InferenceConjure
From data to discovery

One sentence in. A finished effect out.

InferenceConjure is an accelerator for complex creative software. Describe an effect in plain words, and it builds the scene, simulates it, lights and renders it, grades it, scores it and cuts the reel. Every shot it makes teaches the next one.

Now showing

The Screening Room

Every piece below began as a single sentence. Choose one: the house lights dim, the curtain opens, and it plays.

A vintage theatre with red velvet curtains closed
Choose a piece One sentence in.
How it works

From a sentence to a shot

Professional effects software takes years to learn. InferenceConjure carries that learning for you. It works in SideFX Houdini and Maxon Redshift, the tools film studios use, and runs every step itself.

  1. I

    The sentence

    You say what you want to see, in plain words.

    “A wrecking ball smashes through a concrete column…”
  2. II

    The shot plan

    Beats, camera, lens, light, palette and timing are written down before anything is built.

    tension → impact → settle
  3. III

    The scene

    The scene is assembled node by node, with every setting checked, because a wrong setting usually fails silently.

    dozens of silent failures, each now a known fix
  4. IV

    The simulation

    Real physics: sand, smoke, rigid bodies, cloth, particles, fire.

    1.24 M sand grains · 216 concrete pieces · 37,000-point velvet
  5. V

    The render

    Lit and rendered with a cinema colour pipeline, and every frame checked against the renderer's own log.

    ACEScg · motion blur · depth of field
  6. VI

    Grade and sound

    Graded like film. Sound effects are timed to the moments the simulation says they happened.

    impact on the frame the column breaks
  7. VII

    The reel

    A finished piece with a master copy, ready to share in 16:9 or 9:16.

    ProRes master · web and vertical cuts
  8. →

    See the results

    Every step above, finished, in the Screening Room.

    Enter the Screening Room
How it learns

Experience builds the mind

Every run is written down: what was asked, what was built, what failed and why it failed. That record becomes the Hebbian Corpus, and from the corpus a model is trained to do the work itself, on our own machines.

Diagram: Conjure studies, tests, journals and exports what it learns; the Hebbian Corpus pipeline collects, tokenizes, trains, embeds, serves and keeps learning
Conjure does the work and writes it down. The Hebbian Corpus turns that record into a mind.

Measured, not guessed

A colour ramp is set from the measured speeds of the particles, and a collision only counts when the pieces actually move. Each finding is written down where the next shot can use it.

Mistakes are kept

Every dead end is recorded with its cause and its fix. The failures teach as much as the shots that worked, so the model learns the whole craft, not only the highlights.

Sovereign by design

The software, the renders, the record and the models all run on Webspinner's own machines. Your work and what it teaches stay with you.

Beyond video

Engrams and Hebbian Projections

An Engram is your intent, codified: the story, the look, the sound and the beats. InferenceConjure turns an Engram into a Hebbian Projection, a moving image produced by a learned mind rather than recorded by a camera. Refine the Engram and project it again.

The Studio · coming soon
EngramWhat you codify: intent, story, look and sound
InferenceConjureBuilds, simulates, renders, grades and scores
Hebbian ProjectionThe finished moving image