output quality while reducing training time and cost. This is about optimizing how LoRAs are trained, not manually collecting datasets or babysitting training runs. Deep hands-on knowledge of AI-Toolkit and Musubi Tuner is required.
• Base-model fine-tuning — full fine-tuning of open-source base models, with the same optimization mindset toward configs, training efficiency, and output quality.
• Workflow optimization — designing, building, and continuously improving ComfyUI workflows; optimizing node graphs, render efficiency, VRAM usage, batching, and model offloading for repeatable, production-grade output.
• Training pipeline optimization — improving training speed, cost, stability, and quality; managing and optimizing multi-GPU training runs.
• Evaluation & iteration — building systematic evaluation methods to compare configs, checkpoints, and training approaches, and iterating toward measurable quality gains.
Supporting responsibilities:
• Creating consistent character outputs across generations
• Image-to-video / text-to-video, motion transfer, and lip sync workflows
• Upscaling, refinement, and detail enhancement
• Organizing workflow files, models, and documentation into clean, repeatable systems


Work Setup:
• 100% remote / work from home — work from anywhere
• Long-term, ongoing engagement
• You manage your own schedule around agreed deliverables and check-ins

About Us:
We are a fast-growing AI company building AI-powered content systems. We focus heavily on automation, workflow efficiency, and high-quality visual production. We are looking for a technical AI Training & Workflow Engineer to join our team long-term. This is a fully remote, work-from-home position — you can work from anywhere with a stable internet connection.
Your Role

To Apply:
Please include: (1) examples of LoRA training optimization work — configs you've tuned, before/after quality or speed improvements, and sample outputs, ideally using AI-Toolkit or Musubi Tuner; (2) examples of ComfyUI workflows you've built; and (3) a short note on which open-source models and trainers you've worked with most recently.