CUTLINE features

CUTLINE turns a single sentence into a finished short video: script, voiceover, images, motion, and subtitles. No templates, no storyboards. The system acts as both director and editor. This page explains what it does, what makes it different, and what you can control.

All of the following is a factual description of how CUTLINE works. Use the side menu to jump to a section.

What makes CUTLINE different

Most video tools ask you to pick a template, upload clips, or build a storyboard. CUTLINE does not. You give one sentence describing what you want the video to do (e.g. “Explain why coffee makes you feel awake in 30 seconds” or “A short product teaser for our new API”). The system then runs a fixed pipeline: it infers intent, plans the narrative, breaks it into shots, writes the script, generates voiceover, sources or generates images for each shot, adds motion and subtitles, and renders a single MP4. There are no branches or optional paths: every slideshow video goes through the same sequence, so behavior is predictable and debuggable.

You can optionally set duration (10-60 seconds), upload assets (logo, product photos), pass brand colors, or choose a different AI model for that job. But the core value is: one sentence in, one video out, with no manual editing steps required.

One sentence to full video

The only required input is a single sentence. The system uses it to decide who the video is for, what it’s trying to achieve (explain, persuade, tease, etc.), the tone (professional, casual), how complex the topic is, and how long the video should be. If you don’t specify a duration, it is inferred. Everything downstream (narrative, shots, script, pacing, and visuals) is derived from that intent. So you don’t have to write a script, pick images, or time subtitles yourself; the pipeline does it.

End-to-end pipeline (no handoffs)

The video is built in one linear pipeline of 12 stages. Each stage uses the output of the previous one. The stages are: Intent → Narrative → Shots → Script → Subtitles (draft) → TTS (voice) → Subtitle refine → Motion → Asset analysis (if you uploaded assets) → Visuals → Image sourcing → Remotion render. If any stage fails, the whole job fails (with retries for temporary API errors). There are no manual handoffs: you submit once and the worker runs the full pipeline until the MP4 is written.

Two output modes: slideshow and talking object

CUTLINE has two ways to produce a video. Both start from your one sentence and the same early stages (Intent, Narrative, Shots, Script).

AI script and natural voiceover

From the intent, the system plans the narrative (e.g. hook → explanation → conclusion), then breaks it into 8-12 shots and writes the actual words for each shot. That script is sent to ElevenLabs or PlayHT (you choose one provider via TTS_PROVIDER). They return audio and, where supported, word-level timings. So you get a single continuous voiceover that matches the tone and length of your request, with no manual script writing or voice recording.

Image sourcing and motion

Every shot needs an image. For each shot, the AI suggests a search query or image prompt. The system then tries, in order: Unsplash (stock), DALL·E 3 (AI-generated), Pexels (stock), and if needed a simplified query. If everything fails, a placeholder is used so the video still renders. If you uploaded product photos or a logo, the asset-analysis stage can assign some shots to your uploads and use extracted colors and layout hints. Motion (scale, pan, zoom) per shot is computed from the shot list and applied in the Remotion composition. No AI for motion, just rules based on shot purpose and order.

Auto subtitles synced to voice

The script is chunked into subtitle-sized segments and given an initial timing estimate. After TTS, word-level timings from the provider are used to refine when each subtitle appears and disappears, so captions stay in sync with what is actually spoken. You don’t have to time or edit subtitles manually.

Optional inputs: duration, assets, brand, model

Besides the required one-sentence input, you can send:

• durationSeconds: A number between 10 and 60. Overrides the inferred duration for both slideshow and talking_object.
• assetIds: IDs from POST /api/assets/upload. The pipeline runs asset analysis (logo, product photos) and uses the results in the visual spec and image sourcing.
• brandColors: primary (and optional secondary) hex colors passed into the visual spec.
• mode: "slideshow" or "talking_object". Default is slideshow.
• textModel: An OpenRouter model ID. Overrides the default for that job only; every LLM stage in that job uses this model.

All of these are optional. The system works with just the one sentence if that’s all you provide.

Background jobs and polling

Video generation can take a minute or more, so it runs in the background. When you submit, the API adds a job to a BullMQ queue (Redis) and returns a jobId immediately. A separate worker process (npm run worker) picks up jobs and runs the full pipeline. The UI polls GET /api/generate/[jobId] every few seconds until the status is completed or failed, then shows the video or the error. You never wait on the server for the full render: submit, get an ID, poll until done.

Rendered videos and temp assets are cleaned up automatically based on VIDEO_RETENTION_HOURS and related settings, so you don’t have to manually delete old files.