Intent-to-video pipeline
Turns creative intent into images, then video, on Wan and LTX, with character orthos, render profiles, and a continuity ledger for shot-to-shot consistency.
Try Compo ↗Password on request
From a sentence to a clip, in Compo
01Describe the shot in a sentence, or start from a photo.
02The guide turns it into a plan (clip or still, length, size) and queues the take on the home-lab machines.
03The take comes back into the chat. Keep it, reject it, or ask for another.
How it runs
Why build it
Removing the friction between a creative’s intent and what they make. Building it end to end shows where generative video still breaks.
From still to shot
My photographs, directed as six-second shots.
I wrote each photograph a shot (a subject, what happens to it, the camera, the sound), and Compo rendered it on LTX 2.5. Drag across the photograph to see what the model added.
Drag to compare the photograph with the shot
- Subject
- a wide shot from above the bank of a broad river at blue hour. In the middle of the river a small white pedal boat carries a lamp and people in silhouette, pedalling slowly. Behind it a dark wooded hill rises against a deep blue sky, a lit lookout tower on its crest; warm streetlights line the far bank, lit buildings sit on the right, and golden reflections streak the water on the left.
- What happens
- The boat glides at an easy pace to the right, its lamp swinging slightly; a V-shaped wake opens behind it and rolls outward in thin lines. As the wake reaches the golden reflections, they break into trembling pieces and slowly gather again. Near the bank, a moored pedal boat rocks gently as the ripples arrive. A few clouds drift across the blue sky.
- Camera
- The camera remains static.
- Sound
- the boat’s paddle wheel churning in a slow rhythm, water slapping its hull, laughter carrying across the water from the bank, and a tram bell far across the city.