Vacuuming the Desert

The short answer: this AI short film prompt keeps one character consistent across five shots by attaching a character sheet, writing one energy block that every shot inherits, and cutting between one locked wide shot and two close-ups. Our 13-second desert vacuum film was made this way in Seedance 2.5 on Magnific. The full prompt is below.

Most AI short films look like stock footage because the prompt describes a scene, not a performance. The character stands there, the camera drifts, and nothing feels intentional. A good AI short film prompt with a consistent character tells the model how hard the person is working and keeps the same face in every cut.

We at Artea made this 13-second film for Riya's Instagram: she vacuums the Sahara with the focus of someone defusing a bomb. It was made with Seedream 5 Pro and Seedance 2.5 on Magnific. This post gives you the character sheet prompt, the storyboard and the full video prompt.

The tools we used

Seedream 5 Pro on Magnific: the character sheet

Seedream 5 Pro is an AI image model. We gave it three photos of Riya and one outfit photo, and it made a turnaround sheet of the same woman from the front, side and back.

Why it matters: the video model copies her face and clothes from this sheet in every shot, which is what keeps her looking like the same person after each cut.

A storyboard: the shot layout

A storyboard is a row of rough frames that shows what the camera sees in each shot and for how long.

Why it matters: attached as a reference, it tells the video model where to put the camera, so the wide shots match from cut to cut.

Seedance 2.5 on Magnific: the video

Seedance 2.5 is an AI video model that accepts reference images and a multi-shot prompt, and it adds sound effects. The film was generated with it in one run.

Why it matters: one generation means one consistent look, light and grade, instead of separate clips that need matching in an edit.

All three are available on Magnific, which is where we made the whole film.

The character sheet and its prompt

This is the prompt we ran in Seedream 5 Pro. Images 1 to 3 were photos of Riya and image 4 was an outfit reference. Swap in photos of your own character and your own outfit idea.

Create a character reference sheet.

IDENTITY - Images 1, 2 and 3 are the SAME real woman. Copy her face exactly as it appears in these photos: same face shape, eyes, nose, lips, smile, teeth, skin tone, eyebrows and proportions. Do NOT beautify, slim, age, lighten skin or change her features. Her face must look identical in every panel.

- Young Indian woman, early-to-mid 20s, round soft face, warm medium-brown skin

- Black thin rectangular eyeglasses (always on)

- Very long, voluminous, wavy dark-brown hair, middle parting, worn loose past the waist

- Small gold bangle on her right wrist

OUTFIT - Image 4 is for clothing ONLY. Ignore that model's face, hair and body entirely.

- Faded khaki/beige long-sleeve button-up work shirt, sleeves rolled to forearms, open collar with white tank top underneath, light brown dirt stains

- Matching khaki work trousers with white paint splatter

- Dark brown leather belt with brown leather tool pouches on both hips

- Dark worn knee pads

LAYOUT - plain light-grey studio background, soft even lighting, same scale in every view:

Top row: full-body turnaround, standing relaxed with arms at her sides - front, 3/4 front, side profile, back.

Bottom row: head-and-shoulders close-ups - front, 3/4, profile - all with the same face as the reference photos.

STYLE: photorealistic, natural skin texture, true-to-life, clean reference sheet with thin divider lines.

Negative prompt:

different face, face from image 4, ponytail, no glasses, lighter skin, slimmer face, beautified features, changed nose, changed smile, cartoon, inconsistent identity across panels

The storyboard

The storyboard is a simple five-panel layout: three identical wide shots with two face close-ups in between. An early version of it used a different outfit for the character, so the video prompt overrides that and points to the character sheet for her look.

The full AI short film prompt with a consistent character

Attach the character sheet and the storyboard, then paste this into Seedance 2.5.

4:5 vertical, 13 seconds, photoreal, 35mm film grain, muted warm palette, dry absurdist tone. No text, no titles, no logos, no music.

Character (match reference image exactly): young Indian woman, mid-20s, long loose wavy dark-brown hair, black thin-rimmed rectangular glasses, thin gold bracelet on right wrist. Worn beige/khaki button-up work shirt, sleeves rolled to the elbows, open collar over a white tank top, dust and sweat marks on the fabric; matching beige work trousers with white paint splatters; worn dark-brown leather tool belt with two heavy pouches bouncing on her hips; dusty brown leather work boots.

Energy (applies to every shot): she is working HARD - intensely determined, fully committed, as if this is the most important job of her life. Fast, forceful, vigorous strokes with the vacuum wand, shoulders and whole upper body driving every push and pull, knees bent in a wide braced stance, boots digging into the sand, body leaning into each stroke. Sweat on her forehead and neck, damp strands of hair stuck to her face, breathing hard. Jaw set, brows furrowed, eyes locked on the sand. Not smiling. Never slow, never idle, never casual.

Prop: glossy red canister vacuum cleaner, grey ribbed hose whipping with her movement, chrome wand. Sand spraying and dust puffs kicking up at the nozzle with every stroke.

Setting & light (identical in every shot): vast Sahara sand dunes, hazy harsh midday sun behind her, white overexposed sky, long hard shadow falling toward camera. Wind whipping her loose hair.

Shot 1 (0-4s): Static locked-off wide shot. She is small in the lower-centre of frame, red vacuum on the sand beside her, attacking the sand with fast, aggressive back-and-forth strokes, whole body rocking with each push, dust kicking up around the nozzle, hose jerking behind her. Endless dunes behind. No camera movement.

Shot 2 (4-6s): Cut to medium close-up, three-quarter front angle at eye level. Her face clearly visible through her glasses - intense, furrowed brow, gritted jaw, sweat beading on her forehead, damp hair strands stuck to her cheek, breathing hard through her mouth. Shoulders visibly driving the vacuum stroke in and out of frame. Quick swipe of sweat off her brow with her forearm, straight back to work. Dunes soft-focus behind. Static camera.

Shot 3 (6-9s): Cut back to the exact same static wide shot as Shot 1. She is still going hard - even faster rhythm, leaning deep into each stroke, sand spraying.

Shot 4 (9-10.5s): Cut to low-angle medium shot from the front. She is bent low over the wand, both hands gripping it tight, forearms tense, pushing hard through the sand, tool belt pouches swinging, glasses slipping slightly down her nose, eyes fierce and locked on the sand. Static camera.

Shot 5 (10.5-13s): Cut back to the exact same static wide shot as Shot 1. She keeps attacking the sand relentlessly until the end.

Sound: dry desert wind, loud straining vacuum motor, sand hissing and rattling up the hose, her heavy breathing.

Why one wide shot three times? Returning to the exact same locked wide frame between close-ups gives the film a rhythm and makes the effort look endless. Saying "the exact same static wide shot as Shot 1" is what stops the model from inventing a new angle.

How to make your own version

Step 1: Pick an impossible task

Vacuuming a desert works because a vacuum is a normal tool and the desert is a normal place, but together the task can never end.

How to use it: the prop and the place should be ordinary on their own and absurd together.

Step 2: Write the energy block once

The energy paragraph says she works hard in every shot: fast strokes, braced stance, sweat, no smiling.

How to use it: keep this block separate from the shots so every shot inherits it. Without it, AI characters drift into slow, calm movement.

Step 3: Alternate wide and close

Use the same locked wide shot for shots 1, 3 and 5, and two different close-ups in between.

How to use it: give each close-up one specific detail, like sweat on the glasses, so it adds something the wide shot cannot.

Common mistakes when writing an AI short film prompt with a consistent character in 2026

• Letting the character smile, which turns the joke into a happy scene.
• Moving the camera in the wide shots, which weakens the deadpan look.
• Describing the effort only in one shot, so the rest look lazy.
• Adding music when the sound of the effort is the punchline.

FAQ

How do I keep the same character in every shot of an AI video?

Attach a character sheet as a reference image and describe the character once in fixed detail, including small items like glasses and jewellery. Say "match reference image exactly" so the model treats the sheet as the source of truth.

What resolution does this Seedance 2.5 prompt render at?

Our version was generated at 480p as a draft. For a final post, generate again at a higher resolution once you are happy with the motion.

Can I change the prop and the location?

Yes. Keep the energy block and the wide, close, wide rhythm, and swap the vacuum and the desert for any task that can never be finished.

If you came here from the reel, comment on it and follow Riya, and the prompt will land in your DMs too.

In short: an AI short film prompt with a consistent character needs a character sheet, a separate energy block and a locked wide shot you return to. Make the sheet in Seedream 5 Pro, lay out five shots, and run the full prompt above in Seedance 2.5 on Magnific.