New opportunity

Research Assistant (Building a Local AI Personalized assistant for Hand Drawn Animation Artist with ComfyUI)

Nanyang Technological University · Singapore

About this role

The School of Art, Design and Media at Nanyang Technological University seeks a Research Assistant to build a local, privacy-safe AI assistant for hand-drawn animation artist Professor Raimund Krumme. The system should help bring his analogue ink-and-brush process into an AI workflow that he can use independently during his residency from 14 September to 13 November 2026. The role is based at NTU Main Campus, Singapore.

Responsibilities

  • Set up and optimize local, open-source Stable Diffusion and ComfyUI models on high-end workstation hardware without cloud dependency.
  • Train local LoRAs and ControlNets using a curated archive of the artist’s sketches and scans.
  • Package the workflow into a simple interface, such as presets, node groups or a lightweight Gradio front end.
  • Provide on-the-ground technical support during the residency and document the system architecture and user-testing feedback for the final research report.

Requirements

  • Bachelor’s degree in Animation, Film Studies, Media Studies or a related discipline.
  • Hands-on experience with ComfyUI and local AI deployments, such as Automatic1111 or Forge, including LoRA and ControlNet training.
  • Understanding of GPU and VRAM optimization, and ability to assemble a simple working interface; formal UX/UI design experience is not required.
  • Ability to translate an artist’s creative intent into technical parameters and work patiently with a non-technical collaborator.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

ComfyUIStable DiffusionLoRAControlNetAutomatic1111ForgeGradioLocal AI deploymentGPU optimizationVRAM managementUser testingTechnical documentationTechnical support

Your skill match

Checking your profile…

YOUR NEXT STEP

Get interview-ready for this role

A focused preparation guide, built around this job’s responsibilities and requirements.

✦ AI-generated guide
Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.
01

Understand the role

What matters most

  • Run the workflow entirely locally

    Set up and optimize open-source Stable Diffusion and ComfyUI models on high-end workstation hardware, without cloud dependency. GPU and VRAM management are stated requirements.

  • Adapt models to the artist’s work

    Train local LoRAs and ControlNets using the curated archive of Professor Raimund Krumme’s analogue sketches and scans.

  • Translate creative intent into controls

    Work patiently with a non-technical artist to turn his ink-and-brush creative intent into technical parameters.

  • Make independent use practical

    Package the ComfyUI workflow so the artist can run it himself, using an approach such as presets, node groups or a lightweight Gradio interface.

  • Support and document the residency

    Provide on-the-ground support during the 14 September–13 November 2026 residency, and document system architecture and user-testing feedback for the final research report.

Likely challenges to think through

  • Performance without cloud fallback

    A likely trade-off is image quality or workflow complexity versus GPU memory and response time on one local workstation. Prepare to explain how you would profile VRAM use, test representative workflows and simplify settings when needed.

  • Preserving an analogue artistic voice

    Model adaptation may produce outputs that are technically convincing but miss the artist’s intended ink-and-brush character. Prepare an iterative way to review outputs with the artist and adjust training or controls against his feedback.

  • Simplicity without losing useful control

    Exposing every ComfyUI parameter could hinder independent use, while hiding too much could limit creative choices. Prepare to describe how you would test a small set of meaningful presets or controls with the artist and revise them from observed use.

03

Practice your answers

Technical practice

How would you set up Stable Diffusion and ComfyUI to run entirely on a local workstation?

What to coverThe role calls for open-source models on high-end hardware with no cloud dependency. Suggested preparation: Explain how you would verify that models, dependencies, inputs and outputs remain local, and how you would make the setup reproducible.

How would you structure a ComfyUI workflow for an artist who works with ink and brush?

What to coverThe posting asks for an AI workflow grounded in Professor Krumme’s analogue process. Suggested preparation: Walk through the input, generation and output stages, explaining which controls the artist would need and which could stay inside the node graph.

How would you prepare the curated sketches and scans for local training?

What to coverThe stated training source is an archive of analogue sketches and scans. Suggested preparation: Discuss inspecting scan quality, selecting and organizing examples, and checking that preparation preserves the visual characteristics the artist wants to explore.

When would you use a LoRA, and how would you assess its results?

What to coverTraining a local LoRA on the artist’s archive is a stated responsibility. Suggested preparation: Explain its role in adapting a base model, how you would compare outputs against the artist’s intent, and the trade-off between stronger adaptation and flexibility.

What would a ControlNet add to this workflow?

What to coverThe posting also calls for local ControlNet training. Suggested preparation: Explain how conditioning can help retain structure from sketches or scans, and how you would judge whether that control helps rather than constrains the artist’s work.

How would you investigate a ComfyUI workflow that runs out of VRAM or responds too slowly?

What to coverGPU and VRAM optimization are explicit requirements. Suggested preparation: Describe how you would isolate the costly stage and compare changes to model choice, resolution or workflow settings against output quality and usability.

How would you package the workflow for independent use by a non-technical artist?

What to coverThe posting suggests presets, node groups or a lightweight Gradio front end as possible approaches. Suggested preparation: Compare their setup and maintenance costs, and explain how you would reduce routine steps without hiding useful creative controls.

How would you test whether the assistant is useful to the artist?

What to coverThe role includes user testing and aims for a workflow the artist can operate independently. Suggested preparation: Describe observing an end-to-end task, recording points of confusion and comparing outputs with the artist’s stated creative intent.

What would you check if an established local workflow suddenly produced poor or inconsistent outputs?

What to coverOn-the-ground technical support is part of the role. Suggested preparation: Explain how you would distinguish changed inputs or settings from model, node or hardware problems, while keeping a known working configuration available.

What would you document for the final research report?

What to coverThe posting explicitly requires system architecture and user-testing feedback. Suggested preparation: Outline a concise record of components, model and workflow choices, local setup steps, observed feedback and changes made in response.

Behavioural practice

When have you translated a creative goal into technical choices?

What to coverThis role requires translating an artist’s intent into technical parameters. Suggested practice: Use a real situation, your questions and actions to clarify the goal, and the result—including how you checked whether the outcome matched the collaborator’s intent.

Tell me about working patiently with a non-technical collaborator.

What to coverThe assistant must be usable independently by an artist without a technical background. Suggested practice: Describe a real situation, how you adjusted your explanations or interface, and what the collaborator could do as a result.

Describe a time you resolved a technical problem while someone depended on the workflow.

What to coverThe posting includes on-the-ground support during the residency. Suggested practice: Give the situation, your diagnostic and communication steps, and the result, distinguishing an immediate workaround from any lasting fix.

When has user feedback changed a system you built or supported?

What to coverThe role calls for documenting user-testing feedback and delivering an easy-to-run workflow. Suggested practice: Explain a real observation, the change you made in response, and what the subsequent result showed.

02

Sharpen your knowledge

Topics to prepare

  • Local, privacy-safe workflow

    Trace how sketches, model files and outputs would remain on a workstation without cloud dependency. This reflects the posting’s requirement for a local, privacy-safe assistant.

  • LoRA and ControlNet roles

    Explain what each contributes to a Stable Diffusion workflow, then sketch when you would use one or both for ink-and-brush scans. The role explicitly includes training local LoRAs and ControlNets.

  • GPU and VRAM trade-offs

    Compare two plausible workflow configurations and identify where memory use could constrain training or generation. The posting requires understanding of local hardware optimization.

  • Creative intent as technical parameters

    Take an example artistic request, such as preserving the character of a brush line, and outline questions you would ask before changing workflow parameters. The role calls for translating a non-technical artist’s intent into technical choices.

  • Independent use and research feedback

    Outline a short artist-led test and the feedback you would record about ease of use. The posting asks for a workflow the artist can operate independently and for user-testing feedback in the final research report.

Skills to revise

  • Run a local ComfyUI workflow

    Practice setting up Stable Diffusion in ComfyUI, generating an output and restarting the workflow without cloud services. The posting requires hands-on ComfyUI and local AI deployment experience.

  • Train with sample scans

    Using scans you have permission to use, rehearse a small local LoRA training run and a ControlNet training run, recording inputs and results. This mirrors the stated work with a curated archive of the artist’s sketches and scans; the exercise does not assume access to that archive.

  • Diagnose workstation memory use

    On hardware available to you, observe GPU and VRAM use during a local run, adjust the workflow when memory is limiting, and note the effect. This practices the posting’s hardware-optimization requirement.

  • Package an easy-to-run interface

    Turn a ComfyUI workflow into a small set of presets or node groups, or a lightweight Gradio front end, and check whether another person can run it from simple instructions. These are interface options named in the posting, not separate requirements.

  • Support, test and document the system

    Role-play troubleshooting a failed run with a non-technical user, then write a brief architecture note and record their testing feedback. The role includes on-the-ground support and documentation for a final research report.

04

Your preparation checklist

  • Map your evidence to the requirements

    Prepare a brief record of your relevant degree or discipline and specific hands-on work with ComfyUI, local AI deployments, LoRA and ControlNet training. The posting names these as requirements; use only work you can substantiate.

  • Walk through a local deployment

    Prepare a project walkthrough showing how you set up Stable Diffusion and ComfyUI without cloud dependency, including GPU/VRAM constraints, optimization choices and what you tested. Use your own example rather than implying access to the artist’s archive.

  • Explain a sketch-training approach

    Outline how you would assess and prepare a curated set of analogue scans, train local LoRAs and ControlNets, and compare results against the artist’s intent. Present this as a proposed approach, not an existing project outcome.

  • Show an artist-friendly workflow

    Prepare a small ComfyUI example with presets or node groups, or a lightweight interface sketch, that a non-technical artist could run independently. Be ready to explain what you would simplify after user testing.

  • Practice translating creative feedback

    Prepare a short scenario in which an artist describes an ink-and-brush result they want. Show how you would clarify that intent, adjust technical parameters and explain the change without jargon.

  • Plan support and documentation

    Draft a one-page outline for recording system architecture, troubleshooting notes and user-testing feedback for the final research report. Explain how you would provide on-the-ground support during the stated 14 September–13 November 2026 residency.

  • Bring focused questions for ADM

    Ask about the workstation specifications, access to and permitted use of the curated scans, how Professor Krumme prefers to work, and what would make the workflow independently usable. These are points to clarify, not details supplied by the posting.

AI-generated from the job information · Prepared Sep 30, 2026 (UTC)Useful preparation, not a guarantee of eligibility or an offer.

Availability last checked: 4 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in Singapore. Matched to this role, not your profile.

Explore more jobs
Finding similar opportunities…