Pipeline Templates

Tool names are examples, not endorsements, and they change monthly. Each step lists the kind of tool, with examples current as of October 2026. Check the Resources directory for up-to-date options and licenses.

How to read a template #

Each template has:

• Shape: runtime, format and what makes this type distinct
• Roles: the "hats" from the Roles Glossary you'll wear
• Pipeline: steps from idea to delivery
• Time split: a rough guide to where the hours go
• Traps: what usually goes wrong
• Variations: budget, local and hybrid versions

Every pipeline shares the same backbone:

DEVELOP → DESIGN → PLAN → GENERATE → ASSEMBLE → SOUND → FINISH → DELIVER → DOCUMENT

1. Music video #

Shape: 1–4 minutes, cut to an existing track. Mood and rhythm matter more than plot; often a loose narrative or visual concept. Strong for AI: surreal imagery fits, and music carries the emotion.

Roles: Director, Editor, DP, Production Designer, Colorist (light Sound work).

Pipeline:

1. Rights first: confirm you have permission to use the track (your own, licensed, or the artist's commission).
2. Concept & treatment: one-paragraph concept, visual motifs, color arc across the song.
3. Song map: a timeline of song sections (intro / verse / chorus / bridge) with timestamps and BPM. Mark the beats you'll cut on.
4. Look development: style reference frames, palette, the performer or character sheet if there is one.
5. Shot list by section: plan more shots per chorus and slower holds in verses.
6. Key frames: image model with multi-reference for consistency.
7. Animate: image-to-video for most shots; text-to-video for abstract interludes. Generate 20–50% more shots than you need.
8. Performance shots (if any): lip sync to the vocal track, or motion transfer from your own filmed performance.
9. Edit to the beat: cut on downbeats; use speed ramps and match cuts.
10. Grade: follow the color arc.
11. Finish: upscale, grain, deliver in 16:9 plus a vertical cut-down for social.

Time split: Development and design 15% · Generation 45% · Edit 25% · Finish 15%

Traps:

• Clip-reel syndrome: beautiful shots with no visual progression. Plan an arc.
• Lip-sync failures on sung vocals. Prefer angles and cuts that hide them, or capture your own performance.
• Cutting off-beat. Use markers on the beat grid.

Variations: Budget: unlimited low-res tiers for drafts, credits for finals. Local: ComfyUI with an open image-to-video model plus Resolve. Hybrid: film the performer on green screen and generate the worlds.

2. Narrative short (1–5 min) #

Shape: a story with characters, dialogue or performance, and an arc. The hardest type for AI: it needs character consistency, performance and continuity.

Roles: Screenwriter, Director, Casting / Character Designer, DP, Script Supervisor, Editor, Sound Designer, Composer, Colorist.

Pipeline:

1. Script: logline, beat sheet, screenplay. Write to the medium's strengths: fewer characters, controlled locations, more visual storytelling.
2. Characters: a full character sheet for each character, including a canonical description block and references. Consider a character LoRA for the leads.
3. World / locations: a world bible; an establishing key frame for each location; a floor plan if the scene has several angles.
4. Shot list and blocking diagrams, with the 180° line marked for every dialogue scene.
5. Voice: cast voices (voice actors or a voice model) and record or generate dialogue first. The performance drives the timing.
6. Animatic: rough key frames cut to the dialogue track. Lock timing before expensive generation.
7. Key frames: for every shot, using multi-reference and camera-angle tools for coverage.
8. Animate: image-to-video; first/last-frame conditioning for precise actions.
9. Lip sync on dialogue close-ups.
10. Edit: coverage-based, with J/L cuts and reaction shots.
11. Sound: dialogue cleanup, ambience, Foley-level effects, score.
12. Continuity and artifact passes (Field Guide §8).
13. Grade, finish and deliver.

Time split: Writing 15% · Design and pre-production 20% · Generation 35% · Edit 15% · Sound 10% · Finish 5%

Traps:

• Identity drift between shots. Lock it with references and LoRAs; generate faces large in frame.
• Flipped screen direction in dialogue. Plan eyelines in advance (Cookbook §11).
• Weak performance. Use acted reference (motion or performance transfer), real voice actors, and the Kuleshov effect in the edit.
• Skipping the animatic, which leads to regenerating shots that get cut anyway.

Variations: Hybrid: real actors for close-ups with AI environments and wides. Animation style: stylized characters reduce uncanny valley and identity drift.

3. Trailer / teaser #

Shape: 30–150 seconds selling a film that may not exist (concept trailer). Built on structure, rhythm and sound design.

Roles: Trailer Editor, Director, Sound Designer, Motion Graphics / Title Designer, Colorist.

Pipeline:

1. Concept: logline, genre, tone, the "big question."
2. Trailer structure: a three-act trailer — setup (world and hero), escalation (conflict, rising pace), climax montage, title card, button (the final beat or joke).
3. Sound skeleton first: choose or compose the music track and sketch the sound design (hits, risers, silences) before generating. Trailers are cut to sound.
4. Shot wishlist: iconic "trailer shots" (hero reveal, the wide establishing world, action beats, the quiet moment before the drop).
5. Generate: prioritize striking single images over continuity. Trailers forgive gaps in story but not weak shots.
6. Dialogue / VO lines: a few strong lines (voice model or actor).
7. Edit: fast cutting in act two, rhythm on the hits, a strategic black frame before the title.
8. Titles and graphics: title treatment, text cards, release-style end card.
9. Mix: heavy on hits and silence contrast; loudness to platform spec.
10. Grade and deliver: 16:9 plus a vertical teaser version.

Time split: Concept and structure 15% · Sound skeleton 15% · Generation 35% · Edit 25% · Titles and finish 10%

Traps:

• No structure (a montage of cool shots). Use the three-act trailer form.
• Generic titles. Design a real title treatment.
• Sound as an afterthought. In trailers, sound is the edit.

4. Spec ad / product spot #

Shape: 15–60 seconds, one product or brand message, often several aspect ratios. A strong AI portfolio piece for commercial work.

Roles: Producer, Director, Art Director, DP, Editor, Motion Graphics, Clearance.

Pipeline:

1. Brief: product, audience, single message, call to action. For a real brand without permission, label it clearly as an unofficial spec ad and check the rights primer. Using a fictional brand is safer.
2. Script and boards: 3–8 shots; a storyboard for each frame.
3. Product accuracy: product photos as references; consider a product LoRA or 3D model so shape, logo and label stay exact. Composite the real logo and text in post.
4. Key frames: hero product shot, lifestyle shots, detail macros.
5. Animate: controlled moves (slow orbits, push-ins, macro slides).
6. Variants: use transformation tools to make versions (different settings, talent, colors) from one base.
7. Edit: rhythm, product on screen at the key moment, end card with the call to action.
8. Sound: music bed, product sounds (clicks, pours, fizz), VO.
9. Deliver: 16:9, 9:16, 1:1, and 6s/15s/30s cut-downs.

Time split: Brief and boards 15% · Product accuracy 20% · Generation 30% · Edit and versions 25% · Finish 10%

Traps:

• Wrong product details (logo, label, shape). Clients notice instantly. Composite the real ones.
• Commercial-use licensing. Use models and plans that allow commercial work, or licensed-data models if the client asks for provenance.
• Forgetting cut-downs and aspect ratios.

5. Documentary-style / essay film #

Shape: 2–15 minutes. Narration or interviews plus visuals; AI used for reconstruction, illustration or archival-style imagery.

Roles: Researcher, Writer, Director, Editor, Sound, Clearance.

Pipeline:

1. Research: sources, facts, interviews (real people with consent).
2. Script / narration draft; record a scratch VO.
3. Visual strategy: decide what is real footage, what is archival, and what is AI reconstruction. Plan how you'll tell the viewer which is which.
4. Record the final narration or interviews.
5. Edit the radio cut (audio only) first.
6. Visuals: real or licensed footage plus AI illustrations; consistent stylization for reconstructions (e.g. an illustrated or painterly look) so they aren't mistaken for real footage.
7. Edit to the narration; B-roll; lower thirds.
8. Fact-check pass.
9. Captions, sound mix, grade, deliver.

Time split: Research and writing 30% · Recording 10% · Edit 25% · Visuals 25% · Finish 10%

Traps:

• Photoreal AI presented as documentary truth. This is an ethics and trust problem, and possibly a legal one. Label reconstructions on screen and stylize them.
• Synthetic depictions of real people without consent.
• Unverified facts.

6. Vertical micro-drama / social series #

Shape: 30–90 second episodes in 9:16, a hook in the first second, a cliffhanger at the end, a recurring cast. Built for TikTok, Reels and Shorts.

Roles: Showrunner, Writer, Character Designer, Director, Editor, Social Media Manager.

Pipeline:

1. Series bible: premise, characters (full sheets with LoRAs or references), world, episode arc.
2. Episode beats: hook (0–2s), escalation, cliffhanger.
3. Batch production: write, design and generate several episodes at once so the look stays consistent.
4. Frame for vertical: close-ups and medium shots; faces in the upper-middle; keep the bottom ~20% clear of key action because of UI overlays.
5. Burned-in captions (most viewers watch muted).
6. Fast edit, with sound hits on the hook and cliffhanger.
7. Release cadence: schedule, thumbnails, titles; track retention.

Time split: Series development 20% (up front) · Per episode: Generation 45% · Edit 30% · Captions and publish 25%

Traps:

• Character drift across episodes. Use LoRAs or locked references and the same model version.
• A slow open. Viewers swipe away in under 2 seconds.
• Designing in 16:9 and cropping.

7. Weekly challenge sprint (10–60s) #

Shape: AiFC weekly format. One craft skill, a short runtime, a 7-day deadline.

Pipeline (by day):

Traps: spending six days generating and one day on everything else. Protect Friday and Saturday for post.

Pipeline variations by budget & setup #

Choosing a pipeline #