How Cinovix turns a prompt into a film

1. Step 01

   Describe your story

   You write what the film is about in plain language and choose a length between 30 seconds and 10 minutes and one of four visual styles. No shot list, no storyboard, no script is required at this point.

2. Step 02

   Answer the questionnaire

   Cinovix reads the prompt and asks only the questions it actually needs answered — who the film is for, what the tone is, what must appear. The questionnaire is adaptive, so a clear prompt produces fewer questions than a vague one.

3. Step 03

   Storyboard and chapter plan

   The prompt becomes a concept, a scene count derived from the requested runtime, and a description per scene. Films past roughly three minutes are first split into five to seven chapters, each with its own setup, tension and payoff, so a 10-minute film stays one story instead of 60 disconnected beats.

4. Step 04

   Generate the master reference

   Before any video exists, Cinovix generates a single master reference image of the subject and reads it back to extract what the subject actually looks like — face, build, wardrobe — plus the look of the environment. That description, taken from real pixels rather than from the prompt, becomes the film's authority on appearance.

5. Step 05

   Write the scene scripts

   Every scene gets its own image and motion description, written in parallel so a long film does not take proportionally longer to plan. Each scene is also classified as taking place in the master reference's setting or in a new one, which decides how its anchor image gets made.

6. Step 06

   Run the continuity pass

   A continuity supervisor reads all planned scenes as one set — not one at a time — and validates wardrobe, weather, lighting and how each scene's closing pose hands off to the next scene's opening. Scenes that break continuity are rewritten or sent back for regeneration before anything paid is rendered.

7. Step 07

   Generate anchor images

   Each scene gets a still anchor frame. Scenes set in the master location are generated against the master image so the place stays identical; scenes in a new location get a fresh image that keeps the subject's identity but establishes a different place. Every anchor is inspected before it is allowed to become video.

8. Step 08

   Render every scene

   Scenes render from their anchor image, many at a time rather than one after another. Each finished clip is checked for coherence against what the scene was supposed to show, and scenes longer than a single generation are split into sub-clips that all stay anchored to the same reference.

9. Step 09

   Master the film

   Narration is generated and timed against the cut, an instrumental score is generated and mapped to each chapter's tension, and the mix uses sidechain ducking so music drops under the voice. Then the film is levelled, graded, stitched into one continuous cut and exported twice — a high-quality master and a web version — with a thumbnail.

The film arrives finished, not as raw material. If you want to change something, Director's Cut shows every scene with its status, duration and preview — trim it, reorder it, regenerate it or enhance it, and the film re-stitches itself around the change instead of re-rendering from scratch.

More answers in the Cinovix FAQ, background on why long-form AI video is hard, and definitions in the AI video glossary.