01
Brief
The Act idea (bridge, growth loop, ICE reason), its trend evidence, the brand's BRIEF and VOICE.md, the top past examples, the channel's format spec and the owner's learned corrections — assembled into one brief per package.
From an approved idea to a finished reel — in the creator's own voice and face.
ContentMachina turns a growth hack into platform-native pieces that sound like the brand because it learned the voice from the creator's own posts. Then the Creator Studio produces the video: looks, storyboard, keyframes with the consenting creator's likeness, image-to-video, cloned voice, lipsync, captions, music bed and a provenance label on every export. It creates. It never posts.
Picks up Act ideas from SignalMachina within 15 minutes · hands approved pieces to DistroMachina
piece
tiktok.videoes-ARcritic 8/103 hooks
Complaints in reviews: charges and price
A competitor's users complain about surprise charges. The piece shows the assistant doing the same task free, in the brand's voice, with the loop in the CTA: forward it to the group that always pays late.
Hook A (contrast) · Hook B (benefit) · Hook C (curiosity) — pick one, the choice is remembered.
first studio run 6-scene board · 4 image scenes narrated + 1 image-to-video scene · 27.4 s · −14.2 LUFS · 67 aligned caption words · US$0.42
reel
parodycharacter_videolook parodia_cine25.1 s
Titanic · the bow
The creator is the lead of a recreated scene: a generic first-class passenger asks something product-related, she turns and answers in her own cloned voice, a chat card shows the app replying, then the close.
4 character shots + 2 cards · keyframes keep her real face · same co-star across every shot · cloned voice + lipsync · US$3.81 on the ledger
01 The shift
"Generate more content" was never the job. Sounding like you was.
A piece is useful when it carries a specific idea, in the brand's register, in the platform's shape, with the share step written in — and when the person who owns the brand can approve it in one pass. ContentMachina optimises for that pass: fewer edits, a critic that catches what you would, and a memory that stops you correcting the same thing twice.
02 How it works
From Act to pieces in under twenty minutes, with zero paid tokens by default.
A job every 15 minutes picks up Act ideas; the owner's "Draft content" button does the same on demand.
02
Draft
One call per package: a piece per enabled channel and format, publishable text in the brand's language, owner notes in the owner's, three hook variants (curiosity, benefit, contrast) on every format with a hook, the loop made explicit in the CTA.
03
Critic
A separate brand-guardian call scores voice fit, platform fit, specificity and loop clarity, flags off-voice lines, no-go topics, unverified facts and limit breaches. Voice fit below 6 triggers one redraft with the critique.
04
Review
Pieces side by side in the Review Queue: pick a hook, edit inline (versions kept), approve or reject with a why. Edits and rejections are summarised by cm voice-learn into the brand's learned memory, injected into the next draft.
05
Assets
Approved pieces get three 9:16 covers and a narration automatically under their caps; b-roll is a button, never automatic. Covers render text-free; the hook is typeset locally, which is the fix for garbled on-image text.
06
Studio
For a finished reel, the Creator Studio: look → storyboard with thumbnails → approve scene by scene → render → assemble with captions and music → export with a provenance sidecar. Change one scene or the look and re-render only that.
03 What it makes
Text, voice, covers, video — all from the same brief.
Platform-native, not reformatted
TikTok video (hook ≤ 3 s, beats, on-screen text, sound, CTA, shot list), Instagram reel, carousel and story, YouTube short, WhatsApp forward or broadcast, ad creative with three hooks, X post — the set per brand family, overridable per brand.
Learned from the creator's own posts
VOICE.md (tone, vocabulary, do and don't, registers such as voseo, CTA style, no-go) written by the owner or proposed from examples; past posts and scripts as few-shot; every correction summarised into a learned file that never overwrites the owner's.
Three hooks, one critic, measured
Every scripted piece ships with three hook variants and the critic's scores. The Quality page tracks critic averages per channel and prompt version, the blocking rate and the share of pieces approved with at most two edits.
Covers, narration, b-roll
Three cover variants per piece, Spanish or English narration, 5-second b-roll on demand — each under its own monthly cap, each row carrying model, prompt and cost. Failures land on the row and are retried from the panel, never silently.
Voice and likeness, with consent
A cleaned 60–90 s voice reference, a cloned voice, and a talking head by lipsync over the creator's own clip bank or an avatar from a still. Word-level captions aligned to the known script, app-screen overlay, provenance comment and sidecar. No clone exists until consent is recorded; one command forgets it.
Storyboard → keyframes → video
4–6 scenes from a script, an Act idea or a free brief, each with a beat, a text-free prompt, a kind (talking head, video, image, character in scene, screen card) and the exact words spoken. Looks as versioned config set palette, fonts, captions, cuts, music mood and voice preference.
The creator inside an iconic scene
Scene recreation with generated settings and generic co-stars — no film footage, no real actors' faces, no franchise marks. Keyframes composed from 3–4 crops of the creator plus one co-star reference and a setting plate, so the lead is recognisably her and the co-star stays the same person across the reel; chat and end cards drawn locally in the brand's font.
Captions, music bed, −14 LUFS, sidecar
Hard cuts or dissolves, captions aligned to the script plus an SRT, a music bed from the library ducked under the voice, two-pass loudness to −14 LUFS / −1 dBTP verified, beat-synced cuts when the look asks for it, and an export that carries the AI-generated metadata, the look version, every model used and the platform labels to set.
Caps before every call
Local models draft and critique by default. Paid renders go through one queue with a spend guard: thumbnails at a third of a cent, keyframes US$0.15, image-to-video US$0.14 per second, cloned speech per 1k characters, lipsync per minute. The panel prints the estimate before you press render.
04 The Creator Studio
A storyboard you approve before any video money moves.
Five stages, each cheap to redo on its own. Every project starts from a signal: an approved script, an Act idea or a brief you type.
01
Look
A versioned file per visual style — palette, fonts, caption style, shot bounds, dissolves, music mood, voice preference. Shipped: a clean product look, a sci-fi film look, a stadium look, a lo-fi music-video look and a cinematic parody look. Changing the look restyles the whole board.
02
Storyboard
A local model proposes 4–6 scenes with a thumbnail each; you approve, edit the beat, the prompt, the spoken line or the duration, or regenerate a single scene. No video money is spent before this step.
03
Keyframes
For scenes starring the creator, a still composed from her reference crops, the co-star reference and the setting plate — accepted and kept as her real face on all production keyframes to date. Upscaled only when needed; the panel shows the real still.
04
Render
Image scenes get the title typeset locally and a slow push-in; video scenes animate the approved still with image-to-video, faces and wardrobe locked; her lines come from the cloned voice and are lipsynced over the shot, or stay a silent placeholder track you record yourself.
05
Assemble & export
Cuts, voice track, speaker-coloured dialogue captions, optional music bed, loudness mastering, cover, SRT and the provenance sidecar. Re-voice a line without regenerating the video. Three 23–25 s parody reels cost US$3.8–4.5 each.
05 Autonomy & safety
Drafts only. Consent first. Labelled always.
ContentMachina runs at draft for every brand: nothing leaves the review queue without you. The clone and studio add their own rules.
Never postsPublishing is DistroMachina's, behind its own gate. The studio exports files and rows, not posts.
No clone without consentThe owner records who consented, when, to what and how to revoke it before a voice can be created. One command forgets the voice locally; the owner deletes it at the provider.
AI labels mandatoryEvery export carries an AI-generated metadata comment and a sidecar listing the models used; the platform's own label (TikTok, Meta, YouTube) is required on upload and the distributor checks it.
No real faces but the creator'sParodies use generated settings and generic co-stars — no film footage, no actors' faces, no franchise marks. The creator's lines can stay a silent track she records herself.
Facts stay verifiedThe critic blocks no-go topics and invented facts; [verify] markers survive the draft and are stripped only from narration.
Caps before every callEach render kind has a monthly cap; the guard refuses before any upload and leaves no row. Video b-roll is never automatic.
06 In the panel
The ContentMachina group of the GrowthMachina Panel.
Review Queue
/contentPackages → pieces side by side → critic → hook variants → approve all or per piece; per-piece history with field diffs.
Quality
/content/qualityCritic averages per channel and prompt version, blocking rate, approvals with ≤ 2 edits, the last corrections grouped by action.
Voice Memory
/content/voiceVOICE.md rendered, examples uploaded from the panel, learned corrections, 'regenerate memory'.
Asset Studio
/content/assetsSpend per kind against its cap, failures with retry, the recent grid; clone status and provenance badges.
Creator Studio
/content/studioProjects from a piece, an idea or a brief with a look and a character; storyboard grid; 'render approved (≈ US$X)', assemble, export; library of looks, characters, brand assets, uploads and music.
Command Center
/Agent Swarm shows the creator's live state; the API spend tile includes every render ledger.
07 Status · 2026-10-04
What is shipped, what is partial, what is next.
Shipped
- Packages and pieces per channel from Act ideas; brand-guardian critic with one redraft; three hook variants; Review Queue with history; Quality page
- Voice memory: VOICE.md, examples, learned corrections regenerated on demand
- Asset Studio: covers, narration, b-roll under monthly caps
- Creator clone: voice reference, cloned voice, lipsync or avatar talking head, aligned captions, provenance
- Creator Studio: looks, storyboard with thumbnails, per-scene render, assembly with captions, music bed and loudness mastering, export with sidecar
- Parodies v2: identity keyframes, image-to-video, cloned voice with lipsync, chat and end cards, dialogue captions, re-voice without re-render
Partial
- Cloned voice quality is judged per creator; where it falls short, her lines stay a replaceable track and are muxed or lipsynced from her own recording
- Reference-to-video models refuse real faces; the keyframe path is the production path
- "Pieces ready to review" notification waits for the Telegram bot
Deferred
- Performance transfer (the creator's own take driving the avatar)
- Restyling existing footage; music generation (the bed comes from the library today)
- "Viral looks" derived from trending sounds; batch variants per look
- Repurposing from a dropped-in URL or transcript; changelog-to-content; newsletter
Early access · limited cohort
Your voice, your face, your approval — at the pace of the trend.
Bring ten to twenty past posts and a creator who consents. The first drafts arrive the morning after the first Act.