How Rommel Villa made Morgue Cat with AI

Jonesy holds a mouse while floating inside the observation dome — A victory-float study for Scene 12. Jonesy has caught the mouse.

Jonesy can turn off the gravity on a space station. She still has to move like a cat.

I asked the agent for a board of her reactions: cute, excited, alert, scared, angry, ready to attack, heartbroken. Then I wanted her running, falling, floating, soaked in water. I asked about books on cat behaviour. Her ears, her eyes, her whiskers and the way she holds her body have to do the acting.

In Morgue Cat, she lives with a pathologist who plays fetch with her using tuna-can pull tabs. She brings him a mouse as a gift. That mouse turns out to be something much worse. We spend the opening getting comfortable with her before the film becomes terrifying.

The cat’s two eyes are different colours, on specific sides. That went into the project rules in Absurd. A new angle or another person working on a scene shouldn’t give us a different Jonesy.

Rommel Villa outdoors at dusk — Rommel Villa · Morgue Cat

Jonesy, an orange-and-cream cat with different-coloured eyes — An early expression study. The character board keeps her identity available to the team.

A retro-futurist morgue with teal drawers and a central autopsy table — Morgue look development. The same room has to work for play, a delivery and an attack.

We were making different parts of the same film

Robin worked through the opening. He wanted Jonesy strutting home to the morgue, turning the drawers into jumping boards and making a mess before Dr Graham arrived. Mengqing worked on the doctor, Tech Harris and the specimen-jar handoff. Daniella worked through the chase, from the medical hallway into the vents and greenhouse. I spent a lot of time in the observation room, working out what happens when the gravity goes off.

Those conversations lived on the same Absurd canvas. We had separate pages for the scenes, alongside Characters, Environments and Environments Locked. Someone working on a chase could use the character references already developed for the opening. We could copy a room onto a scene page while leaving the environment reference where everyone expected to find it.

Robin asked the agent to bring his chosen establishing shots onto the production page and place them beside their corresponding shots. He also reviewed generations from Mengqing’s drawer-play work. Feedback about the overhead light went back with the selection. The decision stayed attached to the work someone else had started.

I still needed room to try things. At one point I told the agent to treat an opening experiment as a Rommel test. Having the other conversations available is useful, but I need to say when I’m exploring a departure from what the team has settled on.

Scene 12 · The observation module

Make a mess. Then make it float.

The room has to be beautiful. I wanted teal nebulae, stars, a view that could distract Jonesy in the middle of catching her prey. I kept asking for the sky to feel more romantic, more ethereal. Some attempts looked fake; others put the comet somewhere her eyes weren’t looking.

A comet passes beyond the curved glass observation dome — A comet study. Its position has to agree with Jonesy’s eyeline.

Before the gravity changes, Jonesy and the mouse race across the controls. Buttons light up. Drawers fly open. Objects spill out. When everything lifts, we need to understand where all that debris came from.

I asked for the scene as a series of multishots and discussed different plans with the agent. We tried camera moves that stayed close to Jonesy and others that travelled out into the room. I wanted the camera to feel like another cat watching her.

Scene 12 entrance. A working take of the approach to the observation room.

The victory float. One of the team’s working takes, with Jonesy carrying her catch.

Stills on the left. Clips on the right.

For each sequence, I wanted its own container in Absurd, with the images on the left and video generations on the right. I asked for the shotlist to sit with that sequence too. When another row of takes arrived, the blocks below needed to move down. I didn’t want clips overlapping or floating around without a home.

That arrangement lets us compare the intended shot with what came back. The references can stay the same while we change the timing, the camera move or Jonesy’s reaction. I can ask the agent about a take before deciding to spend another round on it.

I’m quite particular about that last step. I want a plan before it generates. When the agent went ahead after I’d asked a question, I reminded it that we were still talking. We used accept and reject to review the results, but approving an attempt and discussing an idea are separate decisions.

A mug becomes part of the gravity sequence. It needs a recognisable shape, and we need to know where it is before Jonesy uses it. We made a reference sheet, then kept it with the other scene references. The props have to survive the same continuity checks as the cat.

Three views of the cream-and-teal mug used in the observation-room sequence — The mug’s reference sheet. A small prop with a job in the action.

Much later, I was looking for an older clip: Jonesy sees the comet, realises the mouse has escaped and opens the gravity-button cover. I described it to the agent, found the take and compared it with the newer shot. The blocking, colour and lighting didn’t match. We had to rebuild the transition.

We also had the agent label the clips by scene: the intro, the chase, the gravity button and the catch, with take numbers carried into the downloads. That makes handing off several versions less confusing than sending a folder of unrelated filenames.

There is one less practical rule in the project. When I say “good job”, I asked the agent to start its reply with “Meowwwww!”

Q&A with Rommel Villa

Edited from a conversation for length and clarity

What drew you to AI, and how has your perspective changed?

Rommel Villa

My background is in systems engineering and computer science, so I’ve always been interested in technology. While studying engineering I realized I was especially interested in how technology helps people be creative. That’s what AI does for filmmakers. It offers the potential to tell stories beyond Hollywood budgets and the usual limits of performances and environments. The universe is the limit.

Since using AI, I’ve become much better at communicating what I want. I think AI is a writer’s medium: you have to describe your vision and needs precisely. Good writing and storytelling help you get the results you need.

As a director and editor, how has AI changed your filmmaking process?

Previsualization is one of the biggest changes. You can practically shoot a film before production and test ideas that would otherwise get only one chance on set. AI makes it possible to request more takes and to alter characters, looks, and environments while you’re already editing. Traditionally those stages happen one after another; now you can edit while you generate. If I’m missing a shot in post-production, I can generate it in Absurd, and the agent remembers the project context.

You can also test characters and wardrobes across many settings before location scouting or costume tests. Instead of committing to a handful of options, you can explore hundreds of looks, colors, hairstyles, and interactions with environments. It makes pre-production much more efficient.

Walk us through the way you organize and use Absurd.

I create a new project, invite my collaborators, and build pages for characters, script, locations, references, and tests. Once we’re producing, I add a page for each scene or sequence. I like everything extremely structured. Agents across the project understand what’s on the different pages, so I can retrieve information or assets from another scene easily.

I like the script on the left, character references at the top, and image references alongside video generations and the shot list. I treat the canvas almost like editing software: it should let me review clips and give feedback in an organized way. I tell the agent how I like to work and it helps me structure the pages accordingly.

How do you direct a believable AI-generated cat in Morgue Cat?

That was one of our biggest challenges. I started with an Emotion Tests page for our locked cat character: happy, frightened, angry, chasing, and so on. The first results were fairly accurate but lacked emotion. When I asked for more, they became cartoonish.

So I gave the agent homework: public-domain books, PDFs, and references on cat behavior. Cats don’t smile like people. Their ears, whiskers, tails, posture, and head movements communicate emotion. I also supplied videos of domestic cats and larger cats like jaguars and pumas for the predatory chase behavior. Once I asked for the same emotional tests again, the improvement was night and day. More accurate information enabled far more believable performances.

How much research did you give the agent?

Several books, many hundreds of pages long, on domestic cats, wild cats, and behaviors in different environments. In our film, the cat is affectionate around her owner but also has hunting and chase sequences. I wanted both sides to feel convincing. I used the agent to summarize some of the books I had not finished myself, and we provided images and videos too. It was able to apply that material in the generations.

How did multiple filmmakers coordinate on Morgue Cat using Absurd?

I began by uploading and revising the script, then talked with the agent about the world, retro-futuristic visual style, orange-and-teal palette, characters, and shooting approach. My core direction was to treat the camera as another cat: following the protagonist, running alongside her, and generally experiencing events at cat level, with occasional wider perspectives for action.

I put those rules into Absurd and presented the team with an overview. If collaborators had questions about camera style, lenses, or references, they could ask the agent, because it had the context of my creative decisions. The agent became a bridge between director and team. We worked in teams of two per episode with separate pages, reviewed generations at daily check-ins, and I edited footage as it arrived to work out what shots we needed next. Keeping the knowledge in one project made collaboration much more efficient.

What would you say to traditional filmmakers and editors who are skeptical of AI?

I understand their concerns; I had reservations too. I compare it to how the internet changed newspapers, radio, and telecommunications. Once a technology exists, it’s not possible to simply turn back time. We can complain about change, or learn to adapt.

I don’t want AI to replace me. I think learning how to use it as a creative collaborator is one of the best ways to protect the work we love doing and execute our visions ourselves.

Are you afraid of AI? And is it true that you meow at your agent and it meows back?

I’m not afraid of AI itself, but I worry about what it could do in the wrong hands. It can be used for art, education, and technological advances, and it needs responsible safeguards. I also think it’s important to have fun with it.

Since I spend so much time chatting with the agent, I personalized the interaction. I told it that whenever it did something well I would say “good job,” and asked it to reply “meow.” Now when it generates a shot I need, I say “good job,” and it meows back. It makes the work more fun.