Across Every Universe: How I Built an AI Film from Zero
Mahesh Bheemaraj
Across Every Universe: How I Built an AI Film from Zero
A case study in AI filmmaking workflow, creative direction and the art of knowing when to stop generating
I'm a traditional advertising filmmaker. I've spent years directing commercials — real crews, real locations, real budgets. When AI video generation arrived I saw it immediately for what it was: not a threat, but a new medium. But I needed to prove I could work in it. Not just generate clips — actually make a film.
I had no actors. No locations. No crew. I had AI generation tools, a laptop running DaVinci Resolve, and one question: what should the film be about?
The answer came quickly. Make it about the only subject I know completely — the experience of being a creative professional. The impossible briefs. The endless revisions. The deadlines that arrive like emergencies. Every filmmaker, every writer, every designer knows this experience. It's universal. And it's funny.
The concept: a multiverse showreel. Across every genre, every era, every universe — there is always a Mahesh, always dealing with the same creative problems. I called it "Mahesh Bheemaraj: Across Every Universe."
This is how I made it.
Chapter 1 — The Concept
The hardest thing about a showreel is that it's a showreel. Everyone knows what they're watching and why. The defences go up. You're being sold to and you know it.
I wanted to make something people would watch because they wanted to — not because they were evaluating my work. So I decided the showreel had to be a film first and a credentials piece second. If the film worked, the credentials would be implicit.
The multiverse concept solved three problems simultaneously. It explained why every scene looked different — different genre, different universe. It gave the film a spine — one character, one theme, across every world. And it made the creative professional's experience the subject rather than the backdrop. The client revision joke isn't background colour — it's the whole point.
"Across Every Universe" is a comedy about creative work. The showreel is almost incidental.
Chapter 2 — The Workflow
I learned this methodology the hard way — through failed generations, inconsistent characters and expensive mistakes. By the end of the project it was second nature. Here's what works:
Build all references before touching any generation tool. Character turnarounds, location references, prop references — everything locked before the first storyboard panel is written. Every hour spent on reference saves five hours in generation.
Storyboard every scene before generating a single clip. Not as a planning document — as a generation anchor. A panel crop used as a Kling start frame is worth more than the most detailed text prompt. The storyboard is pre-production and production simultaneously.
Chain start frames. The last frame of every clip becomes the first frame of the next. Character consistency lives or dies here. Break the chain and the character drifts.
Match the tool to the shot type. Kling for dialogue and camera movement. Seedance for facial performance and micro-expression. Higgsfield for environment and atmosphere. Using the wrong tool is the most expensive mistake I made repeatedly before I stopped making it.
Edit in DaVinci Resolve throughout — not at the end. The edit is where the film actually lives. Every generation is raw material. The film gets made in the timeline.
Chapter 3 — The Scenes That Worked
Apollo 13 came together in one session. The brief was clear — mission control, seven revisions, a new logo. The ensemble casting gave me natural shot variety. The comedy was in the situation, not the dialogue. I knew exactly what I wanted and the tools delivered it.
Nat Geo was the surprise. A single voiceover-driven scene — a filmmaker waiting for client feedback, framed as a wildlife documentary about patience. The kingfisher moment. No dialogue, no complex blocking, just atmosphere and a voiceover that recontextualised everything. It taught me that the simplest scenes often carry the most emotional weight.
The Puppet was the most unexpected success. A marionette version of me in a workshop, saying the truest thing in the film: "Nobody remembers the tools — they remember how you made them feel." Eight seconds. One shot. The best generation in the project.
The Pixar scene came from a difficult day — a brainstorm session when I needed to step back from individual scene problems and think about the film as a whole. I asked myself what was missing. The answer was emotion — genuine, unperformed emotion. A scene with no joke. No punchline. Just a person finishing something and feeling something about it. The result became the emotional heart of the film.
Chapter 4 — The Scenes That Fought Back
Sherlock Holmes cost me the most in time, money and frustration. The concept was right — Holmes examining a paradoxical creative brief, the client asking "Can you make it pop?" — but the execution fought me at every step. Arcane animation style. Two characters. Dialogue generation. Background consistency. Each of these alone is manageable. All four together, in Seedance, proved brutal.
What I learned: the generation is often telling you something. When a scene fights back this hard, the tool is revealing the genuine complexity of what you're asking. I eventually solved Sherlock by switching to Kling for the dialogue clips and accepting a voiceover-driven cut for the deduction sequence. The lesson — know when to change your approach rather than repeat the same attempt.
Wes Anderson was the scene I abandoned. Two full storyboard regenerations. A forensic analysis of seven consistency failures across nine panels. Background drift, prop scale changes, the Bellboy appearing in two places simultaneously. I spent significant time and credits before accepting that the specific combination of perfect symmetry, two-character blocking and consistent props across nine panels was beyond reliable generation quality at this stage of the tools. I moved on. That decision was correct.
Chapter 5 — The Edit
I had been editing in DaVinci Resolve throughout the project — learning the software at the same time as using it. That combination of learning and doing is uncomfortable and I recommend it completely. You learn what you need when you need it.
The running order changed three times. The Puppet moved from the closing scene to mid-film — between Nat Geo and Sherlock. The discovery: the Puppet line works better as a thesis statement that the film then proves rather than a conclusion. State what you believe, then demonstrate it through Sherlock's impossible brief, Inception's impossible world and the Pixar scene's emotional truth.
The Pixar scene moved to the penultimate position — just before the desk return. The animated Mahesh saving his film, the cold open frame appearing on his monitor, the lamp clicking off. Cut to the real Mahesh at his real desk. The multiverse collapses back to the one true universe. The loop closes.
Watching the full film end-to-end for the first time — rough cuts, imperfect generations, gaps still in the timeline — I knew the structure was right. The film felt like a film. Not a showreel. A film.
Chapter 6 — Ten Things I Learned
1. The prompt is a director's brief, not a search query.
Specificity of emotional intention matters more than technical description. "He looks up with the quiet curiosity of a filmmaker encountering something extraordinary" generates a better result than "man looks up at buildings."
2. Reference sheets are pre-production.
Every hour spent building character turnarounds, location references and prop references saves five hours in generation. I learned this after the first three scenes and never forgot it again.
3. The start frame is everything.
Character consistency lives or dies on the chain of last frames becoming first frames. Break the chain once and you're regenerating from scratch.
4. Text cannot override a strong visual reference.
If the reference image shows three monitors, no amount of "ONE monitor only" in the prompt will remove them. The visual reference always wins. Choose your references carefully.
5. Know which tool does what.
The wrong tool costs money and time. I learned this by using the wrong tool repeatedly.
6. The storyboard is a generation tool.
A panel crop used as a start frame is worth more than any text description. Build the storyboard properly — it does double duty as planning document and production asset.
7. When a generation fights you, listen.
The Wes Anderson scene fought back because what I was asking was genuinely hard. The generation was right. I should have listened sooner.
8. The edit reveals what the film needs.
No amount of planning replaces watching the rough cut end to end and trusting what you feel. The running order I planned at the start was wrong. The running order I arrived at through editing was right.
9. Constraints are creative.
Five credits. One close-up. One line. The Inception scene's best version came from having almost nothing to work with. The constraint forced the most elegant solution.
10. The film is never finished in generation. It's finished in the edit.
Every clip is raw material. The film lives in DaVinci Resolve. Generation is photography. Editing is filmmaking.
The closing frame:
The film runs two minutes and thirty five seconds. It was made without actors, without locations, without a crew, without a budget in the traditional sense.
It demonstrates action, comedy, documentary patience, intellectual wit, visual spectacle, emotional depth and Pixar-quality animation. It has a thesis, a structure and an ending that earns its emotion.
More importantly — it exists. It is posted. It is being watched.
The tools will change. The models will improve. The workflows will evolve. But the filmmaker's eye — the instinct for what a scene needs, when a generation is lying to you, when to fight and when to accept — that doesn't come from the tools. That comes from the person holding them.
Nobody remembers the tools.
They remember how you made them feel.
Like this project
Posted Aug 18, 2026
I made an AI film. It launched my career as an AI filmmaker. Old clients hired me. New conversations started. The launchpad delivered.
