AI PRODUCTION
WORKFLOWS
This page documents how the production system operates: control layers, QC, revisions, asset management and delivery. For engagement scope, timing and starting budgets, see Services.
01 · Why AI stallsTeams adopt AI expecting speed. They hit inconsistency, revision hell, and asset chaos instead.
Inconsistent output
Twenty generations, twenty different brands. The product, the model, the lighting. Nothing stays locked across a campaign.
Video falls apart
Flicker, drift, morphing faces. The moment frames become a sequence, temporal instability destroys the shot.
Revision hell
A client asks to move one object. The only option is regenerate everything and hope it matches. It won't.
Asset chaos
Thousands of generated images. No tags. No search. No way to find the good ones without scrolling for hours.
No QC owner
Everyone generates, nobody reviews. The final output is whatever the model happened to give you. No quality gate exists.
Pipelines break
The workflow crumbles when the team changes, the model updates, or the project scales. It was never designed to survive.
Not a prompt. Not a clever ComfyUI node graph. The complete system that turns unstable generation into reliable commercial output, the same way a traditional pipeline turns raw footage into a finished film.
| Component | What it solves |
|---|---|
| Model selection & testing | Which tool fits this job, not which one is trending this week. |
| Reference & control systems | Consistent style, character, and product across hundreds of generations. |
| Generation pipeline | Controlled batch output instead of one-at-a-time experimentation. |
| QC checkpoints | Human review at defined stages, before anything reaches the client. |
| Asset management & tagging | Finding the right frame out of thousands using AI-powered search. |
| Post-production integration | VFX, compositing, colour, and retouching that turn raw output into finished content. |
| Revision protocol | Targeted changes. "lengthen the sleeves" doesn't mean "regenerate everything." |
| Delivery & documentation | Client-ready files with consistent naming, formats, and usage tracking. |
A workflow gets you a production.
Audit the current workflow
I map your existing creative production. What works, what's expensive, what's inconsistent, where AI could genuinely help, and where it would quietly make things worse.
Decide where AI fits, and where it doesn't
Not everything should be generated. Some shots need a real camera. Some need 3D. Some need compositing. I identify which parts can be automated and which still need human craft.
Design the production pipeline
End-to-end system: model selection, reference architecture, generation pipeline, QC checkpoints, asset management, post-production handoff, revision protocol, delivery standards.
Build and test on real content
I test the pipeline on real content, find the failure points under production conditions and fix them before handoff.
Hand over a system the team can run
The deliverable is a documented, repeatable production system with templates, test files, operating notes and a clear owner for quality control.
A national UK commercial, 95% AI-generated
When Gigaclear needed a national broadcast commercial with AI-generated environments, a virtual crowd, and seamless integration with live-action footage, the work ran through a full production workflow, not a prompt window. A custom crowd pipeline turned AI character images into 3D assets and auto-placed 100 of them inside a Cinema 4D scene in a single day. The Giganaut hero character was built as a rigged 3D model because raw generation couldn't hold its form across shots. Of the thousands of images generated, nearly two-thirds were near-duplicates, the visible trace of a real creative process: generate, compare, reject, regenerate. The result: a national broadcast commercial without location scouting, travel, set construction, or crowd casting, a fraction of the traditional cost, none of the compromise.
The lesson that defines everything on this page: AI alone wasn't enough. The workflow around it—the 3D control layer, compositing, QC and asset intelligence—is what turned generation into a finished national campaign.
Read the full case study →Production Map
A shot-by-shot plan showing where AI, 3D, live action, VFX and post each belong.
Control System
Approved references, model choices and locked inputs for character, product, style and camera consistency.
Tested Pipeline
A workflow proven on real project content, including known limits and recovery paths.
QC & Revision Protocol
Named review gates and targeted revision rules that protect approved elements from unnecessary regeneration.
Asset Library
Searchable naming, versions, selects and usage records so the team can find and reuse approved material.
Documentation & Handoff
Templates, operating notes and training for an internal team—or a clean specification for managed production.
Built from VFX, not just AI
My background is in traditional post-production: compositing, 3D, colour and editing. That finishing layer is what turns AI output into commercial content. AI is the tool, not the identity.
I build tools, not just use them
When off-the-shelf workflows are not enough, I write custom ComfyUI nodes, build automation scripts and design pipelines around the production problem.
I flag when AI is the wrong answer
Sometimes a traditional shoot is faster, cheaper and better. I will say so. The job is to solve the production problem, not sell AI into it.
It changed where the work happens.
Insights from Production
Articles about what AI production actually looks like, written from experience on commercial projects.
The One Small Change
AI can generate a commercial in seconds. The trouble begins when someone asks it to generate the same commercial twice.
Read → // ArticleWhat AI Production Actually Looks Like
Thousands of images, a month of work, and why 95% AI doesn't mean 95% less effort. An honest account from the Gigaclear commercial.
Read → // ArticleThe 7-Stage AI-VFX Pipeline
From concept to final frame, a detailed breakdown of every stage in the pipeline and why generation is only one of seven.
Read → // ArticleWhy AI Video Still Needs VFX
Compositing, colour grading, sound design, the craft that separates AI output from broadcast quality.
Read → // ArticleHow to Choose the Right AI Model
When to use FLUX vs GPT Image vs Kling vs Cines vs LTX, a practical guide based on production experience.
Read → // ArticleWhy Node-Based Control Beats Prompting
Cost, quality, timeline, an honest comparison and a decision framework for when each approach makes sense.
Read → // ArticleBuilding Automated UGC Systems
Lock the character, the location, or the style, then batch-generate hundreds of on-brand videos a day, every batch checked before it ships.
Read → // ArticleThe Real Cost of AI Production
What $4K–$40K actually buys. A transparent breakdown of AI production pricing, tier by tier.
Read →More articles coming as production experience grows.
You have a production problem.
AI might be part of the solution. It might not. The fastest way to find out is an audit. Tell me the goal and how you measure success. I will map the workflow, show what is breaking and explain what to automate, what to keep traditional and what production-ready delivery requires. Agencies are welcome; white-label work is available.
See the Production AuditNeed the system itself built? ComfyUI consulting & pipeline development →
Mainly taking on projects. Open to embedded, day-rate, or full-time roles for the right production — resume here.