AI video · Case study
Hat Trick: what six days behind a ten-minute AI film involved
A maker with media-production experience spent about 72 hours and $390 in tools on this film. Watch the result, then trace script, generation, voice and manual editing. Rework in continuity, pronunciation and on-screen text is documented in his account; failed clips are not public.
Published · Reviewed
Viewable result · independent short filmAn approximately ten-minute film and maker's production account
The maker presents the film as evidence of learning AI production. Roughly $390 is his self-reported tool spend, not revenue, profit or full production cost; no order or payment is documented.
Check this resultRead the context and editorial assessment
Geoff Nelson made the film to move his existing media-production skills into AI tools. A media veteran loses his job, learns AI and rebuilds an identity across five acts with different visual styles. The film is viewable, and its maker documents six days of generation, voice and editing problems. Hat Trick helps answer how much human judgment a finished AI film still takes; it does not establish that making one earned money.
01
Verified material
Watch the film: what counts as the result?
An approximately ten-minute, five-act narrative is finished and public on the maker's page and Vimeo.
Hat Trick follows a media professional through job loss, AI learning, five invented cartoon heroes and a new chapter. The maker uses different visual languages across its five acts, including live-action-like imagery, cinematic framing, animation and mixed media. Watch first for recognizable characters across shots and how sound meets picture, then use the production account to trace the decisions behind them.
The public record establishes a finished film and the maker's account. The Vimeo entry is titled Hat Trick Short Film V2 and lasts about 9 minutes 47 seconds; the maker rounds it to ten minutes. The page supplies no exhibition contract, platform settlement or paid-viewing record.
Sources for this section: Geoff Nelson: firsthand Hat Trick production account · Geoff Nelson: original Hat Trick Short Film V2
02
Public account
Six days and about 72 hours: from five acts to shots
Script, shot breakdown and character references preceded the costlier motion work.
The maker reports six roughly twelve-hour production days. On day one, he used Claude to develop a five-act script and edited its voice and comic timing himself. Across days one and two, he split the acts into 15–30-second scenes with camera, lighting, placement and style instructions. On days two and three, he made 36 images in ChatGPT and Midjourney for references and character design. That is a count of generated images, not 36 finished shots.
Across days three to five, Runway was his primary motion tool and Pika supplemented animation tests; he rendered at least 55 clips in Runway alone. The 55+ is a render or clip count, not 55 distinct final shots or a success rate. Voice, manual editing and graphics filled the final two days. Script, recognizable characters and final rhythm still depended on his decisions across tools.
Documented process
The maker's six-day sequence
Compare the maker's timetable with the film. These are stages of one project, not an archive of every raw iteration.
Day 1 · story
Claude helped structure five acts and dialogue; the maker reshaped voice and comic timing.
Read the day-one accountDays 1–3 · shots and references
After shot breakdowns, he generated 36 reference and character-design images for five heroes.
Read the shot and image stagesDays 3–5 · video and rework
Runway rendered at least 55 clips; continuity and generated audio added rework.
Read the generation stageDays 5–6 · sound and edit
He used ElevenLabs for voice, LALAL.AI to remove unwanted music, then assembled and typeset the film manually in an NLE.
Return to the finished film
Evidence limitThe film shows the final result; hours, counts and sequence come from the maker's account. The public materials do not include all 55 raw renders, a complete inventory of 36 images, same-shot failed and corrected moving clips or reproducible generation settings, so they cannot establish a success rate.
Sources for this section: Geoff Nelson: firsthand Hat Trick production account · Geoff Nelson: original Hat Trick Short Film V2
03
Public account
Continuity, pronunciation, audio and text caused rework
Four problems called for different responses: re-rendering, voice changes, music separation and manual graphics.
Continuity was the maker's hardest visual problem: separate 15–30-second renders could differ even with the same references in the same tool. He made multiple references for five heroes, checked each clip and chose what to redo. The page does not publish moving before-and-after versions of a specific shot, so this is not a controlled A/B test.
Audio needed separate work. The maker says Runway clips arrived with unwanted music and sometimes combined dialogue with voice-over, so he used LALAL.AI to separate music and ElevenLabs for narration and characters. Geoff was mispronounced; spelling it phonetically as Jeff in voice instructions prevented that problem after costly voice re-renders. The account gives a concrete failure and response, but not the bad audio files or per-take costs.
Generated on-screen text also could not reliably spell words, so the maker built titles and lower thirds manually in his editor. Generated clips did not always leave enough head or tail for transitions, requiring either earlier shot planning or edit changes. The film shows selected footage after voice, mix and manual typography; none of that is captured by the 55+ video-render count.
Sources for this section: Geoff Nelson: firsthand Hat Trick production account · Geoff Nelson: original Hat Trick Short Film V2
04
Public account
What the roughly $390 covers and what it leaves out
The maker reports about $90 in subscriptions and $300 in credits; invoices, labor valuation and revenue evidence are absent.
The maker estimates approximately $390 spent: about $90 in subscriptions and $300 in generation credits and tokens across tools including Runway, Midjourney, ChatGPT, ElevenLabs and LALAL.AI. He also reports around 72 hours of labor. The breakdown shows why re-renders matter financially, but the page offers no individual invoices, subscription allocation or shot-level usage, and labor is not priced into that figure.
The film's economic role is closer to a portfolio piece and proof of learning: one finished project demonstrates a move from existing production skills into new tools. The maker does not document a resulting client job or screening revenue, and the public record has no viewing payout, customer payment or net profit. Dividing $390 by ten minutes does not produce a reusable per-minute quote or evidence of earnings in six days.
Sources for this section: Geoff Nelson: firsthand Hat Trick production account
05
Our analysis
For a first attempt, test continuity across two shots
Control story, character, frame and audio task first; record usable footage and rework cost.
Five acts, five heroes and at least 55 Runway renders can make the finished film's scale look like a starting point. A more useful first test is whether two shots tell one story: recognizable character, intelligible sound and controlled typography. The reader exercise above turns those checks into evidence you can keep; it is BitShovel's method, not the maker's published success formula.
Before buying several subscriptions, check billing periods and export terms; specify name pronunciation before generating voice; decide who will typeset screen text before motion renders. The maker makes the film and account publicly viewable, which does not grant permission to republish footage, thumbnail, likenesses or music. BitShovel embeds and links to his original media only.
Sources for this section: Geoff Nelson: firsthand Hat Trick production account · Geoff Nelson: original Hat Trick Short Film V2
A BitShovel exercise · not the maker’s tutorial
Try a two-shot story that cuts together
The maker spent substantial time on character changes across shots, voice and titles. BitShovel turns those lessons into a bounded continuity test, rather than copying his six-day workload.
Before you begin
- Write an original short story and prepare character, voice and references you may use; set frame, runtime and a spending cap first.
- Use one character, one location and two consecutive actions; there is no need to subscribe to every tool the maker listed.
Break the story into two necessary shots
Write what shot one shows and what changes in shot two. Note character position, clothing, light and handles for the edit.
- Keep
- A two-shot list and one-sentence story turn.
- Check
- Removing either shot should lose an action or turn, rather than merely one attractive image.
Fix the character and repeatable conditions
Prepare usable front and profile references and record wardrobe, props, place and name pronunciation; keep the actual references and inputs used.
- Keep
- A one-page character-state sheet, references and input log.
- Check
- The same character and space remain recognizable in both candidate shots without relying on a title.
Generate separately and keep the rejects
Make short candidates separately within the cap, logging time, generation spend, inputs, character drift and usable seconds; retain at least one rejected take for comparison.
- Keep
- Candidates for both shots and a rework log.
- Check
- Compare face, wardrobe, screen direction and light; mark an attractive but uncuttable shot as rejected.
Handle voice and text separately
Place usable shots on a timeline, test pronunciation and audio with one voice line, and typeset title and captions in an editor rather than relying on generated screen text.
- Keep
- A two-shot rough cut with clear audio and manually typeset text.
- Check
- Audio does not obscure speech, the name is intelligible, captions are readable and the cut does not jar.
Show it and count the real input
Ask someone unfamiliar with the story to retell it and point out distracting identity or audio changes; total attempts, kept seconds, labor time and cash spend.
- Keep
- Feedback notes, a usable-footage and spend sheet, and a list of one or two changes for the next pass.
- Check
- Record finished sample, audience reach and actual income separately; this test establishes only your own output and input.
Make one final choice: If the character drifts across shots, simplify action, angle or styling. Stop at two shots when the cap is reached; extend into multiple acts only after viewers understand the cut and rework is affordable.
This is a BitShovel exercise based on the maker's account, not Hat Trick’s original prompts, shot-by-shot project files or a proven way to earn. Complete rejects and client payments are not public.
Sources and limits
Where the material comes from
- Geoff Nelson: firsthand Hat Trick production accountMaker's case studyChecked
The six-day timeline, tool roles, 55+ Runway renders, 36 generated images, rework and roughly $390 spend are self-reported. The page has no explicit publication date; the maker's sitemap lists 2026-06-15 as its last-modified date, which is not a publication date.
- Geoff Nelson: original Hat Trick Short Film V2Maker's finished Vimeo filmChecked
Original Vimeo film and thumbnail. Its oEmbed metadata gives a 587-second runtime and a 2026-05-28 upload date; the film itself cannot verify hours, costs or earnings.