Inside AIM Studio · Production Methodology

How we produced No Man's Land

An eight minute dramatic short, made in Google Flow. This is the actual method: how the story was locked, how characters stayed the same person across hundreds of generations, and how it was finished in the edit.

We are not replacing Nollywood. We are telling the stories the big companies pass on, and giving them everything we have.

Watch the film Download this guide

8:17Runtime

8Part shot prompt

5Pass review before a take is kept

0Cameras

The philosophy

Flow is a film set, not a prompt box.

The strongest results came from treating Flow as a controlled production environment rather than a wishing well. Every generation had a specific dramatic purpose, a defined camera position, one dominant action, and a continuity target. We never asked the model to discover the scene for us. The story was already known; the model was directed toward a precise screen moment.

The screenplay, the emotional arc, and the scene objectives all existed before Flow was opened. A technically impressive image still fails if the audience cannot tell what the character wants, what changed, or why the shot exists.

Sequencing the action

Not a limit on how much can happen. A limit on how vaguely you can ask for it.

Short generations fall apart when a prompt stacks actions without structure. Asking for enter the room, speak, hand over the paper, flip it, reveal the drawing, and react gives the model an entire edited scene to solve at once, and you get rushed blocking, broken hands, and emotional confusion.

The failure there is not the number of actions. It is that none of them were sequenced, no hand was named, no contact point was specified, and no end position was set. A well-structured prompt can carry several actions in one generation, because each one is written in chronological order with explicit physical detail. An unstructured prompt cannot reliably carry even two.

The decision is also editorial. Coverage exists so the editor can control pacing, hide continuity differences, and cut on movement. Even when one long take succeeds, separate shots often serve the scene better.

Continuity is a system, not a memory

Characters and props both get locked before anything moves.

Every principal character has a continuity profile: identity, exact wardrobe including which side each item sits on, a performance baseline for how they carry grief or fear, voice direction, and a list of the features the model likes to change. Approved reference images are kept clean and shot from multiple angles.

Props that carry story meaning are treated as characters too. The drawing, the paper, the tray, the watch. Each gets its own reference and placement rules.

The eight part shot prompt

A prompt is a directing document, not a pile of adjectives.

1. Shot identityShot size and camera position, stated first.
2. Continuity lockThe exact character, wardrobe, prop, and location references.
3. Starting compositionWhere every visible person and object begins.
4. Dominant actionOne primary action, in chronological order.
5. Performance directionEmotional restraint, eye behavior, breath, tempo.
6. Camera behaviorStatic, push in, pan, handheld, rack focus.
7. Visual treatmentLighting, lens feel, depth of field, texture, grade.
8. ExclusionsThe specific failures that would make the shot unusable.

Weak prompt versus directed prompt

Emotion tags and exclusions

Two habits that saved more regenerations than anything else.

Delivery is written inline, in brackets, immediately before the line. It tells the model how to play the beat instead of hoping the words carry it. A tag is short, physical, and unambiguous.

Every shot then carries its own exclusion list. Not a generic wall of avoid-words, but a short set of specific, observable failures written for that shot alone. The categories that matter most:

Keep it targeted. An enormous catch-all dilutes the positive direction. The strongest protection is still a clear composition and precise action language, with exclusions guarding only the handful of things that would actually ruin the take.

Directing emotion

Broad adjectives produce melodrama. Physical behavior produces truth.

Choosing the right Flow mode

The mode is chosen by the problem, not by habit.

The five pass review

A take is not kept because it is almost right.

A wrong principal face is regenerated, never colour corrected into hiding. A hand error during an essential transfer is regenerated or split into smaller shots. A good action with an unstable final second is simply trimmed before the instability.

Hands, props, and the things that break

• Hands are treated as a dedicated effects problem. Reduce simultaneous contact, show the grip clearly, keep finger paths short, and cut between phases of a transfer.
• A natural handoff often needs two generations: the giver extends and holds, the receiver secures and withdraws.
• The paper has a defined front, back, top edge, and hand ownership. The flip direction is specified relative to camera.
• Eyelines matter as much as performance. Correct screen direction is what makes separately generated coverage edit together.
• The generated clip does not all get used. The first seconds often contain settling motion and the last second often deforms.

Sound is not optional

Native audio is a creative option, not an obligation.

The soundscape is a post production mix, never accepted automatically from each clip. Dialogue carries the information, one consistent room tone makes the cuts feel continuous, foley makes paper and clothing believable, and score enters late and sits below the breaths. Silence is a tool: the drawing reveal is given space before any music expands.

Ten things the production proved

1. The script remains the foundation. The emotional power came from the written situation, not visual novelty.
2. Character consistency starts with a locked reference system and becomes an editorial discipline.
3. A complicated action should be decomposed into coverage, never forced into one generation.
4. Hands and props require their own master frames, blocking, and acceptance rules.
5. Restrained emotion is generated more reliably through physical behavior than through adjectives.
6. The best camera movement is usually the smallest movement that improves the beat.
7. Frames to Video earns its keep when a reveal needs a precise final composition.
8. Professional sound design decides whether the scene feels real.
9. The editor is the final continuity engine. Cuts can unify separate generations, but cannot rescue unclear storytelling.
10. A naming and versioning system saves more time than improvising after hundreds of assets pile up.

Consistency is the whole craft

The word does double duty in this work.

Consistency is the technical problem. The same face across a hundred generations, the same watch on the same wrist, the same light in the same room, the same voice in every line. Everything in this document exists to protect it.

Consistency is also the working practice, and that is the part that decides whether anyone gets better. Nobody’s first generation is good. Improvement does not arrive because a better prompt was discovered; it arrives because the same system was applied again, and again, and the standard never dropped.

This is why the naming convention, the continuity lock sheet, and the generation log matter more than they look. They are not bureaucracy. They are what turns a lucky take into a repeatable result, and a repeatable result into a body of work.