BitShovel
中

AI video · Case study

The Final Plate: choosing a storyboard direction and finishing a short film

Open the finished film, compare three real storyboard directions, then trace the falling-dish shot from anchor frames and video prompt through the edit.

Published · Reviewed

BitShovel editorial diagram for The Final Plate: one of three storyboard directions leads to six shots and a final edit; not a film frame.View original work ↗
Open the maker's roughly 80-second final film; GitHub may require View raw.Poster source: BitShovel editorial artwork; open the maker's files for original footage · Poster source · View original work
Documented outcome · maker's projectOne finished film, six shot clips and three storyboard directions

These files establish a finished project and its process, not audience size, commission, receipts or profit.

Check this result
Read the context and editorial assessment

A cook enters a competition kitchen, loses a finished dish, improvises a replacement and waits for the judges. This is Samin12's public short-film project, not an official MasterChef production or a verified money-making venture. Its value lies in the unusually traceable chain of three storyboard directions, selected frames, shot files and the final assembly.

Follow the projectContents6
01

Verified material

Start with the finished film

The original repository provides a final edit, individual shots and the maker's notes.

The final file, story-attempt-1-final-smoothed.mp4, runs about 80 seconds and includes a live-action introduction before the cook-off story. A separate nine-second intro-to-contest seam is not the full film. GitHub may not preview the larger final file inline; View raw opens the original.

The approved brief uses a pink cup as a recurring prop and says the film borrows a cook-off genre without using a show's name or trademark. The film and project files establish a completed piece; the repository offers no audience metrics, client contract or revenue ledger.

Sources for this section: Samin12: The Final Plate project overview · Complete final film file · Approved creative brief

02

Verified material

Three storyboard directions: which one was selected?

Three contact sheets explore the same project; the maker selected the first as the working direction.

Attempt 1 is the selected storyboard direction; Attempt 2 explores closer, more physical framing, and Attempt 3 wider, cleaner hero compositions. The approved files follow Attempt 1, whose contact sheet matches the approved version. Open the three original sheets to compare framing and emphasis.

These are creative alternatives, not three failed video generations. The maker does not publish a point-by-point reason for choosing the first direction or the settings behind every image. A useful practice is to compare visual routes against one story goal before generating long clips, while keeping our visual reading separate from the maker's undisclosed reasons.

Documented process

Three visual directions for one film

The maker's own storyboard alternatives and selection record; open the originals to compare them.

  1. Attempt 1 · selected

    The broadcast-style route that became the approved working set.

    Open Attempt 1 image

  2. Attempt 2 · unselected

    The maker describes closer, more physical framing; no shot-by-shot rejection reasons are given.

    Open Attempt 2 image

  3. Attempt 3 · unselected

    The maker describes wider, cleaner hero compositions; no shot-by-shot rejection reasons are given.

    Open Attempt 3 image

Evidence limitOriginal images remain linked at the maker's repository. Separate permission for media reuse, reference photos and likenesses is not documented, so BitShovel does not copy or embed them.

Sources for this section: Attempt 1 storyboard and notes · Attempt 2 storyboard and notes · Attempt 3 storyboard and notes · Maker's storyboard approval log

03

Verified material

How one shot moves from approved frames to video

Trace Shot 04's falling dish through approved frames, anchors, prompt and actual clip.

The maker uses the completed dish in F4 and its fall in F5 as Shot 04's start and end anchors, with a video prompt and a roughly twelve-second clip. The fall is a story beat, not a model failure. Following this one shot shows what the still frames constrain and what the final edit must connect.

The project's shot overview names Seedance 2.0 and a six-shot first/last-frame chain; its reference plan names Nano Banana for character, kitchen and dish assets. It does not publish complete model logs, so exact settings, retry counts and costs remain unknown. Other tools listed in the root README are options, not proven tools used in this film.

Documented process

Shot 04 · the falling dish

Original inputs and output for one shot, each linked to the maker's file.

  1. F4 · approved start frame

    The completed dish before the movement begins.

    Open F4 image

  2. Shot anchor frames

    The shot folder pairs start and end images as the movement's constraints.

    Inspect anchor-frame files

  3. Shot-specific video prompt

    The written prompt describes the dish's fall and visual change.

    Read the maker's prompt

  4. F5 · approved end frame

    The fallen dish as the visual endpoint of the shot.

    Open F5 image

  5. Actual twelve-second clip

    The maker's Shot 04 video file, later assembled with the other shots.

    Open the Shot 04 clip

Evidence limitThe selected inputs and output can be checked; rejected video takes, every revision and complete generation settings are not public.

Sources for this section: Shot 04 approved frames and anchors · Shot 04 video prompt · Shot 04 video file · Six-shot video prompt overview · Visual reference plan

04

Public account

The actual fixes: frame rate, sound and identity

The maker's fix notes are specific, although no failed-and-revised video pair is published.

The stitching notes document mismatched resolution, frame rate and audio between the live-action intro and generated transition. The maker normalised them to 1280×720, 24fps and 44.1kHz stereo and used a white flash to mask the seam. Finishing choices, not just generated shots, determine whether viewers notice a break.

An identity comparison note says the protagonist's facial match was soft in some frames, prompting extra identity-lock references. Comparison images are present, but the original prompt, retry count and decision for each image are absent. This is not a documented bad-prompt/good-prompt experiment.

Sources for this section: Stitching and finishing notes · Identity comparison notes

05

Our analysis

What a reader can borrow—and what remains unproven

The reusable part is the documented decision and shot handoff, not a market-return claim.

A small trial could start with one story turn and one shot: state the goal, make two or three framing options, record the selection, then keep start and end references, video prompt, output and actual finishing problems. That record makes an unsuccessful result diagnosable at the right stage.

This project does not establish views, customer demand or earnings. The repository has an MIT code licence but no separate statement covering reuse of every generated film asset, live-action reference, likeness or music element. Studying the structure differs from reusing the media, identity or a cooking-show brand; check your own rights first.

Sources for this section: Samin12: The Final Plate project overview · Repository MIT licence · Approved creative brief

A BitShovel exercise · not the maker’s tutorial

From three boards to one turning-point shot

BitShovel designed this original trial from The Final Plate's three published storyboard routes, Shot 04 anchor frames and stitching notes. Build a roughly 10–15-second film around an unexpected change to your own prop, then see whether viewers follow the turn. It is not Samin12's workflow or prompt.

Before you begin

  • Bring your own one-sentence story, a person and prop you photographed or may use, and a usable setting. Do not copy the maker's pink-cup images, live-action likeness or any show branding.
  • Have tools for still frames, short motion and editing. Set a cap of two motion attempts and a time and spend limit you can afford.

  1. Choose a visible story change

    In one sentence describe the original goal, the prop mishap and the person's recovery. Limit the event to one person, one setting and your own prop; mark when the viewer should understand the change.

    Keep
    A one-page original brief with the person's goal, mishap, recovery, aspect ratio and target duration.
    Check
    The mishap is a story beat, not a desired generation error. Without an explanatory caption, could someone see that the situation changed?

  2. Board three framings for the same story

    Draw or generate before-and-after stills in three small boards: A preserves the space, B moves closer to the action and C gives the recovery more room. Hold person, prop and story constant; note what each route makes easier to see.

    Keep
    One labelled A/B/C board comparison and a written reason for the choice.
    Check
    Can a viewer identify the prop, action and cause from the frames? Select the clearest route for the story; redraw if none works before generating video.

  3. Lock the chosen shot's start and end

    Take the before and after stills for the chosen action. Mark face, clothing, prop position, screen direction and light. Change only the story state that needs to change, and confirm your materials may be used.

    Keep
    Numbered start and end frames plus a continuity sheet.
    Check
    Can the same person, prop and camera direction plausibly connect the frames? Fix unexplained identity, prop or light shifts first.

  4. Try one action shot and log failure

    Use your own anchor frames in a tool that accepts them. Describe one visible action, duration, camera limits and any extra people or props to avoid. Try at most twice; save the input, settings, output and reason for rejection rather than pasting the maker's Shot 04 prompt into your story.

    Keep
    One candidate motion shot, its input and a log of no more than two attempts.
    Check
    Check action, identity, prop and endpoint frame by frame. If the face drifts or motion fails, simplify the action or rebuild the anchors instead of generating more takes.

  5. Edit a complete segment and inspect the seam

    Join the turning-point clip with your own setup and reaction, cutting to roughly 10–15 seconds. Align aspect ratio, frame rate and sound level to this project's target. Inspect the edit frame by frame; move a cut or add a short transition if useful rather than treating the maker's white flash as a universal fix.

    Keep
    A short film with setup, turn and response, plus an edit-issue list.
    Check
    Are there black frames, audio pops, jumps in person position or contradictory prop states? Match format to your destination rather than copying the maker's exact settings.

  6. Let fresh viewers guide the next version

    Show the first cut to two people unfamiliar with the story. Ask what changed and where the film felt discontinuous; record their words and timestamps. If publishing, separately record platform, date, actual audience and engagement measures, and promotion spend. Two viewers do not validate a market.

    Keep
    Feedback notes and a decision to change one specific point next; if released, a separate source-backed response log.
    Check
    Can both explain the mishap and recovery without flagging the same break? If not, return to the relevant board, anchor or edit. Stop and review when the budget cap is reached.

Make one final choice: If none of the three boards conveys the turn, do not buy more video generation. If two attempts still lose the person or prop, simplify the action. After story and edit checks pass, plan a small release test, keeping reach, feedback, spend and actual receipts separate.

This BitShovel exercise draws on public project files; it is not Samin12's original procedure or prompt. The repository documents three boards, approved frames, Shot 04 and stitching notes, not every retry, audience response or business account. Your result cannot inherit a market claim from the film.

Compare character correction in another film

This project preserves board selection, anchor frames and a clip, but facial mismatch appears mainly in fix notes. Tradecraft also shows the maker's wrong character image and revised wording. Together they inform identity checks before motion, without becoming one project's generation log.

Compare another open drama project

ZenStory has traceable text, checks and a 24-second sample; this project adds visible storyboard alternatives and six shot clips. Neither stands in for the other's workflow.

Sources and limits

Where the material comes from