New opportunity

Generative Workflow Developer

Xanadu Studio · Mumbai, India

About this role

Xanadu Studio is seeking a Generative Workflow Developer in Mumbai to build and experiment with AI image and video workflows for filmmaking, advertising and digital content. The full-time role focuses on turning experiments with open-source generative models into reliable production workflows.

Responsibilities

Build and improve ComfyUI image and video workflows and create agentic workflows with Griptape.
Test and integrate new models, including MiniMax H3 and LTX 2.5; write Python tooling and custom nodes when needed.
Develop workflows using LoRAs, ControlNets, reference conditioning, relighting, upscaling, video-to-video and emerging techniques. Evaluate new models to determine their production value.

Requirements

A background in technical development, VFX/compositing, technical art, generative AI or a similar technical-creative field.
Comfort with Python, Git/GitHub, open-source AI tools and models, basic Linux and command-line workflows, generative image and video models, and technical troubleshooting.
Hands-on experience with at least one of Nuke, ComfyUI or Griptape is mandatory. Candidates need not know all three, but should be able to learn the others.

Preferred experience includes PyTorch, custom ComfyUI nodes, LoRA or ControlNet training, Hugging Face, Docker, CUDA, cloud GPUs, MiniMax H3, LTX 2.5, FLUX, Wan, Qwen Image, Nuke or other VFX tools. Applicants should submit a CV and examples of work they have built, such as GitHub projects, workflows, Nuke work or generative experiments. The studio emphasizes demonstrated building ability and learning speed over formal qualifications.

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

Skills for this role

ComfyUIGriptapeNukePythonGitGitHubLinuxCommand lineGenerative image modelsGenerative video modelsAI agentsLoRAControlNetReference conditioningRelightingUpscalingVideo-to-videoVFX compositingTechnical artTechnical troubleshootingPyTorchHugging FaceDockerCUDACloud GPUsMiniMax H3LTX 2.5FLUXWanQwen Image

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Make image and video workflows production-ready
The role explicitly calls for building and improving ComfyUI workflows and turning experiments into reliable production workflows. Prepare a built example that shows what you changed between an initial experiment and a more dependable workflow.
Build agentic workflows with Griptape
Griptape workflow development is a stated duty. Prepare to explain the steps and decisions in a workflow you have built, using relevant work if you have it; experience with all three of Griptape, ComfyUI and Nuke is not required.
Evaluate and integrate emerging models
The studio names MiniMax H3 and LTX 2.5 and asks for quick judgments about new models' practical usefulness. Prepare to describe how you would compare model outputs and integration effort for a filmmaking, advertising or digital-content use case.
Extend and troubleshoot technical-creative pipelines
Python tooling, custom nodes where needed, and workflows involving LoRAs, ControlNets, reference conditioning, relighting, upscaling and video-to-video are part of the role. Prepare a concrete example of diagnosing a workflow problem or adding a capability, rather than listing tools.

Likely challenges to think through

Balancing novelty against repeatability
Preparation hypothesis: a promising new model may produce compelling results but be difficult to use consistently in production. Prepare to explain what evidence would justify adopting it, what failures you would check for and when you would retain an existing workflow.
Preserving creative control across multiple stages
Preparation hypothesis: combining reference conditioning, relighting, upscaling or video-to-video could improve an output while introducing unwanted changes elsewhere. Prepare to describe how you would isolate stages, compare outputs and identify which step caused a regression.
Learning across tools without assuming identical workflows
Preparation hypothesis: work spanning ComfyUI, Griptape and Nuke may require transferring ideas between different kinds of tools. The posting requires hands-on experience with at least one, not all three; prepare to explain how you would learn an unfamiliar tool while keeping an existing workflow usable.

Practice your answers

Technical practice

How would you structure a ComfyUI workflow for generating and revising images or video?

What to coverThe role involves building and improving ComfyUI workflows for both media types. Preparation suggestion: explain how you would organize inputs, model stages and outputs so changes can be tested without losing track of what produced a result.

What would you put in a Griptape agentic workflow, and how would you check its output?

What to coverThe posting names building agentic workflows with Griptape as a duty. Preparation suggestion: describe a bounded task, the steps the agent would perform and how you would detect or handle an unusable result.

How would you decide whether a newly released image or video model has production value?

What to coverThe role includes testing and integrating new models, including MiniMax H3 and LTX 2.5. Preparation suggestion: compare outputs on representative tasks and discuss quality, consistency, runtime and integration effort rather than novelty alone.

When would you write Python tooling or a custom ComfyUI node instead of using existing components?

What to coverThe posting calls for Python tooling and custom nodes where required. Preparation suggestion: identify a concrete workflow gap, explain the smallest useful implementation and describe how you would test that it behaves reliably.

How would you choose among a LoRA, ControlNet and reference conditioning for a visual task?

What to coverAll three techniques appear in the role's workflow duties. Preparation suggestion: distinguish adapting a model from guiding structure or using a visual reference, then discuss the control each offers and where results may fall short.

How would you assess a workflow that combines relighting and upscaling?

What to coverThe posting explicitly lists relighting and upscaling. Preparation suggestion: explain how you would inspect lighting coherence and recovered detail, including cases where an apparent improvement introduces artifacts.

What would you check when a video-to-video result looks good in individual frames but inconsistent in motion?

What to coverVideo-to-video is a stated workflow area. Preparation suggestion: discuss temporal consistency, subject continuity and comparisons across frames, then identify which workflow settings or stages you would isolate first.

How would you turn a successful generative experiment into a repeatable production workflow?

What to coverTurning experiments into reliable production workflows is an explicit responsibility, and Git/GitHub comfort is requested. Preparation suggestion: cover recorded inputs and versions, repeatable execution, failure checks and how changes would be tracked.

How would you troubleshoot an image or video workflow that fails after a model change?

What to coverThe posting asks for technical troubleshooting, basic Linux and command-line comfort, and work with open-source models. Preparation suggestion: explain how you would inspect errors, verify model and dependency compatibility, and narrow the failure to a specific stage.

How could Nuke or compositing work fit alongside generative outputs?

What to coverNuke is named as a relevant tool, but hands-on experience is required in only one of Nuke, ComfyUI or Griptape. Preparation suggestion: if relevant to your experience, discuss how you would inspect and refine generated material for a filmmaking or advertising deliverable without assuming the posting prescribes a specific handoff.

Behavioural practice

Tell me about an experiment you made reliable enough for repeated use.

What to coverThe role emphasizes moving generative experiments into production workflows. Preparation suggestion: use a real situation, the changes you made to improve repeatability and the observed result; do not present an untested experiment as production-ready.

Describe a difficult technical workflow failure you diagnosed.

What to coverTechnical troubleshooting is explicitly requested. Preparation suggestion: state the symptoms and stakes, the checks you performed in order, what you changed and how you confirmed the result.

Give an example of learning a new tool or model quickly to complete a build.

What to coverThe posting values learning speed and says experience with one of Nuke, ComfyUI or Griptape can be paired with learning the others. Preparation suggestion: describe the initial knowledge gap, how you tested what mattered and what you ultimately built or learned.

Tell me about a time you balanced visual quality with technical constraints.

What to coverThe work combines generative tools with filmmaking, advertising and digital content, making this an inferred preparation topic rather than a stated interview requirement. Preparation suggestion: give a real situation, the trade-off you made, the action you took and the resulting output or feedback.

Sharpen your knowledge

Topics to prepare

Production value of new models
The role involves testing open-source image and video models, including MiniMax H3 and LTX 2.5, and deciding what is useful in production. As a revision exercise, define a comparison checklist for output quality, consistency, runtime and workflow fit, then explain how you would make an integration decision.
Image and video conditioning
The posting names LoRAs, ControlNets, reference conditioning, relighting, upscaling and video-to-video. Revise what each technique controls; for one filmmaking or advertising shot, sketch which you would use and why.
Agentic workflow design
Building agentic workflows with Griptape is a stated responsibility. Sketch an agentic workflow with clear inputs, steps, outputs and failure points, and explain which steps need review before an output is used in production.
Experiment-to-production reliability
The studio wants experiments turned into reliable production workflows. As preparation, compare an exploratory image or video workflow with a repeatable version: identify inputs to record, likely failure points and checks on the output.
Technical-creative handoff
The role spans generative workflows and filmmaking, advertising and digital content; Nuke and VFX/compositing are also named. As a preparation suggestion, describe how you would evaluate whether a generated image or clip is usable for a compositing task, distinguishing technical faults from visual ones.

Skills to revise

Build a ComfyUI workflow
Building and improving ComfyUI image and video workflows is a stated responsibility. Practice constructing one workflow, changing a conditioning step, and explaining the effect of each change. The posting requires hands-on experience with at least one of ComfyUI, Griptape or Nuke—not all three.
Prototype in Griptape
Building agentic workflows with Griptape is part of the role. If Griptape is a tool you use, practice a small workflow and trace its inputs, steps and outputs; otherwise, treat this as preparation for learning a named tool, not an additional experience requirement.
Write and debug Python tooling
Python comfort is requested, and the role may involve tooling or custom nodes. Practice a small Python utility that validates workflow inputs or outputs, then diagnose a deliberately broken case. Custom ComfyUI node experience is listed as a bonus.
Troubleshoot with Linux, Git and GitHub
The posting requests comfort with basic Linux, command-line workflows, Git/GitHub and technical troubleshooting. Practice reproducing a workflow error from the command line, recording the fix in Git, and writing concise steps that someone else could follow.
Demonstrate work in a named tool
Applicants are asked for examples of things they have built, and hands-on experience with at least one of Nuke, ComfyUI or Griptape is mandatory. Select a relevant existing project or practice artifact in one of those tools and rehearse a short explanation of what you built, the problem encountered and how you tested the result.

Your preparation checklist

Select a hands-on project
Choose a project demonstrating real use of at least one of Nuke, ComfyUI or Griptape; the posting requires hands-on experience with one, not all three. Prepare a short walkthrough of what you built and your contribution.
Package evidence of your work
Prepare a CV and relevant examples, such as a GitHub project, workflow, Nuke work or generative experiment, as the posting requests. Make the inputs, outputs and your role in each example clear.
Rehearse an image or video workflow walkthrough
If your work includes one, map its stages from input to output and explain choices such as reference conditioning, LoRAs, ControlNets or video-to-video. These are techniques named in the role, not requirements to demonstrate every one.
Prepare a model-evaluation exercise
As interview practice, outline how you would test a new image or video model on a filmmaking or advertising task, compare results and decide whether it is useful in production. State what you would measure rather than claiming tests you have not run.
Document a troubleshooting example
Bring one concrete account of a workflow failure, how you diagnosed it and what changed. Be ready to show relevant Python, Git and basic Linux or command-line steps if they were part of your work; the posting calls for comfort with these tools.
Explain a path from experiment to production
Sketch a small design for making a generative workflow repeatable, including where Python tooling or a custom node might help. This is a preparation exercise based on the role’s focus on reliable production workflows, not a stated interview assignment.
Write focused questions for Xanadu Studio
Ask how the team judges a model’s production value, where ComfyUI, Griptape and Nuke fit in its workflows, and what reliability problems this role would tackle first.

Availability last checked: 5 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 India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…