How did KURACONV shoot the falcon at 300 km/h?

They did not chase it. A camera moving at 300 km/h is impossible to fake convincingly with AI, so the team either mounted the camera to the bird's trajectory and banked with it, or held the frame still and let the falcon tear past a fixed horizon. The middle ground, a tracking shot at full speed, was abandoned because that is where AI invents physics that do not exist.

What is the one-anchor rule in AI cinematography?

It is KURACONV's universal shot rule, born on Dubai Falcon. One anchor in the frame, one subject, one cause driving the motion, and one practical light source. Stacking more than one of any of these is when AI models start guessing and the image begins to look generated instead of directed.

Can AI films include shots that are physically impossible to capture?

Yes, and KURACONV argues they should. Dubai Falcon includes the POV of a single raindrop falling over Dubai, the falcon's shadow treated as a living entity, and visible heat coming off the wings. These shots are unbuildable on a real set, which is the reason the film exists. In the words of co-founder Diogo Felipe Silva, the laws of physics do not apply to AI.

Anatomy of a Scene: Dubai Falcon

Dubai Falcon is a cinematic AI short built around a bird the camera could never follow. It is the project where KURACONV stopped simulating a real camera and started directing AI as its own medium. Three decisions shaped the cut. Each one became a rule the studio now carries into every film it directs.

Decision 1: We stopped chasing the falcon

The first pass tried to track the bird at full speed. A camera moving at 300 km/h does not exist in the physical world, and models trained on real footage do not know how to invent one. The image collapsed into motion soup. The fix was a binary choice we now apply to every velocity shot.

Mount the camera to the movement and bank with the bird, so frame and subject share one trajectory.
Hold the camera still and let the falcon tear past the lens, so speed is read against a fixed horizon.
Never average the two. The middle option is where AI invents physics that do not exist.

Direction over generation begins here. We did not prompt harder. We changed the shot.

Decision 2: One anchor, one subject, one cause, one practical light

The studio's universal shot rule was born on this project. Every frame in Dubai Falcon obeys it. One anchor holds the geometry, a dune ridge, a falconer's glove, a horizon line. One subject carries the scene. One cause drives the motion, wind, dive, breath. One practical light source, the desert sun, a lantern, a window of sky. Stack more than one of any of these and the model starts guessing. Strip it back, and the image stops looking like AI and starts looking like cinema.

This is the Engenharia pillar of Sentimagem in operation. AI is the new lens. Direction is the eye. The council of 22 minds filters every shot against this rule before a single frame is generated.

Decision 3: We kept the impossible shots impossible

The first two decisions are about respecting physics. The third inverts the test. If a shot could never be captured by a physical camera, we do not hide that. We lean in. Dubai Falcon holds three sequences designed to be unbuildable on a real set:

The POV of a single raindrop falling over Dubai, before it meets the sand.
The falcon's shadow treated as a living entity, moving across the desert with its own intent.
Heat coming off the wings, made visible as a layer between bird and sky.

These are not effects. They are the reason the film exists. A traditional crew would cut them in pre-production. An AI-directed pipeline can hold them, if the direction is willing to commit.

The laws of physics do not apply to AI., Diogo Felipe Silva

What the Lab study documents

The full breakdown on the KURACONV Lab shows the shotlist, the identity locks, the continuity filter for the falcon across scenes, the original score written for the cut, and the failed first passes. The failures matter. They are the reason the rules exist.

Dubai Falcon was directed and finished inside the studio's standard two to three week window, with GPT-Image-2, Seedance, Higgsfield, Kling and Veo running under one direction, not five separate prompts. The piece sits alongside Everest and São Paulo Blues as part of the studio's selected work.

In AI cinematography, respect physics where it grounds the image and break it only where the shot could never exist any other way. One anchor, one subject, one cause, one practical light.