What is the KURACONV three-question filter for AI frames?

It is the internal test KURACONV applies to every AI-generated frame before approval, drawn from the São Paulo Blues director's note. The three questions are: Is there ether (atmosphere with presence)? Does the light reveal (narrative work, not just illumination)? Is there a deliberate detail (a director's choice, not a model default)? A frame that cannot answer all three does not ship.

How does KURACONV avoid generic AI content?

By directing AI rather than prompting it. Every frame passes the three-question filter, every creative decision runs through a 22-mind council against slop, and the studio refuses categories where the filter cannot operate: performance ads, stock photography, projects without a brief, and generic AI content.

What is Sentimagem and how does it connect to the frame test?

Sentimagem is the KURACONV method, built on three pillars: Presença (mood as brand), Engenharia (camera direction in motion, with AI as the new lens), and Narrativa (story engine first, image emerges). The three frame questions map directly onto them: ether is Presença, light that reveals is Engenharia, the deliberate detail is Narrativa.

Three Questions per Frame: How We Filter AI Slop

Most AI frames fail not because the model is weak, but because no one asked them to mean anything. At KURACONV, every frame, still or motion, passes a three-question test before it is approved. The questions came out of the director's note for São Paulo Blues, our 2026 original film and soundtrack, and they have since become the studio's filter against AI slop.

The test is short. Three questions, applied to every frame, every project, every client. If a frame cannot answer all three, it does not ship. That is the entire method: a narrow gate at the end of a wide pipeline.

Question 1: Is there ether?

Ether is the word we use for atmosphere that has presence. Not weather, not haze, not a filter. A frame has ether when the air in it feels inhabited, when the mood persists for a beat after the eye moves on. Most AI output has the opposite quality: it evaporates the moment you stop looking at it.

Ether is testable. Cover the subject with your thumb. If the frame still holds a feeling, there is ether. If only the subject was carrying it, the frame is decoration, and decoration is the first symptom of slop.

Question 2: Does the light reveal?

Models default to illumination. We ask for revelation. Illumination makes pixels visible. Revelation does narrative work: it tells you where to look, what the character knows, what the scene is hiding. Light that reveals has a source, a direction, and a reason. Light that only illuminates has none of the three.

When we review a frame, we name the light source out loud. If we cannot name it, the model invented it, and an invented light source almost always lights an invented story. We send it back.

Question 3: Is there a deliberate detail?

Every approved frame must contain at least one detail a director chose, not one the model defaulted to. A specific object, a specific gesture, a specific edge of wardrobe, a specific imperfection. Defaults are how AI announces itself. Deliberate details are how a frame announces a human was in the room.

Ether: does the air hold a mood that survives the subject?
Light: can we name the source, the direction, and the narrative job?
Detail: what did the director choose that the model would not have chosen?

Why three questions, not thirty

A council of 22 minds runs against every creative decision at the studio, photographers and directors, philosophers and strategists. The council is broad because the work is broad. The frame test is narrow on purpose. Three questions are the most a director can hold in working memory while reviewing a sequence at speed. Any more, and the filter stops filtering.

The three questions also map cleanly to Sentimagem, the studio method. Ether is Presença: the mood is the brand. Light that reveals is Engenharia: direction is the eye, AI is the new lens. The deliberate detail is Narrativa: story engine first, motion follows, image emerges.

We direct AI. We do not prompt it. Prompt-and-pray produces frames the model is proud of. Direction produces frames the studio is proud of.

What the filter rules out

The three questions are also a quiet definition of what we do not make. No performance ads, no stock photography, no projects without a brief, no generic AI content. A frame that passes the test could not have come from prompt-and-pray, and a studio that ships only frames like that cannot, by construction, produce slop at volume.

This is why the work moves the way it does. Editorial AI Films of 15 to 89 seconds, editorial stills, editorial carousels, brand identity, original music on every project, no stock. Delivered in two to three weeks where traditional production takes three to six months. The speed is not the point. The filter is the point. The speed is what is left when the filter is doing its job.

Use it on your own work

The test is portable. Take any AI frame, yours or someone else's, and ask the three questions in order. Ether, light, detail. If two pass and one fails, the frame is salvageable through direction. If two fail, the prompt was the director, and the prompt is not qualified for the job.

Three questions stand between a frame and the feed: is there ether, does the light reveal, is there a deliberate detail. A frame that cannot answer all three is slop, no matter how clean the render.