Behind the Scenes of Finding Weegio’s Voice for Wego at Pressure Cooker Studios
We’ve been dodging the AI conversation, not because we didn’t want to join in, but because we were waiting for the right project to really test its limits. That moment finally arrived, and there was no ignoring it.
So here it is: our first official blog on using AI in the studio, told by the people who actually used it. Because every creative has their own take, and there’s no one-size-fits-all approach to AI in the creative process.
Spoiler alert: it’s not about prompt-engineering our way to perfection.
When it came to designing the voice of Weegio, a small, lovable alien, we quickly realised no algorithm could replace good old-fashioned creativity, teamwork, and… blowing a kazoo into slime.
This wasn’t just about generating a voice with AI, it was about pushing the limits of creativity to craft a character that felt truly authentic in its own right. AI was part of the toolkit, but it was the strange experiments, the late-night creative sparks, and collaboration that brought Weegio to life.
We didn’t use AI to replace creativity, we used it to amplify it.
What was the original brief for Weegio’s voice, and what initially excited you about it from a creative perspective? How closely did you stick to that brief as the project evolved?
Denys said the project was unique because the main objective was to combine traditional workflows with AI, which meant figuring out how to create a voice that wasn’t human. The team had to sit down and brainstorm entirely new approaches, making the process feel open, experimental, and exciting from the start.
Alex added that the brief pushed things even further by asking them to invent an alien language using AI, something consistent enough to function like a language, but still full of personality and emotion. That balance between structure and expression was the real creative hook. As the project evolved and Weegio’s dialogue shifted into English and Arabic, it grounded the character and made his humour clearer to audiences. But throughout, the guiding challenge remained the same: pushing AI beyond the obvious to find a truly original voice.
How did you approach using AI in this project and what made the process feel different from a typical audio job?
Denys explained that the team tried just about every AI tool they could find, from sound effects and voice generators to text-to-speech and voice-to-voice converters. They even experimented with building an entirely new language, complete with grammar, which they then fed into a generator. The whole process was unlike anything they’d done before, but the results from AI alone weren’t giving them what they needed.
Meanwhile, Alex pointed out that generating a phonetic language for a voice model seemed like the obvious route, but it sounded flat and generic, with no real control over tone or emotion. Pushing further, they experimented with feeding non-human sounds into the models, which produced wonderfully strange, alien textures. And when the brief later shifted to English and Arabic, they built a new workflow, layering multiple voice models in sequence to create Weegio’s final voice.
What were some of the weirdest or most unexpected things you tried while chasing the right sound?
Denys recalled how text-to-speech tools never delivered the alien qualities they were after, and sound effect generators proved too inconsistent. So the team began feeding non-human sounds into a voice generator, which transformed quick, sharp noises into syllable-like “sentences”. Trying to mimic these themselves only resulted in baby talk, not the energetic, sassy character they needed. The challenge became translating full English lines into this gibberish while still shaping pitch, inflection, and delivery.
At the same time, Alex described how they pushed the experiments further, throwing everything from squeaky toys and alpaca cries to whale calls and creaky doors into the AI. The strangest, and most effective test, involved blowing bubbles in yoghurt through a kazoo. Layering pitched voices with bubbly textures gave Weegio a quirky, otherworldly quality they loved so much that it remained part of his alien identity, even once the dialogue shifted into English and Arabic.
Were there moments where it felt like it wasn’t working? What helped you break through?
Denys explained that short, fast sounds produced the most convincing AI “words”, but early bubble experiments with yoghurt and water were messy and inconsistent. Slime worked far better, and when paired with a kazoo it created a unique mix of raspy, creature-like tones with airy, bubbly textures. Performing lines through this setup, while exaggerating pitch shifts, finally gave the team material the AI could shape into something compelling.
For Alex, the biggest challenge was getting the AI to deliver lines with a consistent tone. Working with multiple voice models meant generating countless takes before landing on the right one, a frustrating process, but one that ultimately pushed the team to streamline their workflow and find creative ways around the AI’s built-in habits.
Everyone’s talking about AI, but in your view, what can it not do?
Denys noted that working closely with AI quickly revealed its limits, humbling, but useful. The tools often defaulted to the average, and the most exciting results came from their mistakes rather than polished outputs. AI opened doors to ideas he couldn’t have reached otherwise, but every result still needed human refinement, editing, shaping, and countless small creative choices. For him, it always came back to this: a human idea, transformed by AI, then made meaningful again by human hands.
Alex agreed the biggest challenge was guiding AI itself. Voice models, trained in narrow ways, often required endless takes just to get the right tone or pronunciation, sometimes slower than simply recording a human. But where AI excelled was in the non-human space: by pushing models beyond their comfort zones, they found the alien qualities that made Weegio unique, something no human performance could fully replicate.