AI Video Workflow: From Brief to Shot List to Final Cut

Most AI video projects do not fall apart at the render step. They fall apart much earlier — in a vague idea, an unstated visual reference, or a shot list that lives only in someone's head. A generator can produce a beautiful six-second clip; it cannot decide what a story needs. The distance between a folder of disconnected clips and a piece that genuinely feels cinematic is almost always workflow, not model choice. What follows is a repeatable production pipeline you can run on any project, from a ten-second social cut to a two-minute brand film or a music video, with Orelon as the generation surface.

Start With a Brief, Not a Prompt

A prompt is an instruction. A brief is a set of decisions. Write one page before you open a generator, so every later choice has something concrete to be measured against.

The five-line creative brief

• Logline. One sentence: who wants what, and what stands in the way. "A courier races a storm to deliver a single letter" is workable. "Cool cinematic video" is not.
• Tone references. Name two or three existing works and say precisely what you are borrowing — the palette of one, the camera movement of another, the pacing of a third.
• Visual rules. Palette, contrast, lens feel, grain, aspect ratio.
• Sound direction. Music, ambience or voice. Decide this before you animate a single frame, because it changes pacing choices.
• Deliverable spec. Runtime, aspect ratios, caption needs, and the platforms the piece has to fit.

Turn the brief into acceptance criteria

"Done" should be measurable before you generate anything: eight usable shots, one recognizable protagonist, no frame where the horizon breaks, no shot where hands are the subject. Write those as a checklist. Rejecting a clip in the first minute of review is far cheaper than rebuilding an edit around it later.

The brief also settles arguments quickly. When two people disagree about a shot, the useful question is not "which do you prefer" but "which one serves the logline and the visual rules". That conversation takes ninety seconds instead of twenty minutes.

Write down the audience and the screen

A teaser watched on a phone with the sound off has completely different demands from a landscape film played on a monitor. Shot size, contrast, text size and pacing all change. Choosing the destination first prevents a rebuild later, and it makes the safe-area decision obvious before any footage exists.

Build the Shot List Before You Generate

Shot lists are the highest-leverage document in AI video production, precisely because generation is the bottleneck. A shot list means you generate only what survives the edit instead of building an archive you will never open again.

Beat, shot, duration, purpose

Structure in beats, break beats into shots, then assign durations and a purpose to each one.

Beat	Shot	Duration	Purpose
Setup	Establishing wide, slow push	3s	Orient the viewer
Setup	Detail insert, hands	1.5s	Texture and scale
Turn	Medium shot, character reacts	2.5s	Emotional hinge
Payoff	Wide, movement resolves	3s	Release

Purpose is the column people skip and later regret. A shot without a purpose becomes filler in the edit and wasted time in generation. A thirty-second teaser usually lands at eight to twelve shots, averaging two to three seconds on screen. Anything past four seconds needs genuine internal motion to justify the time.

Naming and versioning discipline

Use a convention that survives a month of iteration, such as shot_03_storm_v2. Keep two folders: generated for raw output and selects for anything that survives a first pass. Version the edit rather than the clips — cut_v1, cut_v2, cut_v3 — so the timeline is always the source of truth.

Decide what should not be generated

Titles, lower thirds, overlays and simple product rotations are usually faster and cleaner in a design tool. Reserve generation for what is genuinely hard to shoot: weather, scale, impossible camera moves, period settings, and performances you cannot schedule. If you need stills first — character sheets, location plates, look frames — Create Image is a quick way to lock the look before you spend time on motion.

Budget the session, not just the shot

A shot list also tells you when to stop. Give each shot a maximum of three attempts, and give each generation session a fixed wall-clock limit. When the limit hits, you move to assembly with what you have. Sessions that end on a timer produce better selects than sessions that end when the folder is full.

Choose the Right Generation Path for Each Shot

Text to video

Best when a shot is about motion, atmosphere or environment and the subject does not need to match anything else: fog rolling over a ridge, traffic at dusk, a slow orbit around an object. Keep prompts concrete and short. Prompts with five competing clauses tend to produce muddled frames.

Image to video

Best when continuity matters. Supply or generate a keyframe, then animate it. Because the first frame is fixed, likeness, wardrobe and composition stay far more stable across a sequence. This is the default path for character-driven work and for any shot that has to connect to a neighbouring shot.

Hybrid sequences and the aspect-ratio decision

Most finished pieces mix both methods: image-to-video for hero shots where identity matters, text-to-video for inserts, atmospherics and transitions. Treat each shot as a separate decision rather than applying one method to the whole project.

Decide aspect ratios up front. If you need both 16:9 and 9:16, generate with a wide safe area or plan separate vertical framings for hero shots, because centre-cropping horizontal footage routinely cuts the subject or the horizon. Making that call after generation costs you your best takes, not your worst ones.

A three-question decision rule

Before generating any shot, ask: does identity need to hold across shots? Does the shot need to match a specific look frame? Is the shot purely atmosphere with no recurring character? The first two answers point to image-to-video; the third points to text-to-video. If the answers are mixed, split the shot into two simple ones.

When to change paths instead of rewriting

If a shot fails twice with one method, switch methods before you write a third prompt. Motion problems in a static subject usually mean the shot needs a fixed first frame. Identity problems in an animated clip usually mean it needed a reference image from the start. Switching paths is faster than refining wording.

Prompting With Structure Instead of Adjectives

A five-part formula

Subject, action, environment, camera, light and style — in that order.

"Salvage diver in a heavy canvas suit (subject) pushes open a rusted hatch (action) inside a flooded engine room (environment), slow handheld push-in with slight lens flare (camera), cold blue light from a single overhead work lamp, shallow depth of field, 35mm grain (light and style)."

Each part does one job, which means you can diagnose which part failed instead of rewriting the whole prompt.

Camera language that changes output

• Movement: slow push-in, dolly left, crane up, whip pan, static tripod, orbit.
• Framing: wide establishing shot, medium close-up, over-the-shoulder, low angle, top-down.
• Optics: shallow depth of field, wide-angle distortion, long-lens compression, anamorphic flare.
• Tempo: time-lapse, slow motion, real-time handheld.

One movement per shot. Two movements in one prompt usually produce a camera that does neither well.

Build a prompt library, not a prompt

Keep blocks that produced good results — lighting phrases, movement phrases, texture phrases — and reuse them instead of writing from scratch every session. Curated starting points are available in Prompts, and model-specific notes such as the Seedance 2.5 prompt page help when you want a particular look without reinventing the phrasing.

Define a stop condition

State what you do not want where the tool supports it: no text overlays, no watermarks, no extra limbs, no jump cuts. Then set a rule: three failed attempts on a shot means the shot is too complex, not that the prompt needs another adjective. Simplify it, split it in two, or switch generation paths.

Keep a failure log

For every shot that takes more than three attempts, write one line about what went wrong: "camera move too fast", "two characters in frame", "lighting direction ambiguous". Patterns appear within a week. Most creators discover that their recurring failures come from the same three habits — too many subjects, contradictory lighting, and camera moves that fight the action.

Continuity: References Beat Adjectives

Consistency is the hardest problem in AI video, and it is solved with references rather than adjectives.

Character sheets and reference frames

Build a character sheet before principal generation: three to five images of the same person from different angles, plus one neutral expression at the framing you will use most. Reuse those frames as the input for every shot featuring that character. When you need a variant, change one variable at a time — wardrobe, lighting, angle — and hold everything else fixed.

Name your reference files the way you name shots, so a collaborator can pick up the project without a call. char_lead_neutral_4k, char_lead_profile_left, char_lead_wardrobe_wet tells a whole story about what exists and what it is for.

Three continuity anchors

Pick three anchors and enforce them across every shot: a palette, a lens character, and a texture such as grain amount or contrast curve. Anchors do more for perceived production value than the fidelity of any single shot. When one clip looks sharper and cleaner than the rest, it reads as a different film — match it down rather than pushing the others up.

Write the anchors into the brief as literal values if you can: a hex range for the palette, a focal length, a grain percentage. Vagueness is what drift feeds on.

Catching drift early

A common failure is mid-project drift: someone generates a beautiful new shot in a slightly different style, adds it, and the sequence quietly splits in two. Review the full timeline at thumbnail size once a day. Mismatched shots reveal themselves instantly when the images are small and the cut is playing.

Location and prop continuity

Characters are not the only thing that drifts. A room changes wall colour, a car changes colour between shots, a coat changes length. Keep a simple continuity sheet with one line per recurring element and its approved reference frame. Ten minutes of bookkeeping saves an afternoon of regeneration.

Editing, Sound, and the Finishing Pass

Assemble for rhythm

Cut on motion, not stillness. If a clip ends with the camera settling, trim before the settle. Keep the piece slightly shorter than feels comfortable; generated clips rarely benefit from lingering on screen.

Lay out the timeline in beats first and fill shots into beats, rather than ordering shots as they were generated. The generation order is a production artifact and almost never the best narrative order.

Let sound carry the illusion

Viewers forgive visual softness far more readily than bad audio. Lay three layers: ambience (room tone, wind, crowd), effects (footsteps, cloth, impacts), and music. Add a subtle sound effect on each cut and the sequence instantly feels intentional. For dialogue or narration, write the line first and animate to it — pacing improves when the audio exists beforehand.

Grade once, across everything

Apply a single grade to the whole timeline so all shots share contrast and saturation, then unify with grain or a light halation pass. Ready-made structures for common formats live in Templates, which can save an afternoon of layout work on cutdowns and social versions.

Review loops that stay short

Keep one line of feedback per version so you know why a shot was dropped. On team projects, review at final size and final playback speed; a clip that looks fine in a slow preview often falls apart at real tempo.

Time budgeting for small teams

Solo creators and two-person teams can run this pipeline without a producer, but only if they budget time honestly. A realistic split for a thirty-second piece: an hour for the brief and shot list, two to three hours for generation and selects, two hours for editing and finishing, and one hour for review passes and export versions.

The useful signal is the ratio. If generation eats more than half the schedule, either the shot list is too long or individual shots are too complex. If editing eats more than half, the shots were not planned against the beat sheet. Splitting responsibilities the same way each time — one person owns the brief and shot list, one person owns generation and selects, both review at thumbnail size — keeps decisions from being relitigated at the finish line.

Quality Control and Common Mistakes

The seven-point review

• Does every shot have a purpose in the beat sheet?
• Is the protagonist recognizable across shots?
• Do exposure and white balance match across cuts?
• Any morphing artifacts at the first or last frame?
• Are hands, faces and text legible at final size?
• Does the audio peak consistently, without clipping?
• Does the piece work with sound off, and again with eyes closed?

Review on the worst screen your audience will realistically use, not your best one. Phone speakers and small screens hide detail and expose pacing, which is exactly the environment most short-form video lives in.

Mistakes worth naming

Generating before writing. Without a brief and a shot list you produce options, not shots, and the edit becomes salvage work. Fix: one page of decisions before the first render.

Overloading prompts. One clear look beats three borrowed looks stacked together. Fix: five-part formula, one camera move, one light source.

Ignoring the head of the clip. Many artifacts appear in the opening moments. Trim them rather than trying to repair them in post.

Chasing fidelity over coherence. A slightly softer, unified sequence reads better than one flawless clip wedged between mismatched ones. Fix: match down, then grade once.

Leaving audio for last. Sound changes pacing decisions, and retrofitting it usually means recutting the entire piece. Fix: a temporary music bed from day one.

Generating vertical by crop. Cropping horizontal hero shots loses the subject and the horizon. Fix: generate vertical framings for the shots that matter most.

No stop condition. Endless attempts on one shot burn the session. Fix: the three-strike rule, then simplify.

Treating a good clip as a good shot. A beautiful clip that does not serve a beat is a distraction. Keep it in a parked folder for a future project and move on.

A Worked Example: A Thirty-Second Teaser

Brief: a teaser for a coastal drama about a lighthouse keeper during a storm.

1. Brief. Logline: a keeper must choose between the light and a rescue. Tone: cold, tactile, practical light only. Rules: 2.39:1, 40mm feel, visible grain. Deliverable: thirty seconds plus a vertical cutdown.
2. Shot list. Eleven shots — coastline, tower interior, lamp mechanism, keeper's face, storm exterior, hatch struggle, beacon flare, boat silhouette, hand on lever, wide resolve, title card.
3. Generation plan. Image-to-video for the keeper's three shots from one character sheet; text-to-video for weather, water and inserts.
4. Prompting. Each prompt follows subject, action, environment, camera, light — with the same three continuity anchors appended.
5. Assembly. Cut to a temporary music bed at 100 BPM, average shot length 2.3 seconds, all settles trimmed.
6. Finishing. One grade, unified grain, three-layer sound design, then vertical recomposition with a wider safe area.

The output is not a stack of clips. It is a sequence with intent, and that difference shows in the first three seconds.

A useful test after the first assembly: describe each beat out loud in one sentence. If a beat cannot be described without naming the shot, the beat is doing structural work that the story should be doing, and the piece needs one more planning pass.

FAQ

How long should each generated clip be?

Generate longer than you plan to use — four to six seconds for a two-to-three-second edit — so you have handles for trimming and transitions. Very long generations rarely pay off, because the middle usually gets discarded and the end tends to drift.

Do I need a separate workflow for vertical video?

Yes. Generate vertical framings for hero shots rather than cropping later. Centre-cropping horizontal footage frequently cuts the subject or loses the horizon, and the loss is visible immediately on a phone. For supporting inserts, a wide-safe-area generate-and-crop approach can be good enough.

How many attempts should one shot get?

Three. If a shot fails three times, the problem is usually the concept's complexity rather than the wording. Split the action, simplify the camera move, or move from text-to-video to image-to-video with a fixed first frame.

Can I mix generated and real footage?

Yes, and it often looks better. Real inserts — hands, texture, actual locations — give generated shots something to sit against. Match grain, contrast and lens character so the seams disappear, and cut on motion where the two meet.

Do I need professional editing software?

Not for short pieces. Any timeline editor that supports frame-accurate trimming, three audio tracks and a single adjustment layer for grading is enough to start. Upgrade when your workflow, not your ambitions, demands it.

What single change improves output quality fastest?

Move your effort earlier: a clear brief, a written shot list and fixed reference frames. Prompt polish helps, but structure and references drive most of the perceived quality gain. The second fastest change is committing to one grade across the entire timeline.

How do I keep a series consistent across episodes?

Freeze the anchors and reuse the same character sheets, palette and grain settings across episodes. Save the look as a project preset, then treat any deviation as a deliberate choice rather than an accident. A series that drifts between episodes loses the recognition value it spent time building.

How do I know when a project is finished?

When the acceptance criteria from the brief are all met, not when the shots stop improving. If you can tick every item on the checklist and the piece holds up at final size on an average phone, it is done. Continuing past that point usually means refining details the audience will never see.

Put Your Next Idea in Motion

A workflow is what turns a promising idea into something watchable: brief first, shot list second, references to hold continuity, sound to carry the sequence, and one grade to bind it all together. Start with a fifteen-second test of three shots to pressure-test your pipeline — one wide, one insert, one character beat — then scale the same structure to longer work.

When you are ready to move, take a first frame and animate it. Create Video is built for exactly that, and the Orelon blog has more breakdowns of prompts, formats and finishing passes to fold into your own process.