AI video, with direction built in
Cadenza turns one brief into a directed sequence. It plans the hook, locks continuity, and routes each shot before a render credit is spent.
Story-first. Continuity-locked. Model-routed.
Brief to directed sequence
Demonstration dataBrief
Launch film for a trail shoe, 15 seconds, vertical, pre dawn runners.
Story plan
Tentpole mode. Hook, two development beats, payoff.
Shot sequence
Five shots, each with a narrative and emotional purpose.
Continuity locks
Character, style, and lighting references pinned across shots.
Model routing
Dialogue routes to Veo, reference motion to Seedance, per policy.
Result
A provenance signed sequence, ready to refine or publish.
The proof
Most AI video tools begin by rendering. Cadenza begins by deciding what the sequence should mean, what each shot contributes, and what must stay consistent. This is what one directed sequence looks like before generation.
Shot plan
Demonstration dataBrief: a 15 second launch film for a trail running shoe, vertical, for people who run before sunrise.
Hook
route: veo
Stop the scroll: a single striking image of the shoe cutting through dark trail mist.
35mm slow push in, slight low angle, high contrast key light.
Development
route: seedance
The runner moves. The world widens and the product proves itself in motion.
50mm lateral track at eye level, motivated practical lighting.
Development
route: seedance
Peak intensity. Full stride, terrain flying past, the transformation undeniable.
24mm fast dolly in, dynamic low angle, harder key with sharper contrast.
Payoff
route: veo
Arrival. The runner crests the ridge at first light, the product commanding the frame.
85mm slow pull back to hero, warm key at its richest.
Continuity locks
Story Intelligence checks
Advisory by design. Cadenza flags weaknesses and you always keep the final say.
The problem
Rendering first means paying to discover problems. Industry benchmarks we track put effective cost near $1.65 per finished 8 second clip on credit tools, roughly three renders for every keeper. Cadenza attacks the denominator: decide first, render once more often.
Without direction
With Cadenza
Direct it
Story Intelligence plans the sequence and gives every shot a purpose before anything renders. Five structural checks run on the plan, and they advise rather than block.
Keep the world
Continuity references lock character, product, lighting, and palette when a sequence starts. Later shots inherit the same references instead of re rolling the dice.
Refine surgically
A lighting note should not replace your actor, camera, and approved shots. Refinement is scoped to the channel you asked to change, and the rest stays locked.
How it works
01
Describe the outcome, or point Cadenza at a product URL. Brand facts land in editable Brand Memory.
02
Cadenza proposes a mode, an arc, and a shot plan with story checks. You adjust anything before spend.
03
Shots render with locked continuity references and policy driven model routing, with failover if a model is down.
04
Refine scoped channels, export per platform, and confirm every post explicitly. Nothing publishes on its own.
The same four steps work from the studio, the REST API, or an agent over MCP. One identity, one plan, one continuity store.
Director Modes
A Director Mode decides camera movement, lensing, pacing, lighting, and blocking for every beat of the arc. 14 modes ship today, defined as reviewable configuration. Auto selection reads your brief and picks one, and your explicit choice always wins.
Handheld energy, continuous-thought cuts, UGC immediacy.
Audience feeling: I am there.
Camera grammar
handheld, in-the-moment — no polish, all energy. Opens on a 28mm handheld, resolves on a 35mm whip pan to result.
Pacing
fast, cuts on energy change, montage compression
Product URL to advertisement
Give Cadenza a product URL and it extracts the product name, palette, call to action, and hero image into Brand Memory as an editable draft. Nothing is trusted silently: you correct the facts, then the product image is locked as a continuity reference so the real product appears in every shot.
Extraction respects page access and falls back to manual entry when a page cannot be read. Brand facts never leak into public share pages.
Paste the product URL
Cadenza reads the page and drafts product facts from it.
Correct the draft
Name, palette, call to action, and hero image are editable Brand Memory, never silently trusted.
Lock the product
The hero image becomes a continuity reference so the real product appears in every shot.
Direct the ad
A story arc is proposed around the product and you generate when the plan looks right.
Directed with Cadenza
Creators can opt in to publish how a sequence was directed: the mode, the arc, each shot's purpose, and the continuity proof, with every brand fact scrubbed at the boundary. The gallery grows as beta creators choose to share.
Agent-native
The full model set and the same economics are available through MCP and REST, not a reduced tier. Connect once with a browser approval, no API key pasted into any config file.
/plugin marketplace add nikegunn/cadenza/plugin install cadenza@cadenza-marketplaceRichest bundle, MCP tools + /connect + /direct + the creative-director agent + a skill.
Then just ask
"Create a 15 second premium launch film for this product. Plan the story first, preserve the product in every shot, and show me the plan before starting."
Cadenza returns the shot plan, the Director Mode it chose, story checks, and locked continuity references. Billable work starts only after the plan.
Read the agent integration docsBuild in public
Every phase on this board maps to test gates and verification runs recorded in the repository. A phase reads live only after its gates pass. The three seams still open are stated plainly: live provider renders, live payment checkout, and live platform posting.
See the full build logVideo and image model registry with intent-aware routing policy.
Idea to directed shot to guarded render path with provenance.
Per-shot routing across Kling, Veo, Seedance, Runway, and Luma paths.
Locked character, style, and lighting references across every shot.
Hook, arc, shot purpose, mute test, and attention-cut checks before render.
Policy checks, consent gating, provenance, watermarking, and audit logs.
Locked brand facts, URL ingest, and continuity-ready product references.
Fourteen camera grammars with auto selection from the creative brief.
REST, MCP, plugin, skills CLI, guided prompts, and server-issued scoped tokens.
Supabase Google, GitHub, magic link auth, and editable creator profiles.
Platform previews for TikTok, YouTube, Instagram, and Facebook with explicit confirmation.
Public direction pages, brand scrubbing, remix handoff, pricing, quality feedback, and failover.
Pricing philosophy
Estimates are visible before generation, usage is visible after, and refining one channel of one shot should never bill like a new project. Checkout is in beta; tiers below are the published plan structure.
Full pricing and FAQFirst cut
$0/mo
5 finished videos included
Creator
$29/mo
40 finished videos included
Developer
$49/mo
100 finished videos included
Agency
$199/mo
500 finished videos included
Trust
Every generated asset carries C2PA style provenance, a watermark, and a disclosure flag. No asset returns without it.
Real person likeness is deny by default and requires consent. Political and election content is blocked in every geography.
Platform posting is never autonomous. Each destination requires an explicit confirmation after preview.
Every generation, block, and publish is written to a hash chained audit log.
The beta includes the full direction layer: story planning, continuity locks, Director Modes, and the agent surface.