Back to feed

Creating a Mystical Series with Neural Networks

Creating a mystical mini-series on the Black Sea using neural networks and manual editing.

Idea and Character Preparation

The project "Lighthouse Keeper" tells the story of a woman arriving at a lighthouse where her grandfather served thirty years ago. Despite his funeral, someone lights up the light every night at 3:17 AM. The author aimed to create a series with a clear hook and world rules, rather than just a collection of beautiful shots.

The Nano Banana model was used for image generation. Full-body character cards from four angles and close-ups of faces were created for each hero. Approximately twelve images with locations and interiors were prepared after several revisions.

Storyboard review allowed identifying errors before video generation began. The neural network detected inaccuracies in lighting and character placement, helping save resources on regeneration.

Video Generation and Sound Work

Video material was created in Vidu Q4 at 540p resolution. Prompts specified ARRI Alexa 35 camera parameters and Kodak Vision3 film settings to achieve realistic skin texture without the plastic effect.

The neural network generates background sounds independently, but Ukrainian speech came out with a strong accent. Therefore, all dialogues were re-dubbed with live voices, and phrase duration was sped up by 15–18% to sync with lip movement.

Editing was performed via ffmpeg. Cuts, fades, and fog transitions were used. Sea sound was lowered under dialogue to improve comprehension.

Economics and Tool Comparison

The full cycle of creating an episode from script to editing took two to three hours. The cost was about four to five dollars, including expenses for image and video generator APIs.

Vidu showed an advantage in cost and quality of live skin transmission for dialogue scenes compared to competitors. However, other solutions are better suited for complex camera movements.

5.8K views

More from this channel Metavershche and AIshche

Similar in this category Technology