This is where I document what’s working, what’s breaking, and what I’m learning about building with AI tools.
I’ve been a filmmaker for 15 years. For the last three, I’ve been stress-testing generative AI to see if it can hold up to professional storytelling standards. Sometimes it does. Often it doesn’t. Either way, I’m documenting it here.
This isn’t a portfolio. It’s a workshop. You’ll see experiments that succeeded, workflows that failed, and questions I’m still trying to answer.
If you’re also trying to figure out how to preserve human voice while working with these tools — welcome. Let’s figure it out together.
— Bruno
AI expands what you can make.
It does not expand who you are.
That’s the artist’s job.
Kaera Varan: What If Music Was in a Language You Couldn’t Understand?
The Concept:
I created a band called Kaera Varan that sings entirely in Lumirian — a language that doesn’t exist outside of this project.
Why?
I wanted to explore something specific: what happens when you remove linguistic meaning from music? When you can’t parse the words, your brain stops trying to “understand” and just feels.
I use music while I work, but lyrics in English or Portuguese pull my attention. My brain tries to process the words. I wanted background music that was deeply human (real melodies, real emotion) but didn’t occupy linguistic space in my mind.
So I invented a language for it.
The Process:
The Result:
Songs that feel deeply human but bypass linguistic processing. You can’t understand the words, but you can feel what they mean.
Example: “Thoria Velor” (Time Within)
A song about an ancestral community that lived in sync with nature and the cosmos. Sung entirely in Lumirian. The listener doesn’t know what the words mean — but the melody, the atmosphere, the pacing — they carry the meaning anyway.
What I Learned:
Language is one layer of meaning. But rhythm, tone, melody, space — those are older languages. They work even when the words don’t.
This experiment taught me something about AI too: the tools don’t care what language you’re working in. They respond to structure, emotion, and musical logic. Lumirian worked just as well as English.
Status: Kaera Varan has released multiple tracks. It exists as a real band with a real aesthetic, singing a language that will never be spoken.
Why This Matters for AI:
If you can teach an AI to work with a language that doesn’t exist — you prove it’s responding to structure and pattern, not just mimicking what it was trained on. That’s useful to understand.
Tools Used: Claude (language creation), Logic Pro, Suno
[Embed Kaera Varan track or link to music]
As part of the Kaera Varan project, I wrote this press release — treating a fictional band as if it were real:
For Immediate Release
Kaera Varan Unveils Ethereal New Single “Thoria Velor”
August 3, 2024 — Lumirian band Kaera Varan is thrilled to announce the release of their latest single, “Thoria Velor,” a mesmerizing journey through time and cosmic harmony. This new track, sung in their self-created Lumirian language, beautifully encapsulates the band’s signature atmospheric and introspective style.
“Thoria Velor,” which translates to “Time Within,” tells the story of an ancestral community that lived in perfect sync with nature and the cosmos. Through haunting melodies and poetic lyrics, the song explores themes of time, wisdom, and the eternal dance of life.
“Our goal with ‘Thoria Velor’ was to create a piece that captures the essence of timelessness and the deep connection our ancestors had with the natural world,” says Kaera Varan’s lead vocalist. “It’s a reflection on how living in harmony with the cosmos can reveal profound truths about our existence.”
Kaera Varan’s unique blend of atmospheric, bossa nova, and melancholic music shines in this single, drawing listeners into an otherworldly soundscape. “Thoria Velor” continues the band’s tradition of creating deeply emotional and thought-provoking music that transcends the ordinary.
About Kaera Varan:
Kaera Varan is an ethereal band known for their atmospheric and melancholic music sung in Lumirian. Their songs transport listeners to otherworldly realms, exploring themes of dreams, fears, and introspection. With “Thoria Velor,” the band delves into the timeless wisdom of an ancient community, offering a poignant reminder of our connection to the cosmos.
Building My AI Band: What Happens When You Give Suno a Human Foundation?
The Experiment:
I wanted to test something specific: if I give an AI music tool a real human performance as the foundation — my voice, my guitar, my lyrics — can it build around me without erasing me?
The Process:
What Happened:
It worked.
The AI added drums, bass, keys, atmospheric layers — a full band. But my voice stayed my voice. My guitar playing stayed recognizable. It didn’t homogenize me into “AI singer voice.” It collaborated.
Why This Matters:
Most people use Suno by typing: “Make me a sad indie rock song.” The output is generic because the input had no human specificity.
But if you give it something human to build around — your actual voice, your actual playing — it becomes a collaboration tool, not a replacement.
The Difference:
Status: Successfully released. The track sounds like me with a band I could never have afforded to hire.
Tools Used: Logic Pro, Suno
What I learned: AI doesn’t erase the artist if the artist shows up first with something specific and human. The prompt isn’t enough. The performance is.
How To Be Human: Making a Film Accessible to Everyone
When I directed How To Be Human (a VFX-heavy sci-fi short), I partnered with the Royal National Institute for Blind People (RNIB) to create an audio-described version.
What is audio description?
A narrator describes the visual elements of the film — body language, settings, action — in the gaps between dialogue. This allows blind and visually impaired audiences to experience the story fully.
Why I did this:
Innovation means nothing if it isn’t accessible. I wanted to know: could a film designed around visuals still work as a sonic experience?
What I learned:
Audio description isn’t just accessibility. It’s a creative constraint that makes you better at visual storytelling. You realize which details actually matter — and which are just decoration.
The result:
The audio-described version screened at festivals and was used by RNIB in their training programmes. More importantly, I learned to think about reach differently. The work was never about the medium. It was always about finding the person who needed to hear this — and removing every obstacle between them and the story.
Tools Used: Professional AD scriptwriter, RNIB consultation, custom audio mix
ComfyUI: Why I’m Learning to Think in Node Workflows
Most AI image tools are black boxes: type a prompt, get an image. ComfyUI is different — it’s a visual programming environment where you build generation pipelines using nodes.
Why I’m investing time in this:
What I’m building:
A custom workflow for generating storyboard frames that match the cinematography style of Darker Days. Training it on reference images from the shoot so the AI understands our specific lighting and framing language.
Status: Still learning. But already more useful than MidJourney for this specific use case.
[Placeholder for ComfyUI node graph screenshot]
Think AI: A Framework for Creative Collaboration
I’m developing a teaching framework called Think AI — a way of working with generative tools that preserves human voice instead of diluting it.
The Problem:
Most people approach AI like a genie in a bottle: “Make me a poster.” “Write me a script.” “Generate a song.”
The output is technically correct but emotionally empty. Because the prompt didn’t carry intention. It carried a request.
The Framework:
1. Start with intention What are you actually trying to say? What feeling are you trying to create? Who needs to experience this, and why?
2. Think in systems, not outputs AI tools are best used as part of a pipeline — not as a replacement for thinking. Build workflows that allow iteration, not just generation.
3. Bring context Your voice comes from your life. Feed the tool references, constraints, and perspective. Otherwise it will give you the average of everything everyone else has already made.
4. Evaluate ruthlessly Does this output carry your point of view? Or could it have been made by anyone? If the latter — iterate or start over.
The Goal:
AI should expand what you can make, not replace who you are.
The Gap Between Imagination and Execution
There’s still a distance between what I can imagine and what I can make AI tools produce.
I know exactly what a shot needs to feel like before I render it. The light angle. The texture. The weight of the atmosphere. But the prompt doesn’t carry that knowledge. Neither do the reference images.
The tools weren’t built for someone who already knows what they want. They were built for exploration and generation.
That gap isn’t a failure. It’s the most interesting place I’ve ever worked.
Question I’m sitting with: How do you train a model to understand cinematic intent, not just visual aesthetics?
The Legacy Protocol: What If We Could Digitize the Soul of a Lens?
The Idea:
Cinema lenses have character. An Arri Master Prime renders differently than a Panavision C-Series. That character comes from optical imperfections — how the glass bends light, how bokeh forms, how flares behave.
The Question:
Could we train a generative model to understand the optical characteristics of legacy cinema lenses? Not just mimic the aesthetic — but actually reproduce the physics?
Why This Matters:
Indie filmmakers can’t afford $100k lens sets. But if we could train AI to apply those optical characteristics to footage shot on affordable glass — we’d democratize access to cinema-quality imagery.
Status:
Pure R&D. I’m researching optical physics, talking to lens technicians, and exploring whether this is even technically possible. No promises. Just curiosity.
What I’d need to test this:
If you’re a lens nerd or an AI researcher interested in optics — let’s talk.