Story and structure
Claude helped me develop the premise, screenplay logic, character rules, and a production brief that connected the emotional arc to a practical shot map.
AN INDEPENDENT AI FILM EXPERIMENT · 2026
Operation
Grassfire
A cowboy, a grasshopper, and one very large promise.
Operation Grassfire is my very first movie made with AI. I made it from scratch with Claude and Higgsfield. What began as a five-minute film became 9 minutes and 36 seconds. Here is the finished film and the production experience behind it: what worked, what failed, what it cost, and what I learned.
Watch the full film
9 min 36 sec · 16:9 · English dialogue
The film
A small life. A global rescue.
Bonan, a Marine veteran turned Yellowstone park ranger, rescues a small grasshopper named Hopper—and gets the surprise of his life when she thanks him out loud. A tiny brass keepsake is the only clue to relatives scattered generations ago. Their search crosses the Pacific and leads to a fictional trafficking annex beneath a Seoul palace, where a rescue mission becomes an unruly celebration of courage, family, and second chances.
Why I made it
From a blank page to a finished film.
I approached Operation Grassfire as a first-principles experiment: can one person take an original story from idea to finished movie with today’s AI tools? The answer is yes—but not in one click. The real work was directing a system that does not reliably remember: defining characters, translating story logic into checkable shots, reviewing every generation, correcting continuity, and assembling the final cut outside the generation platform.
How it was built
The workflow behind the screen.
Continuity system
I created reusable character, wardrobe, prop, voice, geography, and insect-anatomy constraints, then restated the critical rules inside every individual shot prompt.
Generate and audit
I generated clips in Higgsfield, inspected each one immediately, and corrected failures in sequence: wording, pose, camera angle, then—when necessary—the shot structure itself.
Edit and finish
The platform generated individual clips, not a finished timeline. I trimmed, repaired, ordered, and assembled the movie externally, preserving cause and effect across the complete cut.
Production record
These figures combine records from multiple sources rather than a single platform screenshot.
~14calendar span
76generated videos
13:22raw generated footage
$489.99personal tool spend
What the experiment revealed
Four things my first AI film taught me
Mechanisms matter more with AI
AI increases the speed and volume of production, but it does not remove the need for discipline. Explicit standards, clear constraints, and immediate review are essential. Without them, faster generation simply produces more inconsistency.
Quality failures compound
An early defect rarely stays isolated. Problems in the concept, script, character design, or visual continuity become increasingly difficult and expensive to correct once later work depends on them. Quality must be checked at every stage—not only at the end.
Every addition has a cost
The original plan was a five-minute film. It became 9 minutes and 36 seconds. Every new scene introduced more than runtime: it required additional assets, continuity, sound, transitions, revisions, and opportunities for failure. Creative ambition needs an explicit complexity budget.
Human judgment is the integration layer
AI generated many of the assets. It did not decide what belonged in the film. Selecting, rejecting, sequencing, reconciling, and refining those outputs required human judgment. The final work emerged from those decisions—not from generation alone.
Practical lessons from the shots
Four lessons I would carry into the next film.
Continuity must live at shot level.
A character bible is not enough if the generation model never sees it. Any detail that matters—wardrobe, anatomy, voice, props, and temporary states—has to appear in the prompt for that shot.
Camera geometry can solve what words cannot.
After a wording fix fails, change the pose or angle. A structurally clearer shot often prevents occlusion and anatomy errors more reliably than another paragraph of prompting.
Story logic has to be visible.
Motivation, setup, and location transitions cannot remain in the writer’s head. Each shot must earn the next one on screen, or the audience will feel the missing link.
Scope is a production decision.
The target grew from five minutes to 9:36 through individually reasonable additions. Runtime, character roster, cost, and regeneration budget need to be locked and tracked from day one.
If I started again
I would lock every recurring character before generating the first frame, complete a cause-and-effect audit of the whole shot map, paste a universal continuity checklist into every prompt from day one, cap each shot at two or three story beats, and maintain a live ledger of credits, cost, time, attempts, and accepted footage.
Final reflection
The expense was real, and so was the discarded work. But the project changed my understanding of AI filmmaking. The scarce capability is not simply generating attractive images; it is judgment—knowing what must stay consistent, what the audience needs to see, when to redesign a shot, and when the story is finally complete. Operation Grassfire is my first film, and the most valuable output may be the production system I learned to build along the way.
Fiction, AI & independence disclosure
This is a fictional, AI-assisted film. All characters, organizations, incidents, military action, and the palace service annex are invented. No government, military, museum, heritage institution, or nationality is depicted as participating in wrongdoing. AI-assisted images of real places are dramatic representations, not documentary footage. The project was created independently, outside working hours, without employer resources, approval, sponsorship, or endorsement. Views and creative choices are personal.