BitShovel
中

AI video · Case study

Tradecraft: character correction, a four-shot prompt and a 61-second film

Watch the film, then trace the shared character file, environments, storyboard, twelve-second four-shot prompt and edit decisions.

Published · Reviewed

BitShovel editorial illustration of a maker correcting an overly muscular character design and planning four shots; not Tradecraft footage or an original process screenshot.Open the creator’s playerBitShovel illustration · not film footage
Play the maker's roughly 61-second complete film, then inspect the making-of video and original files below.Poster source: BitShovel AI editorial illustration, not film footage; maker's film and production files are linked in the case · Poster source · Watch original film
Documented outcome · independent short film61-second film, 10:34 breakdown and six shared process files

The work and part of its process are inspectable; no client order, income or profit is evidenced.

Check this result
Read the context and editorial assessment

Tradecraft began with Ryan Duggan's own fantasy-football story and became a short film. He did not send one long prompt straight to a video model: his Story Guru and First A.D. projects helped examine the script and shots, followed by characters, environments and still storyboards before Seedance 2.5 clips and editing. The public files include an actual twelve-second four-shot prompt, while the making-of retains a character image that turned Danny into an unintended muscular man. The useful question for a reader is how a mistake changed the character description, how images were prepared before motion and how editorial choices made the film work.

Follow the projectContents8
01

Verified material

Start with the 61-second film

The maker's channel provides both the complete film and a separate process breakdown.

Tradecraft is a complete film of roughly 61 seconds that turns a fantasy-football trade between brothers into a spy joke. At the end, a child catches Uncle Danny at Dad's computer again. A separate 10:34 breakdown walks through conception, characters, environments, boards, generation, audio and spend. Both YouTube videos are credited to Ryan Duggan. Seeing the film first helps separate the final images from decisions shown in the process account.

This is a maker-led personal project. The public record establishes a finished film and documented production steps, but includes no client brief, campaign dashboard, contract or settlement. Visual polish does not establish a sale or profit.

Sources for this section: Ryan Duggan: complete Tradecraft film · Ryan Duggan: How I made Tradecraft · Ryan Duggan's original post about the personal short

02

Public account

The maker wrote the story; AI helped examine it and plan shots

Story Guru, First A.D. and Hold State served story review, shot planning and record-keeping.

Duggan wrote a spy-flavoured joke about a fantasy-football trade with his brother Danny, then used two Claude projects he built: Story Guru to challenge story logic and First A.D. to organize filmable shots. Their instructions are in the shared folder. The projects did not decide why he wanted to tell the story; he judged which shots served it.

He used his Hold State app to retain prompts, model choices and timing, with assets organized by character and setting. The making-of shows clips and shots prearranged before final video generation. That record makes it possible to revisit a specific character, set or shot instead of guessing among unnamed outputs. The shared pack shows the method, not every generation or his personalized Midjourney style.

Sources for this section: Ryan Duggan: How I made Tradecraft · Maker's six shared production files · Shared pack README and usage note · Ryan Duggan's original post about the personal short

03

Public account

A likeness cue turned an ordinary character into a muscular man

The wrong image appears in the breakdown; the character file documents removing the cue and describing an ordinary middle-aged face.

The character file says a Jason Statham-style head-shape cue led Midjourney to produce a batch of overly muscular versions of Danny. The making-of shows the unwanted look at about 4:33. The problem was not insufficient resemblance: the celebrity cue shifted body and tone away from the ordinary middle-aged brother needed for this story.

Rather than piling restrictions onto the celebrity comparison, he removed it and specified an ordinary face, normal proportions, a close-fitting dark cap on a balding head and short scruffy stubble. The file gives before and revised descriptions and the video shows the bad image. There is no fixed seed, complete generation log or controlled pair, so this is not a one-word A/B proof. Selection and later image work also shaped the final character.

Documented process

A traceable character-description correction

The maker's file records the wording change, his breakdown shows the failed look and the finished film shows the chosen character. Each source answers a different question.

  1. Earlier wording · celebrity head-shape cue

    The character file records the cue and the resulting mismatch with an ordinary brother.

    Read the original character file

  2. Wrong look · around 4:33 in the breakdown

    The video shows the overly muscular result; BitShovel did not recreate it.

    See the maker's failed image

  3. Revised wording · remove the analogy

    Directly describes an ordinary middle-aged face, cap and stubble before the maker selects a usable look.

    Check the revised description

  4. Final character · see the selected look

    The film shows the selected character, not a single controlled after image.

    Return to the film

Evidence limitThe maker's video and file document this correction, not a universal causal effect from one phrase. His personalized Midjourney style and complete reference set are not shared.

Sources for this section: Original Danny character prompt file · Ryan Duggan: How I made Tradecraft · Ryan Duggan: complete Tradecraft film

04

Public account

Build settings and still frames before motion

Location colours drifted, so the maker corrected them in Photoshop and tested composites.

Duggan generated unoccupied settings in Midjourney first. The breakdown and environment file report inconsistent colour across views of one world; he corrected it in Photoshop, then used image tools in Weavy to place characters into the adjusted locations. A felt-like character texture did not carry into Seedance motion, so he accepted a smoother animated look and adjusted later style wording to make the film coherent.

The making-of shows a still storyboard around 5:41; he checked whether images and shots would join before spending on motion. He also reports missing hands or arms and a grey character-card background leaking into a scene, then adding constraints. The public record does not map every bad shot to complete inputs, repaired clips and retry counts, so these remain the maker's account rather than a constructed A/B comparison.

Sources for this section: Original environment prompt file · Ryan Duggan: How I made Tradecraft · Original Danny character prompt file

05

Verified material

An actual twelve-second, four-shot prompt

The shared file separates identity, setting, cast size, timing, cuts and audio.

The maker shares a Seedance 2.5 prompt used for an opening section of about twelve seconds. It establishes a 16:9 three-dimensional toy-like look, asks for sound effects without generated music and keeps exactly two people on screen. @Image1 and @Image2 anchor character identity; @Image3 and @Image4 anchor the exterior and porch. Grey backgrounds from character cards are not meant to become story locations. Assigning each reference a role is more precise than asking the model to use all images indiscriminately.

The file then divides the clip into four beats—0–3.5, 3.5–6.5, 6.5–8 and 8–12 seconds—with actions, hard cuts and camera boundaries, repeating the camera rule at the end. It is one multi-shot generation input, not the sole prompt for the 61-second film. Duggan says music sometimes appeared despite the no-score instruction. The file proves the instruction was given, not that every output obeyed it.

Documented process

From reference roles to one four-shot generation

Open the maker's complete prompt to inspect one opening segment. BitShovel explains its structure without reproducing the full prompt.

  1. Character references · @Image1/2

    Anchor the two identities without importing the character-card backgrounds.

    Inspect the reference roles

  2. Setting references · @Image3/4

    Assign exterior and porch references and constrain the cast.

    Read the setting constraints

  3. Four time windows · action and cuts

    Four windows carry distinct shot tasks, with camera limits stated within and after them.

    Read the complete four-shot prompt

  4. Opening edit · inspect its use

    Inspect the final opening edit; the film is not necessarily an untouched generation.

    Watch the film's opening

Evidence limitThe file gives one actual input, not every retry, all reference images or a controlled prompt A/B. Editing and sound followed, and the twelve-second prompt cannot account for the whole 61-second film.

Sources for this section: Original Seedance 2.5 four-shot opening prompt · Ryan Duggan: How I made Tradecraft · Ryan Duggan: complete Tradecraft film

06

Public account

Generation made footage; the director chose what stayed

The maker reports 23 saved video generations among 155 saved outputs; several scenes were cut for pace.

The maker says five working sittings across three days produced 155 saved generations, 23 of them videos. That is not 155 video runs or 155 accepted shots. His breakdown shows prearranged shots, selected material and finishing; roughly three or four scenes were cut for pace. The final 61 seconds reflect those decisions, not one long-form generation command.

Video clips were chiefly generated through Seedance 2.5 in Weavy; the maker also names Topaz for moving 720p footage to 1080p, Photoshop and After Effects for finishing, and ElevenLabs for audio. These tools served different stages rather than each producing the whole film. The six-file pack is not a complete edit project or audio-mix archive, so it shows division of work and order, not a one-click way to reproduce the film.

YouTube thumbnail for Ryan Duggan's roughly 10:34 making-of video.Open the creator’s player
The maker walks through script, characters, boards, generation, audio and credit use; play it to jump by chapter. · Ryan Duggan's original YouTube video and thumbnail; not hosted by BitShovel · Original material

Sources for this section: Ryan Duggan: How I made Tradecraft · Shared pack README and usage note

07

Public account

“Under $50” covers Weavy credits only

An existing Midjourney subscription and other tool and labour costs are not fully accounted for; shared files are not a media republication licence.

Duggan says the Weavy credits cost less than $50 on top of the Midjourney subscription he already had. That describes one part of generation spend: other tools, existing subscriptions, labour and finishing are not fully itemized. No client contract, platform payout, income or profit is published. Under $50 is neither the total project cost nor evidence of earnings.

The maker shares six process files without signup, including instructions for his two Claude projects, character and setting descriptions, a Seedance shot structure and a README. The README invites learning, use and modification but provides no standard open-source licence or separate permission for BitShovel to copy the film, character images or complete prompt files. We link to the originals and explain them in our own words. A new project should begin with material one may use and keep records of spend, failures and selected shots.

Sources for this section: Ryan Duggan: How I made Tradecraft · Maker's six shared production files · Shared pack README and usage note

A BitShovel exercise · not the maker’s tutorial

Make your own four-shot microfilm

BitShovel designed this small trial from Tradecraft's published character correction, reference-image roles and four-shot opening. Tell an original moment of discovery with one person, one location and one prop in roughly 10–12 seconds. It is not Ryan Duggan's complete workflow or prompt.

Before you begin

  • Use a short story you wrote and may use, plus your own or properly licensed character, empty-location and prop references. Do not reuse the film's characters or images as inputs.
  • Have still-image, short-video and basic editing tools. Choose an aspect ratio and an affordable time and spend limit; cap motion attempts at two per shot needing repair.

  1. Define the turn the viewer should understand

    Write one sentence stating who discovers what and what they do next. Break it into four shots: place, approach to the prop, discovery and reaction. Give each shot one visible action and fit the sequence into roughly 10–12 seconds.

    Keep
    A one-page story card with the premise, four shot tasks, aspect ratio and target duration.
    Check
    Without an explanatory caption, do the four actions convey the discovery and reaction? Revise the story card if they do not.

  2. Give each reference image a single job

    Make a clean character reference for your own protagonist, plus an empty-location image and a prop image. Number them and label their jobs: identity, setting or prop. Keep the character-card background out of the story location.

    Keep
    Three numbered reference images and a table of their intended roles.
    Check
    Are the provenance and usage rights clear? Can the person, place and prop be identified separately without conflicting backgrounds or lighting?

  3. Correct character drift in stills first

    Generate a few character stills using direct traits such as age, face, clothing and build, without a celebrity-likeness shortcut. Select one and record specific reasons for rejecting the others. If body or tone drifts, remove a misleading analogy, revise one concrete trait and try once more.

    Keep
    Selected and rejected stills, before-and-after character descriptions and an error note.
    Check
    Does the selected person fit the story? Do not accept a pretty image that misses age, build or clothes. Stop at the still stage if the revision fails.

  4. Board the four shots before motion

    Place the chosen character, setting and prop into four still frames. For each, note start, end action, camera direction, person position and prop state. Change between frames only what the story requires.

    Keep
    A four-frame board and a person-and-prop continuity sheet.
    Check
    Do the stills tell the event in order? Did the prop change hands, the person jump position or the setting abruptly shift colour? Fix those before spending on motion.

  5. Generate, log and reject bad takes

    For each shot state the identity and setting references, visible action, duration, camera and sound requirements. Use a combined prompt only if the tool supports it; otherwise generate one shot at a time. Save tool, date, input, candidate clip and rejection reason, within the two-attempt cap.

    Keep
    Usable clips, rejected candidates and a shot-by-shot error log.
    Check
    Check cast count, hands, identity, background, prop and unwanted music. If two attempts fail, simplify the action or rebuild the stills instead of generating indefinitely.

  6. Edit, then ask what viewers understood

    Keep only the clips that serve discovery and reaction, align format and sound, and cut roughly 10–12 seconds. Show it to two people who do not know the story; ask what happened and where they became confused. Record their words and the timestamp. Plan channel and cost tests separately before any public release.

    Keep
    A first cut, time-and-cost log, two viewer notes and one specific revision decision.
    Check
    Can both viewers describe the turn, without an obvious visual or audio break? Understanding by two viewers is a clarity check, not evidence of reach or income.

Make one final choice: Do not start motion until the stills work. At the attempt or budget limit, simplify the action or stop and review. If the story is clear and the technical check passes, consider a small release, recording platform response, spend and any income separately.

This is a BitShovel reader exercise, not the Tradecraft maker's original workflow, prompt or cost reproduction. His published four-shot input and wrong-character image inform the lesson; verify your own tools, assets, rights and results.

Next, compare three storyboard routes in another film

Both make choices before motion. This case traces character drift and stills into an actual four-shot input; The Final Plate traces three boards into a shot's anchor frames, clip and edit. Their records cannot stand in for each other.

Sources and limits

Where the material comes from