How Scene Explorer writes the prompt

You don’t write the prompt for a scene render. Cask Film writes it from the scene, and sends it with a guide and your reference images. Knowing what goes in tells you which control to change when a render misses.

What the model receives

For each shot:

1. The guide video: the shot’s range rendered in guide colours on a grey ground, always sharp. Its short side is 1080 px when you render at 1080p or above, otherwise 768 px.
2. The start frame, if there is one. Shot 1 uses the test frame you marked Use as start frame. A shot after a continue uses the last frame of the shot before. Models with Seamless joins take it as their start frame. Models with Soft joins take it as the first reference.
3. The reference images of the objects in the shot.
4. The look reference: the test frame you pinned, or the newest test frame when the model needs more references than the objects supply.

A Test Frame sends the guide still (768 px short side) as image 1, then the references of the objects in frame at the playhead. It uses the aspect ratio closest to your project’s.

The guide is previs

The prompt opens by telling the model that the guide is a 3D previs blockout. It sets the layout and the motion, not the look. The model is told to follow its composition, camera move, timing and where each object sits, and to treat each solid colour as a name tag only: build the real thing with its own colours, materials and detail.

The blockout places only the key objects. The prompt asks for the whole location around them: background, surroundings, set dressing and sky in every part of the frame. A Test Frame asks for the exact composition, angle and framing of image 1 instead of a camera move.

The sentences that follow

After the opening, the prompt runs in this order.

The Vibe, word for word from the Environment tab. The default is “Cinematic and photorealistic, motivated light, real-world materials, subtle film grain.”

A continuity line, on soft-join models only: continue exactly from the start frame.

One sentence per object, named by guide colour, shape and place in frame, for example “The orange person on the left in the foreground is a tall woman in a rust raincoat (reference image 1).” The place comes from where the object sits in the shot camera’s frame. The ending depends on what you gave it:

• a description, with its reference images named,
• reference images only: the object is what the image shows,
• neither: a real version of the shape that fits the scene.

Two objects in the same colour become “red” and “red 2”. Relabel a colour in the Object tab to change the word the prompt uses.

The light, from the sky and sun: “clear daylight”, “golden hour”, “overcast soft light”, “night” or “interior studio lighting”. The sun is “low sun” under 15°, “high overhead sun” over 60°, and is placed relative to the camera: ahead of camera (backlit), camera left, camera right or behind the camera. Studio has no sun.

The lens and the move, for example “Shot on a 35mm lens at f/2.8, slow push-in.” Cask Film names the move from the largest change in the shot: push-in, pull-out, truck, pedestal, pan, tilt, zoom, or “locked-off static camera”. Handheld adds “handheld drift”. A Test Frame has no move.

The focus, stated every time:

• f/2.8 or wider: shallow depth of field, focus on the named object or distance, everything nearer or farther falling into soft blur.
• f/8 or narrower: deep focus, sharp from foreground to background.
• Between the two: moderate depth of field, distant parts soften slightly.
• A focus change during the shot adds a rack focus to the new distance.

The look reference, when there is one: match the look of the pinned test frame.

How references are named

Each model family names its inputs differently, and the prompt follows the model: “@Image1” on Seedance, “reference image 1” on MiniMax, and “image 2” onward on image models, where the guide is image 1.

Long prompts are cut from the end

A shot’s prompt stops at 2000 characters. Cask Film drops object sentences from the end of the list first, then shortens the Vibe, so the opening, the light, the lens and the focus survive. A Test Frame prompt has no limit.

Which control changes what

The render gets this wrong	Change
Layout, framing, timing	The set, the camera and their keys. The guide video carries them.
What an object is	Its Description and Reference images
Mood, era, weather, the world beyond the set	Vibe
Time of day, light direction	Sky and Sun
Depth of field	Aperture and Focus
Overall look between shots	Pin a test frame with Use as reference