Building Hollywood quality commercials in record time with significantly lower cost.

The $500 Video That Beat Our $15k Agency Reel

Six months ago we’d spend $10k–$25k per campaign for “brand videos” that took 6–8 weeks and died after two posts.

Today, we can ideate in the morning, ship by lunch, and beat those agency numbers with a total hard cost under $500—and most of that is paid media to test winners.

This post breaks down the exact system we use to concept, storyboard, generate, and publish AI-native videos using Claude Max, Google Veo 3.1 (+Flow), 11labs, and Canva—so you can do it too.

TL;DR (System at a Glance)

• Creative: Claude Max drafts brief, tone, and comedic beats → produces Veo-ready visual-only storyboard.
• Generation: Google Veo 3.1 renders visuals & SFX (no VO). We enforce an 8s clip limit and stitch continuity with Flow “extend.”
• Audio: 11labs generates dialogue, character VO, and music stems as separate layers for clean mixing and rapid A/Bs.
• Edit: Canva master timeline: assemble clips, mix layers, final grade, export, and version for placements.
• Output cadence: From brief → publish in 2–4 hours; scale concepts to 5–10 variations/day.

Tools (Our Standard Stack)

• Claude Max — creative brief, archetypes, tone, joke mechanics, and production-grade storyboards/prompts
• Google Veo 3.1 (+ Flow) — visuals + SFX only (no VO/music here); use Flow → “Extend” for continuity between 8s shots
• 11labs — voiceover & dialogue (character voices), narration, and background music stems
• Canva — editing, leveling, transitions, and platform-resized exports; quick social versioning (Reels, Shorts, LinkedIn)

Why This Beats the Old Way

The Old Way (Time, Teams, Cost)

The New Way (Agentic AI Workflow)

Impact: We trade big, infrequent “brand films” for continuous creative velocity—more shots on goal, higher odds of breakout winners, lower creative risk.

Our Methods (Step-by-Step)

1) Creative Brief (Claude Max)

Claude is our head of comedy + strategy. We feed it ICP, tone, and a few brand rules. It returns: logline → beats → storyboard scaffolding.

Prompt Starter (copy/paste):

You are Head of Creative at Anyreach, an omni-channel agentic AI platform.
Goal: Write an infotainment creative brief for a <60s video that blends education + comedy with “confident, self-aware, slightly outrageous” tone.
Audience: <insert ICP>.
Output:
1) Big Idea (1-2 sentences)
2) Hook lines (first 3s, 3 options)
3) Story arc in 5–6 scenes (Problem → Twist → Solution → CTA)
4) Comedy triggers per scene (surprise, stereotype, irony, timing)
5) Visual metaphors (2–3 per scene), no on-screen text
6) Safety guardrails (brand tone, PG humor, no cringe, headset continuity if present)
Return a Veo-ready storyboard with shot length ≤8s.

Why this works: It forces short, visual-first scenes and ensures every beat earns its place.

2) 🎯 ICP & Persona Snap (for the brief)

Use this compact block inside your prompt so Claude grounds the humor and pain points:

3) Storyboard (Veo 3 Meta Framework)

We convert Claude’s beats into Veo 3.1-friendly storyboards: visuals + SFX only (no dialogue). Follow the Veo-3 Meta Framework style: shot grammar, camera moves, ingredients, negative prompts, clip length, and continuity locks.

Storyboard Checklist

• Clip length: max 8s (Veo 3.1 constraint)
• Continuity locks: wardrobe, props (e.g., “headset stays on”), character count, lighting mood
• Camera notes: shot size, motion (dolly, pan), rack focus cues
• SFX bed: room tone, foley anchors (beeps, squeaks, clicks)
• Negative prompts: ban overlays, captions, horror lighting, etc.
• Ingredients: optional image refs (character, set, props)

Tip: End each scene with a clean entrance/exit action (turn, door open, head tilt). That motion is where Flow’s “extend” stitches invisibly.

4) Video Generation (Veo 3.1 + Flow)

• Input: For each scene, include start/end frames in the prompt to guide pose and set continuity.
• Render: Generate at 16:9, 1080p, audio ON for SFX (we’ll mute/duck later as needed).
• Continuity: Use Flow → Extend to bridge shots so the final composite feels like one continuous take per sequence.
• Select: For each scene, shortlist 2–3 alternates (A/B/C); keep the same camera grammar so alternates drop in cleanly.

Veo Prompt Skeleton (per scene)

Scene: <short name> (8s, 16:9, 1080p, audio ON)
Cinematography: <shot size + move>; <lighting>; natural skin tones
Subject: <who + wardrobe + fixed props>
Action: <primary beats, 1–2 sentences>
Context: <set + prop logic>
SFX/Ambience: <room tone + foley>
Music: low bed only (no melody); duck -3 dB under key beats
Continuity locks: <wardrobe/props/character count>
Negative prompts: no on-screen text, no glitches, no duplicates
Ingredients: <optional refs>

5) Dialogue, Narration, Music (11labs)

We generate voice layers separately to keep the visuals modular.

Layers we export:

1. VO Narration (neutral or witty guide)
2. Character Dialogue (distinct timbres; pace matched to cuts)
3. Music Bed (simple, loopable motif; 15–30s stems)
4. Optional Stingers (CTA sting, comedic accents)

Dialogue Prompt Starter:

You are the voice of Sam: Black male, late-20s/30s, witty, slightly dry, confident.
Generate 3 takes per line: A) deadpan, B) animated, C) whisper-confident.
Target read length: 95–105% of captions timing (0.8–1.0s per 8 words).
Keep lines under 12 words each; punchlines end on a hard consonant.

Music Stem Prompt Starter:

Brief: Upbeat confident groove, non-intrusive. 100–108 BPM.
Style: modern corporate with light funk percussion.
Export: 15s, 30s loops + 2 stingers (0.8s / 1.2s).
Avoid heavy melody; prioritize rhythm.

6) Edit & Versioning (Canva)

Master Timeline Rules

• Track 1: Picture master (Veo scenes; lock continuity)
• Track 2: SFX bed (from Veo; selectively duck/remove under VO)
• Track 3: Dialogue (11labs; micro-fades on breath)
• Track 4: Narration (or mute if character-led)
• Track 5: Music (sidechain duck −3 to −6 dB during lines)
• Track 6: CTA cards only (end screen; no text overlays earlier)

Finishing

• Level peaks at −1.0 dB, dialog LUFS around −16.
• Quick neutral grade: lift shadows +5, reduce highlights −4, saturation +6.
• Export master at 16:9, 1080p, 20–25 Mbps.

Versioning

• LinkedIn (landscape, 30–50s)
• Shorts/Reels/TikTok (9:16, punchy 20–35s)
• YouTube (16:9, up to 60s)
• Web hero (muted autoplay variant; add burned-in captions if needed)

QA & Continuity: The 12-Point Gate

Before you export, run this checklist:

1. Hook lands in 0–3s (first line or visual shock).
2. No on-screen text before CTA card(s).
3. Headset continuity on Sam (never disappears).
4. Single Sam rule (no duplicates/reflections).
5. Robot look consistent (soft vinyl, LED smile).
6. SFX bed cohesive: same room tone family.
7. VO pace matches cuts (no breath collisions).
8. Music ducking on key lines (−3 to −6 dB).
9. Scene transitions motivated by motion (doors, turns, pans).
10. Color pass consistent skin tones; avoid horror contrast.
11. File hygiene: named stems, versioned edits, export presets.
12. Platform cuts: landscape + 9:16 vertical made and titled.

A/B Testing Framework (Make Winners, Not “Videos”)

• Variable 1 — Hook: Swap the first 3 seconds; keep rest identical.
• Variable 2 — Punchline: Same build, different last line.
• Variable 3 — Music: Groovy vs minimal percussive bed.
• Variable 4 — CTA: “Book a demo” vs “See it answer your calls.”

Readout: 24–48h micro-spend; pick winner on 3s/10s hold, watch-through, clicks. Fold insights back into Claude’s next brief (“prefer disruption + stereotype; cut irony tangent”).

Governance & Safety (Don’t Skip This)

• Brand guardrails baked into the Claude prompt (PG humor, empathy, no punches down).
• Voice cloning consent: use internal/synthetic voices; avoid real-person likeness without written approval.
• Logo usage: only on CTA cards unless specifically approved in brief.
• Accessibility: publish captioned versions (burned-in for paid placements as needed).

Results We See Consistently

• Concept-to-post in hours, not weeks
• Higher hook retention (3s/10s) vs agency reels
• Cheaper iteration → more tries → more hits
• Sales lift: posts that make prospects say “that’s so me”

FAQ

Q: Why separate VO/music from Veo?
A: Clean layers = faster A/Bs, clearer dialog mixing, and platform-specific music swaps without re-rendering visuals.

Q: Can I do longer than 8 seconds per shot?
A: Veo 3.1 caps at ~8s; use Flow Extend to fake longer takes. Plan actions that naturally bridge across clips.

Q: When do I use captions?
A: Keep scenes text-free for cinematic impact; use end-card CTAs. For feed placements where sound is off, ship a second cut with burned-in captions.

Q: Do I need a robot?
A: No—but a recurring character/archetype trains your audience. The point is continuity and recall, not “cute bot.”

Copy/Paste SOP (Internal)

1. Kickoff in Claude: Brief → Beats → Veo storyboard
2. Build 6–8 Veo scenes @ 8s each; Flow Extend between key motions
3. Generate VO/dialogue/music in 11labs (3 takes/line)
4. Assemble in Canva: stems on separate tracks; ducking & micro-fades
5. Export master + platform variants; captioned alt as needed
6. A/B test hooks; roll winners into paid and homepage hero
7. Archive: raw renders, stems, .canva file, final exports, analytics snapshot

The End Result

Bottom Line

AI-native video isn’t a cheaper imitation of live action—it’s a different creative sport. With a tight brief, ruthless scene discipline, and continuity tricks, you can ship memorable, funny, on-brand videos in hours, keep iterating, and win the attention game without burning budgets.

If you want our templates (Claude brief, Veo scene skeletons, 11labs VO packs, and Canva timeline), say the word and I’ll drop a sharable kit.