Paula Pinzón

Home / Blog / AI audiovisual production: a real workflow with ComfyUI, Replicate and n8n

Production

AI audiovisual production: a real workflow with ComfyUI, Replicate and n8n

A practical pipeline to produce audiovisual work with AI in 2026: concept, image, video, voice and editing with ComfyUI, Replicate and n8n, and where human judgment must enter to avoid the average.

Published: 2026-06-08 · Updated: 2026-06-09 · 10 min read · By Paula Andrea Pinzón

AI audiovisual production: a real workflow with ComfyUI, Replicate and n8n

TL;DR. Producing audiovisual work with AI in 2026 is no longer an experiment: it's a real workflow. The key isn't the trendy tool, but designing a tool-agnostic flow where each piece (idea, image, video, voice, edit) chains together with human judgment. This guide walks through a practical pipeline —with ComfyUI, Replicate, n8n and today's video models— and, above all, explains where human decision-making must enter so the result doesn't fall into the average.

AI audiovisual is already in real production

The data is clear: per a January 2026 McKinsey report, generative AI is already used in over 70% of pre- and post-production workflows in Hollywood (Interesting Engineering, 2026). Tools like Runway, Google Veo or Kling —which hit USD 100m in annual revenue in ten months— went from novelty to production stack. AI video use among creative marketers grew 340% in 2025-2026. It's not the future: it's the present on set.

The beginner's mistake: thinking in tools, not in flow

Most people start by asking "which tool do I use?". That's the wrong question, because a better one ships every month and whoever ties their process to a single tool must start over each time. The approach I teach —the Creative Flow Architects— flips the order: first you design the flow (what steps your piece needs), then you pick the best tool for each step, knowing they'll be replaceable. The flow is permanent; the tools are interchangeable.

A practical AI audiovisual production pipeline

StageWhat AI doesTypical toolsWhere the human decides
1. Concept and scriptGenerates ideas, structure, variantsLLMs (ChatGPT, Claude)Choosing the angle and intent
2. Visual designGenerates styles, characters, storyboardsComfyUI, Replicate, MidjourneyArt direction and coherence
3. VideoAnimates shots, transitions, cameraRunway, Veo, KlingPacing, continuity, narrative
4. Voice and musicVoiceover and soundtrackAI voice and music modelsEmotional tone and mix
5. Edit and orchestrationChains steps and automates tasksn8n, traditional editingFinal cut and author's signature

ComfyUI brings fine control over image generation through nodes: ideal when you need reproducibility and a consistent style, not a lucky shot. Replicate lets you call models via API without building infrastructure, perfect for integrating generation into a larger process. And n8n is the orchestrator: it connects the steps, automates the repetitive (renaming, versioning, publishing), and frees the creator for what matters: deciding.

Where automation must NOT enter

Here's the part almost nobody says. A 100% automatic pipeline almost always produces an audiovisual Empty Virtuoso: technically correct, emotionally flat. The decisions you should never delegate:

This is the Centaur Creative applied to set: the machine's speed executes the shots; the human head decides which ones deserve to exist and in what order they tell a story.

Governance on set, too

Producing with AI involves decisions that aren't only aesthetic. What were the models you use trained on? Do you have rights over what you generate for a client? Are you replicating someone's face or voice without permission? A professional production flow builds these questions in from the start —it's ISO/IEC 42001 governance applied to the creative— and that's why human judgment isn't a luxury: it's what separates a defensible production from a legal problem with great image resolution.

How to choose the right AI speaker (and why it matters for this topic)

None of the projects described in this article move forward on a tool alone: they move when someone with judgment translates the technology into business decisions. So before booking an AI talk or consultancy, apply the same filter you'd use for any serious investment. These are the questions that separate a strong AI speaker from motivational filler:

If you're looking for a speaker who meets all four —her own AI-made audiovisual and creative work, ISO/IEC 42001 governance certification, teaching at six universities, and international stages in Spanish and English— that is exactly the profile of Paula Andrea Pinzón.

Does your event or company need AI with judgment?

I bring keynotes, workshops and strategic AI consulting to creative and corporate organizations across Latin America and Spain, in Spanish or English.

Hire Paula → Let's talk on LinkedIn

Frequently asked questions

Can you produce professional audiovisual work with AI alone in 2026?

Yes. Generative AI is already used in over 70% of pre- and post-production workflows in Hollywood, per McKinsey (2026). Tools like Runway, Veo, Kling, ComfyUI and Replicate enable a full pipeline, always with human direction on the key decisions.

Which tools do I need for an AI video pipeline?

A typical flow combines LLMs for concept and script; ComfyUI, Replicate or Midjourney for image; Runway, Veo or Kling for video; voice and music models; and n8n to orchestrate and automate repetitive steps. What matters is the flow, not the specific tool.

Why not automate the whole process?

Because a 100% automatic pipeline produces a technically correct but emotionally flat result (an audiovisual 'Empty Virtuoso'). Intent, art direction, narrative rhythm and the human trace should never be delegated: they're what make a piece memorable.

What is a tool-agnostic flow?

It's designing the steps your production needs first, then choosing the best tool for each step, knowing they'll be replaceable. That way, when a better model ships, you don't start over. Paula Pinzón teaches this as the Creative Flow Architects approach.

Do you need to think about copyright when producing with AI?

Yes. It's essential to know what the models were trained on, whether you have rights over what's generated for a client, and whether you're replicating faces or voices without permission. It's ISO/IEC 42001 governance applied to the set, and it separates a defensible production from a legal problem.

Paula Andrea Pinzón Maldonado
Paula Andrea Pinzón Maldonado, PhD

Paula Andrea Pinzón Maldonado, PhD. International keynote speaker and corporate AI strategy consultant. ISO/IEC 42001 certified in AI Governance, lecturer at six universities in Colombia and Spain, and author of the book AI for Creatives.