How I turned a poem into a 3-minute short film in one day using AI.

Love Is Lost — Creative Experiment #002

Opening Sparks • The Seed Poem

The spark for this short film came from a poem Emma shared with me in 2021. Its emotional depth — the mix of loss, resolve, and reflection — demanded translation into visual and auditory storytelling. Here’s the original poem:

Love Is Lost — Emma (2021)
1 year ago you left me
Without a single word
Now you’re back here begging
Hoping to be heard

12 months I’ve lived without you
I’ve been doing just fine
We are through
I don’t want you to be mine

52 weeks without a sound
Then you come crawling back to me
But I don’t want you around
You’re so dense, why can’t you see

365 days it took you to see
You lost someone who cared
Finally you realize you truly wanted me
As these lies leave your lips, my face stone cold, I stared

Lab note: The poem’s cadence and raw emotion immediately suggested voiceover narration, layered with music and visuals — a blueprint for emotional pacing.

Tools & Trinkets • Software and AI Stack

This was a low-budget experiment, but it relied on multiple AI and open-source tools:

  • Poem: Emma
  • Narration voice: ElevenLabs — Francesca Segretto (Cosmopolitan Woman — High Quality Voice) with emotional audio tags (1:10 minutes)
  • Male dialogue voice: ElevenLabs — Matthew Schmitz (Anti-Hero / Villain / Rogue / Tough Guy) (15 seconds)
  • Music generation: Producer.ai — three music segments, prompts and lyrics written by ChatGPT
  • Image generation: Midjourney (208 images in Relaxed mode)
  • Image-to-video conversion: Grok Imagine — simple prompt: slow pan (46 clips – 6-seconds each)
  • Lip-sync voice-over clip: WAN Video (~10 sec, cost 50¢)
  • Video editing: Kdenlive
  • Audio cleanup: Audacity
  • Operating system: Linux Mint (three-monitor desktop setup)

Lab note: Generating images in Relaxed mode meant slower rendering but unlimited experimentation — crucial for budgetary control.

Mixing & Matching • Music & Narration

The narration and music were produced as three separate segments:

  1. Intro — voiceover reading the poem
  2. Instrumental break with sung lyrics (AI was permitted to stretch, echo, fragment lines for emotional emphasis)
  3. Outro — poem narration closing the story

All clips were eventually synced in Kdenlive. I slowed down interior shots to 90% to fit the timing and create a subtle slow-motion effect. Romantic memory sequences were modified with alpha masking for partial transparency, faded in and out, giving a dreamy feel.

Lab note: Even minor timing tweaks created emotional lift. Slowing a clip 10% can turn a “nice” shot into something poetic. All the interior b-roll clips were composited together and treated as one clip.

Image Layering • Visual Storytelling

The visual strategy used two complementary layers:

  • Symbolic objects: books, fidget toys (a Rubik’s cube subtly on a table), hinting at personality without faces
  • Romantic memories: implied couple gestures — hand-holding, hugs, clothing on the floor — Caucasian, mid-twenties, male and female, lighting consistent across shots, soft golden interior glow, cinematic style
Sample of Midjourney AI images used in the short film.

All Midjourney images were converted into 6-second clips using Grok Imagine. This consistent timing allowed smooth pacing across the three-minute runtime.

Lab note: Constraints breed creativity. The 6-second limit forced careful shot selection and storytelling economy.

Iteration & Discovery • Editing the Story

Every component — narration, music, video clips — went through multiple iterations:

  • Music: 12 iterations in Producer.ai
  • Romantic imagery: alpha masking and fade-ins
  • Dialogue: ElevenLabs emotional tags tuned in Audacity
  • Male character lines: “Wait… what?” and “You’re serious?” tested for perfect timing

The video opens subtly with a phone notification SFX and him saying “Hi… it’s me,” followed by her poem reading. The story progresses, culminating in the door slam and him walking away dejected. Memories fade, closure arrives, and the narrative resolves visually and aurally.

Lab note: Iteration and improvisation wasn’t optional — it was the engine of discovery. Small tweaks shaped pacing, emotion, and clarity.

Time in the Lab • Duration & Workflow

  • Total Midjourney images: 208
  • Working hours: 6–8 hours (with long breaks)
  • Completion: fully online within 24 hours

Constraints encouraged bold creative decisions, rather than polishing endlessly. Quick decisions became serendipitous discoveries.

What Did It Cost? • Budget Breakdown

  • Midjourney: $30/month (Relaxed mode, unlimited images)
  • Grok Imagine: free, with daily limits; workaround via multiple accounts and browsers
  • WAN Video: ~$0.50 per lip-sync clip (We had one clip).
  • ElevenLabs: $5/month
  • Producer.ai: $8/month for 600 iterations
  • Kdenlive & Audacity: free, open source

Lab note: The most expensive element wasn’t the software — it was time spent iterating and experimenting.

Image-to-video conversions were done in Grok Imagine for free.

Closing Reflections

Creating Love Is Lost was more than turning a poem into a video. It was a lesson in iteration, constraint, and discovery. Each tool, each clip, each tweak became part of a delicate alchemy, culminating in a short film that exists because of process, not perfection.

Sometimes, if you work quickly and embrace the unknown, a poem written in the morning can become a short film by the evening.

— Steve Teare
Master Alchemist