The Last Ember — Cinematic Poster

THE LAST
EMBER

AI Cinematic Sequence — Storyvord Internship Assignment
Watch the Film
01 — The Project

The Brief

As an AI Filmmaker intern at Storyvord, I was tasked with creating a short AI cinematic sequence using an original idea. I chose the Fantasy genre and built "The Last Ember" — a story about a young temple keeper who carries the sacred flame through the night to a mountain summit, where dawn renews its power.

The goal wasn't just to generate beautiful images. It was to prove I could:

  • ✦ Maintain character consistency across 16 shots
  • ✦ Build a coherent visual world with cinematic lighting progression
  • ✦ Work within AI limitations and adapt my workflow in real time
  • ✦ Deliver a complete narrative with beginning, middle, and end
Tools Used
▶
Higgsfield (Seedance 1.5 Pro)
Video generation (Audio generated with clips, Seedance audio ON)
◈
Nano Banana Pro
Image generation
◎
Luvvoice
Voiceover narration
⬡
Google Vids
Final assembly, editing & auto-generated transcript captions for voiceover sync
02 — The Process

How It Was Made

This project was built through an iterative collaboration between myself and Kimi (Moonshot AI), an AI assistant that guided the entire workflow — from story concept and character design to prompt engineering, shot-by-shot generation, and technical problem-solving.

Key decisions made together
◎

Simplified the story to one character, one lantern, one journey for maximum consistency

◎

Discovered that Seedance 1.5 Pro handles Chinese prompts better for video generation

◎

Identified that dolly-back camera motions cause AI hallucination outside the frame

◎

Adapted the narrative mid-workflow from "emergency rescue" to "annual ritual renewal" when the generated fire never looked dying

◎

Tested Start+End frame techniques for character continuity across shots

◎

Troubleshot ElevenLabs IP restrictions and pivoted to Luvvoice for voiceover production

This shows the reality of AI filmmaking: it's not about perfect tools. It's about critical thinking, rapid adaptation, and knowing when to change the plan.

Workflow guided by Kimi (Moonshot AI)
03 — Behind the Scenes

Timeline

Image Generation (Shots & Frames)
5–6 hours
Character sheet, 4 location sheets, 16 shot images across all scenes
Video Generation (Animation)
4–5 hours
16 clips generated in Seedance 1.5 Pro, shot-by-shot iteration and selection
Voiceover Production
10–15 minutes
16 narration lines generated in Luvvoice (Elara Wynn voice)
Video Editing
30 minutes
Merged clips, synced voiceovers, trimmed timing, balanced volume
Webpage Build & Deploy
2 hours
Designed and built with Antigravity, deployed via GitHub + Vercel
Total Project Time
~12 hours

From first prompt to live website.

04 — The Story

The Myth

Every year, the temple keeper must carry the sacred flame from the temple through the night to the mountain summit, where the first light of dawn renews its power. At sunrise, she returns to restore the strengthened flame to the temple brazier — a ritual that ensures the village's protection for another year.

Scene 1
The Temple
0:00 – 0:27

Deep night. The keeper begins her ritual. She takes the flame and steps into the dark.

Scene 2
The Forest Path
0:28 – 0:55

Through mist and cold, she walks alone. The forest tests her resolve.

Scene 3
The Summit
0:56 – 1:27

At dawn, the mountain gives back what the night has taken. The flame is renewed.

Scene 4
The Return
1:28 – end

Home breathes again. The cycle is complete.

Lighting Arc
Cool Blue Night
Warm Gold Dawn
Bright Morning
05 — Character Design

Elara, The Temple Keeper

Elara — Character Sheet
Character Reference Sheet — Front · 3/4 View · Close-up
Age
22
Role
Keeper of the Sacred Flame
Outfit
Heavy rust-colored wool hooded cloak, worn leather edges, dark leather belt with brass buckle, small brass lantern with glass panels
Design Strategy
◉
Hooded cloak reduces face exposure, making AI consistency easier across shots
◉
Rust color creates strong visual contrast against cool blue environments
◉
Lantern serves as the story's central visual motif — the literal "last ember"
06 — World Building

The Locations

Location A — Stone Temple Interior
Night
Location A
Stone Temple Interior

Ancient granite walls with carved runes. Circular stone brazier. Arched windows. Cool blue moonlight. Where the story begins and ends.

Location B — Misty Forest Path
Night
Location B
Misty Forest Path

Tall pine trees, moss-covered ground, fallen leaves. Deep indigo night sky. Distant mountain peak barely visible through fog.

Location C — Mountain Summit
Dawn
Location C
Mountain Summit

Rocky terrain, ancient stone marker, low clouds below. Golden sunrise breaking over the horizon. Where the flame is renewed.

Location D — Temple Return
Morning
Location D
Temple Return (Morning)

The same temple, transformed. Warm golden light. Bright brazier flames. Peace and restoration. The cycle complete.

07 — Shot Breakdown

The Sequence

Scene 01
The Temple
0:00 – 0:27
Location A · Night
Shot 1 — The Empty Temple
▶
Play Clip
Shot 01
The Empty Temple
Extreme Wide / Establishing Slow push-in
"The flame burns… but it fades with time."
Shot 2 — The Keeper's Vigil
▶
Play Clip
Shot 02
The Keeper's Vigil
Medium Shot Slow pan
"Before the sun rises, I must walk."
Shot 3 — The Realization
▶
Play Clip
Shot 03
The Realization
Extreme Close-Up Slow push-in
"The mountain calls. The ritual begins."
Shot 4 — The Decision
▶
Play Clip
Shot 04
The Decision
Medium-Wide Slow pull back
"I carry what the village cannot."
Scene 02
The Forest Path
0:28 – 0:55
Location B · Night
Shot 5 — Into the Mist
▶
Play Clip
Shot 05
Into the Mist
Extreme Wide / High Angle Slow push-in
"Through the sleeping forest…"
Shot 6 — The Walk
▶
Play Clip
Shot 06
The Walk
Full Shot / Tracking Slow dolly forward
"…where even the trees hold their breath."
Shot 7 — Against the Wind
▶
Play Clip
Shot 07
Against the Wind
Medium Shot Static push-in
"The wind tests my resolve."
Shot 8 — The Flame
▶
Play Clip
Shot 08
The Flame
Extreme Close-Up Slow push-in
"I guard the ember with both hands."
Scene 03
The Summit
0:56 – 1:27
Location C · Dawn
Shot 9 — The Ascent
▶
Play Clip
Shot 09
The Ascent
Extreme Wide Slow push-in
"The summit waits above the clouds."
Shot 10 — The Stone Marker
▶
Play Clip
Shot 10
The Stone Marker
Medium Shot Slow push-in
"Here, the old world meets the new."
Shot 11 — The First Light
▶
Play Clip
Shot 11
The First Light
Close-Up Slow push-in on lantern
"The dawn gives what the night has taken."
Shot 12 — The Dawn
▶
Play Clip
Shot 12
The Dawn
Wide Shot Slow pan left-to-right
"The flame is renewed."
Scene 04
The Return
1:28 – end
Location D · Morning
Shot 13 — The Temple Reborn
▶
Play Clip
Shot 13
The Temple Reborn
Wide Shot Slow push-in
"Home breathes again."
Shot 14 — The Threshold
▶
Play Clip
Shot 14
The Threshold
Medium Shot Slow push-in
"I return with the light."
Shot 15 — The Offering
▶
Play Clip
Shot 15
The Offering
Medium-Close Slow push-in
"The cycle is complete."
Shot 16 — The Keeper
▶
Play Clip
Shot 16
The Keeper
Extreme Close-Up Static hold
"Sleep well, my village. I am here."
08 — The Craft

Prompt Engineering

The Formula
Subject +
Location +
Lighting +
Camera +
Mood
Every prompt in this project followed this exact structure — no exceptions.
Example Prompt
"Young female temple keeper age 22, rust-colored wool hooded cloak with worn leather edges, ancient stone temple interior, cool blue moonlight and warm ember glow mixing on her face, medium shot waist up, shallow depth of field, photorealistic cinematic fantasy film still, 8K, film grain, shot on Arri Alexa 35mm"
Subject Location Lighting Camera Mood
Consistency Keywords — used in every single shot
"rust-colored wool hooded cloak with worn leather edges" "dark leather belt with brass buckle" "small brass lantern with glass panels" "photorealistic cinematic fantasy film" "8K, film grain, shot on Arri Alexa 35mm"
语
Chinese Prompts for Higgsfield

Discovered that Seedance 1.5 Pro responds better to Chinese prompts for video generation. All video prompts were translated and optimized for the model.

Working With AI Limitations

✕
Dolly back / Pull back / Zoom out

AI hallucinates areas outside the original frame

→
✓
Removed all pull-back motions. Replaced with push-in, pan, or static camera moves.
✕
Face warping in close-ups

AI struggles with facial consistency during motion

→
✓
Kept face shots minimal in movement. Judged face first before keeping any clip.
✕
Hand/finger morphing

AI distorts fingers during complex hand motions

→
✓
Kept hands stable. Let camera and environment do the motion work.
✕
Body transition between poses

Start+End frame risky with large pose changes

→
✓
Used Start+End frames only for subtle transitions. Avoided kneeling or complex pose changes.
Mid-Workflow Adaptation

Originally planned an "emergency rescue" story — a dying flame that needed saving. But discovered that fire never looked dying in generated clips. Rather than forcing a broken narrative, I rewrote the story to an "annual ritual renewal" — matching the actual visual output. This shows critical thinking and professional judgment.

The Voiceover

Platform
Luvvoice
Voice
Elara Wynn — Warm Reflective
Rate
−10% (meditative pacing)
Format
16 MP3 files, one per shot
Total Characters
~500
Style
Ancient, ritualistic, intimate. First-person.
09 — The Final Film

Watch the Complete Sequence

Runtime: ~112 seconds 16:9 · 1080p Voiceover + Audio generated with clips (Seedance audio ON) Hard Cuts
10 / Methodology

Directed,
Not Generated.

Every frame in this sequence was the result of deliberate, structured decision-making — not luck. AI was the brush. The direction, the story, the choices, the adaptation — those were human.

A prompt formula was applied to every single shot, no exceptions. When the AI broke, the workflow changed. When the narrative didn't match the output, the narrative was rewritten to serve the truth of the image.

That's not AI filmmaking. That's filmmaking with AI.

16
Shots — each with a unique structured prompt
12
Hours — concept to deployed website
1
Mid-workflow narrative pivot — when the AI told a different truth
0
Random generations accepted — every clip was judged, iterated, or replaced
Ready to collaborate?

Have a story
that needs telling?

Start a Project ← Back to Portfolio