Creative ML Technologist

What You'll Do

• Compose workflows: You will run focused experiments that answer product questions using ComfyUI graphs that combine perception and generative steps.
• Measure and compare: You will produce side-by-side comparisons with qualitative frames and a set of meaningful metrics to guide decision-making.
• Custom nodes and integrations: You will develop, implement, and deploy custom ComfyUI nodes for novel functionality and create wrappers for new open-source models.
• Document and guide: You will document inputs, controls, expected artifacts, and typical failure modes in concise how-to notes and manifests so others can rerun and extend your work.
• Collaborate and iterate: You will partner closely with creative users to refine and validate the visual output and work with engineering to hand off proven workflows for scaling.

What you’ll bring:

• ComfyUI experience: You have demonstrable experience building multi-step ComfyUI graphs and compile and deploy open sourced projects through wrappers. This includes building projects from source, compiling them as needed, and integrating them to keep workflows readable, reusable.
• Creative and technical skill: You combine strong creative sensibilities with practical, hands-on machine learning skills.
• Python and ML proficiency: Solid Python and PyTorch fundamentals with hands-on experience, including tensor shapes, dtypes, autocast or no-grad, memory-aware batching, and devices.
• Workflow design and evaluation: You have experience assembling perception and generative workflows for visual tasks and evaluating tradeoffs with simple, meaningful metrics.
• VFX/Games production experience: Professional experience in VFX or game development with practical knowledge of content creation pipelines and cross-functional collaboration.
• Clear communication: You have the ability to explain complex findings clearly and effectively to both creative and engineering stakeholders.

What will help you stand out:

• Training experience: You have experience training small adapters, such as LoRAs, when it directly improves a workflow’s performance or quality.
• ComfyUI portfolio: A portfolio with one or more original ComfyUI nodes or wrappers and a short write-up of the design decisions.
• Generative editing quality: Examples of generative editing or scene adjustments that deliver high accuracy and visual fidelity, with speed gains a plus.