It finishes the job
Most tools hand you ingredients and leave the assembly to you. This one renders the actual MP4 — synced voiceover, images on their timestamps, transitions, captions, thumbnails — and tells you when it is done.
One message in. A finished video out.
Describe a video, or upload a CSV of topics. An agent running in the cloud writes the script, records the voiceover, storyboards every beat, generates every image in one locked style, assembles the video with transitions and captions, makes the thumbnails, and — if you ask — uploads it. You never open a timeline.
Bring your own provider keys and every plan is half price — you pay the model vendors directly instead of us.
- 01ScriptWritten to an exact word target with the retention frameworks you choose.
- 02VoiceoverNarrated with word-level timestamps returned alongside the audio.
- 03StoryboardOne image per change of visual subject, aligned to word boundaries.
- 04ImagesEvery scene generated in one locked style, batched with bounded concurrency.
- 05AssemblyTransitions, Ken Burns, burned captions and a ducked music bed.
- 06PublishUpload or schedule to a connected channel. Never marked for kids.
Built around two things most tools skip
Finishing the video, and never depending on one vendor to do it.
No vendor is hardwired
Every model sits behind an adapter. Bring your own key and that call costs you nothing here. Connect an account, or run on ours and pay in credits. Swap a provider on a channel and the next video changes with no code.
Durable, not fragile
Jobs are durable workflows. Close the tab, lose your connection, restart the backend — the job resumes from the last completed stage. Every stage retries on its own and never double-charges.
One or a hundred and twenty
The same pipeline runs for one typed topic or a CSV of a hundred and twenty. Each row is an independent, resumable job, so one failure never blocks the rest.
Automatic until you want in
Pause before the expensive stage, read the script, edit it, resume — and only the stages downstream re-run. Every artifact is editable at any point.
The price is shown first
A cost estimate appears before anything is spent, broken down by stage. Behind it we record the real vendor cost of every call, so the pricing is measured rather than guessed.
Where this differs from the tools it was measured against
Each row is a capability we built because the alternative did not have it.
CapabilityOctanaTypical alternative
End-to-end assemblyRenders the finished MP4Generates ingredients, you edit
PublishingUploads and schedules to YouTubeManual upload
Bring your own keyEvery provider, per userNot offered
Script ceilingTwenty minutes and past it, section by sectionCapped at sixteen
Public APISame tool contract over RESTNone
AutonomyOne message, the agent drives every toolEvery tool a manual step
Pick your models. Change your mind later.
Adding a model is one adapter file and one registry entry — no pipeline change, no redeploy of anything else. These are wired up now.
Language models
OpenRouterAnthropicOpenAI
Voice
ElevenLabsOpenAI TTSCartesia
Images
fal.aiHiggsfieldOpenAI GPT-ImageGoogle Imagen
Assembly
CreatomateRemotion (self-hosted)FFmpeg (self-hosted)
Try it on one video
Type a topic and watch the stages tick over. If the result is not worth the credits, you will know inside a few minutes.