AI Cinematic Video Pipeline Architect — Build a Reference-to-Commercial System (Storyboard → Keyframes → Ad)

## What we are buying We are not hiring you to make one video. We are buying your system. We want a packaged, pre-designed pipeline for cinematic AI video — the prompt architecture, presets, plugins, node graphs, SOPs and decision rules you already use — so a junior operator on our team can drop in a reference and get a broadcast-ready commercial out the other end, repeatably, without you. Today, output quality depends on the operator. We want it to depend on the system. ## The pipeline you will build INPUT: a reference (a film, an ad, a mood video, a still, a brand deck) + a product. OUTPUT: a finished commercial, ready for sponsors, paid ads and organic placement, in every master format. 1. REFERENCE DECODING — extract the DNA of the reference: shot grammar, focal language, color science, lighting, pacing, camera movement, texture, edit rhythm. Output is a structured, reusable "look spec", not a vibe description. 2. SCOPE & BRIEF — a guided intake turning "here is my reference + my product" into a locked creative direction: concept, narrative arc, duration, aspect ratios, CTA. 3. STORYBOARD — break the brief into a numbered shot list: shot type, duration, camera move, subject action, VO line, audio cue, and the exact generation prompt per shot. 4. KEYFRAMES — every shot built first as a still image, separately, with locked character, product, wardrobe and brand consistency across the whole board. 5. ANIMATION — keyframes to video, model chosen per shot type (not one model for everything), with motion control and explicit re-roll criteria. 6. ASSEMBLY & SOUND — VO, music, sound design, transitions, pacing, color matched to the reference. 7. COMMERCIAL LAYER — hook architecture, retention curve, on-screen typography, logo and endcard, masters in 9:16/1:1/16:9, captions, A/B variants. 8. QC GATE — a written pass/fail checklist so the operator knows when a video is done versus when to re-roll, without asking anyone. ## Deliverables — the definition of done This job is complete when we have: 1. A written Pipeline Playbook (SOP) covering all 8 stages, step by step, executable by a non-specialist with no prior AI video experience. 2. A Reference Decoding framework: the template, questions and rubric that turn any reference into a look spec. 3. A prompt library: modular, parameterized blocks for storyboard, keyframe and motion generation — variables, not one-off prompts. 4. The automation layer, packaged and runnable: ComfyUI workflows, n8n flows, Claude Skills/MCP or scripts — importable files that run in our environment, plus a setup guide. 5. A model routing table: which model wins for which shot type and why, with cost per output and fallbacks when a model or platform changes. 6. A QC checklist and re-roll decision tree. 7. TWO complete reference cases produced end to end with the pipeline: we give you two references and two products; you deliver storyboard, keyframes, shot files, the final commercial in all masters, plus a written trace of the decisions. 8. One 30–60s commercial produced live, in front of us, following only the documentation — proving the system works without improvisation. 9. Two recorded training sessions for our operator, plus a Q&A round after they run it alone. 10. Full IP transfer: all documentation, prompts, workflows, presets and assets become our property on final payment. ## Tool stack Video: Veo 3.1, Sora 2, Kling 3.0, Seedance 2.0, Runway Gen-4.5, Wan 2.6, Higgsfield Image: Midjourney, Nano Banana / Gemini Image, Flux, Seedream Audio: ElevenLabs, Suno, Udio Post: DaVinci Resolve, Premiere, After Effects, CapCut Automation: ComfyUI, n8n, Claude Skills / MCP, Python You do not need all of them. You need a defensible opinion on which wins for which shot, and receipts to back it. ## Who this is for You fit if you have shipped cinematic AI video a client paid for and ran as an ad, and you have already systematized your process. You do not fit if your portfolio is one-off text-to-video clips, if character and product consistency across a full board is still a problem, or if you cannot document how you work. ## Engagement Hourly, part-time, roughly 4–8 weeks, run in four phases. P1: paid pilot — decode one reference we send; deliver the look spec and a 6–8 shot storyboard with keyframes. Gate to the rest. P2: documentation and automation layer. P3: the two full reference cases. P4: live production test, training, IP handover. Tell us your rate and hour estimate per phase. We would rather hear your reasoning than have you guess our budget. ## How to apply Skip the generic cover letter. We read the ones that answer this: 1. Link 2–3 finished commercials you made with AI. For ONE, name the reference you were matching and how you hit it. 2. Show something from your existing system — a ComfyUI graph, an SOP page, a prompt template. Proof it exists. 3. The hardest consistency problem you have solved, and how. 4. Your rate, hour estimate per phase, and timeline.