Custom ComfyUI workflows
built for real production.
I design, build, fix, and hand off the AI pipelines behind commercial work. For agencies, brands, and startups that need generated content to be consistent, repeatable, and revision-proof. Not a demo that worked once. A system your production can rely on.
In one paragraph: A ComfyUI consultant turns AI models into production systems. I build custom graphs and batch automation, audit and repair workflows that keep breaking, and hand everything over documented so your own team can run it. The same pipeline discipline delivered a national UK TV commercial that was roughly 95% AI-generated, and a full film from idea to final edit in under 12 hours on local hardware.
01 · The ProblemWhy AI workflows break in production
Most teams get to a good-looking result once. Then the client asks for the same character in a different shot, or version two of the campaign, or four thousand product variations, and the workflow falls apart. The generation was never the hard part. The hard part is control: consistency across shots, quality gates, file discipline, and a pipeline that survives revisions.
That gap between "we made a cool test" and "we can deliver this every week" is exactly what I get hired to close. Usually there is no off-the-shelf tool for the problem, which is why the job is building the tool.
02 · What I BuildFour ways to engage
Custom Workflow Development
ComfyUI graphs designed for your specific production need. Image, video, character consistency, product accuracy. Built, tested on real briefs, and tuned until output survives client notes.
Automation & Batch Systems
When the ask is a thousand videos a week or four thousand images in a batch, hand-driving a graph stops working. I build scripted, API-connected systems that generate, QC, and organize at volume.
Workflow Audit & Repair
Already have a pipeline that misbehaves? I review the graphs, models, and process, find where consistency or quality actually breaks, and fix or rebuild the failing stages. Usually control and QC, rarely the model.
Training & Handoff
Documented graphs, reusable templates, named settings, and training for the people who will run it. Your team operates the system without me, or I stay on for testing, maintenance, and the first production run.
03 · Who This Is ForWho hires me
Agencies and production studios adding AI capability to client work without betting a delivery date on an experiment. Brands with content volume: product imagery, UGC-style video, campaign variations that need to be repeatable, on-brand, and cheap per unit. Startups and tech companies building generative pipelines behind their own products, who need production-grade reliability rather than a research demo.
04 · ProofPipelines that shipped
A national UK broadcast commercial. Roughly 95% of the visual world of the Gigaclear TVC was AI-generated. Rural locations, environments, a 100-character crowd. Through a ComfyUI-centered pipeline with VFX finishing and real actors composited in.Read the full case study →
A film in under 12 hours, on local hardware. The NEON EVM piece went from idea to final edit in a single day, generated almost entirely with Wan 2.2 running locally. Proof of what a tuned local pipeline does for speed and cost.See the project →
Automated volume systems for e-commerce. Under NDA, I have built pipelines for e-commerce brands where every SKU needs its own set of images. Hundreds of on-brand shots per product, across catalogs of different SKUs. Product data goes in one end; generated, QC-checked, correctly named files come out the other. The same discipline runs UGC-style video at volume, the "thousand videos, one week" problem. How those systems work →
05 · Under the HoodTechnical capabilities
For the technical directors evaluating depth. This is the toolbox. If this section isn't for you, skip to how an engagement runs.
- Custom ComfyUI graphs and reusable production templates; custom nodes and Python scripting where the graph alone is not enough
- Local open-model stacks: Wan 2.2, LTX, Flux, Qwen, Stable Diffusion. For privacy, cost-at-volume, and control; cloud models (Seedance, Kling, Veo) where they earn their place
- Consistency systems: ControlNet and reference pipelines, LoRA training for characters and styles, locked looks that hold across shots and batches
- Batch generation and API integration. Queue management, variable-driven prompting from data, automated QC and file organization
- Model selection and testing as a production method: budget, control, privacy, and delivery requirements decide, not leaderboards
- VFX-grade finishing downstream: compositing, cleanup, tracking, grade, the seam between generated and shot material is where quality lives
06 · The ProcessHow an engagement runs
It starts with a short call or a rough brief: what you are producing, at what volume, and what is failing today. I tell you what is feasible, what needs testing, and, honestly. Where AI is not the right tool. Then the pipeline gets designed, built, and proven on a real brief before it is called done. Projects start from $5k; an audit is the usual first step for teams with an existing setup.
07 · FAQCommon questions
What does a ComfyUI consultant actually do?
Designs and builds the node graphs, automation, and quality-control systems that turn AI models into a repeatable production pipeline. Custom workflows, fixes for existing ones, batch automation, and documented handoff so your team can run it without me.
Who hires a ComfyUI consultant?
Agencies and studios adding AI to client work, brands that need content at volume, and startups building generative pipelines behind their products. The common thread is needing output that is consistent and repeatable, not one lucky generation.
Can you fix our existing workflow instead of building a new one?
Yes, an audit is often the right first step. I review your graphs, models, and process, find where consistency or quality breaks, and fix or rebuild the failing part. Most broken AI workflows fail at control and QC, not at the model.
Do you work with local and open-source models?
Yes, that is a specialty. I run Wan 2.2 and LTX locally on my own hardware, which matters for privacy requirements and for volume where per-generation API costs do not scale. A recent film went from idea to final edit in under 12 hours on a local Wan 2.2 pipeline.
Will our team be able to run the workflow after you leave?
That is the point of the handoff: documented graphs, reusable templates, named settings, and training for the operators. I can also stay involved for testing, maintenance, or the first production run.
How does an engagement start?
A short call or a rough brief. What you are making, the volume, what is failing today. I respond with what is feasible, what needs testing, and where AI is not the right method. Projects start from $5k.
Have a pipeline problem
with no tool for it yet?
Send the rough version, a brief, a broken workflow, or just the volume you need to hit. I'll tell you what's feasible, what needs testing, and what it would take.
Mainly taking on projects. Open to embedded, day-rate, or full-time roles for the right production — resume here.