Digital Casting with ComfyUI at Tertulias IA – Universidad Mayor

On March 27, Yutro participated in Tertulias IA, a conversation and exchange event organized by Universidad Mayorfocused on the current state of creative artificial intelligence in Chile. A 7-hour day with 12 panelists and over 2,400 LinkedIn impressions — a clear sign that the industry's conversation around AI is far from exhausted.

The event

Tertulias IA brought together professionals from different disciplines — design, advertising, audiovisual production, education, and technology — to share real experiences with generative AI tools. It wasn't a conference with experts speaking from a stage: it was a horizontal conversation where each panelist shared their concrete practice, their mistakes, and their discoveries.

The format worked. The room stayed active throughout the entire day, with questions overflowing the allotted time and conversations continuing in the hallways.

What we presented: digital casting with ComfyUI

From Yutro, we presented our workflow for digital character casting using ComfyUI. The problem we solve is one of the most frustrating in AI generative production: how do you keep the same character consistent across multiple images, angles, and situations?

The short answer is that no single model does this well by default. The long answer is the workflow we showed: a combination of IP-Adapter, ControlNet, and structured prompting to anchor visual identity without sacrificing compositional variety.

“Digital casting doesn't replace real casting. What it does is open a pre-production stage that didn't exist before: you can explore 40 character profiles in an afternoon and arrive at the client meeting with concrete options, not just ideas.”— Milivoy Yutronic, Yutro

The three stages of the workflow

The workflow we presented has three distinct stages:

1. Identity definition: we work with real or generated reference images to establish the features that must be maintained. This includes facial structure, skin colorimetry, hair type, and base expression.
2. Anchored generation: using IP-Adapter in ComfyUI, that identity is connected to new generations. The character can change outfit, background, lighting, and pose while maintaining recognizable coherence.
3. Consistency validation: a battery of tests is generated — different angles, contexts, and lighting variations — to evaluate how well the identity holds up before taking the character into production.

What we learned listening to other panelists

Beyond presenting, we listened. And that was equally valuable. Some themes that resonated throughout the day:

• The gap between what agencies promise clients with AI and what they can actually deliver remains large.
• The most effective workflows aren't the most sophisticated ones — they're the ones the team understands and can reproduce.
• The question of authorship and identity of generated characters is becoming urgent for brands.
• ComfyUI remains the most powerful tool for professional work, but the entry curve pushes many creative teams away.

Event numbers

Why these spaces matter

Generative AI moves fast, and most of that movement happens in scattered online communities. Events like Tertulias IA do something no YouTube tutorial can: they put people in the same room and generate real conversation about real practice.

For Yutro, participating in these spaces is part of how we build knowledge. We don't just show what we do — we listen to how others are solving the same problems we face. That always comes back in the form of better projects.