Blocking

Plan where everyone stands, how big they are and where the camera is, once for the room and once per shot -- and let that drive the storyboard.

Blocking is working out who stands where in a shot. Doing it with words alone ("Mira crosses to the desk") leaves the picture model to guess, and it guesses differently every time: a character is the wrong size, stands in the wrong place, or the newsroom turns into a different room. The Blocking tool lets you decide it once, by hand, and then hands that decision to the model as a guide.

It has four parts, and the words below are the ones you will see.

The Blocking Set -- the room

Every location can have one Blocking Set: a simple model of the room, shared by every shot that happens there. You build it once.

A new set starts as the smallest useful thing -- one camera looking straight at one flat backdrop that shows the location's picture. Characters will stand in front of that picture, and a close-up is just a narrower view from the same camera. From there you add:

  • Objects -- planes (a wall, a screen, a floor), a box, a desk, a table, a stool, chairs of three kinds, cylinders and a sphere. Each has a name ("news desk"), a position, a turn, a size and a height off the floor. They may overlap freely; there is no collision.
  • Blocking marks -- named places on the floor where a character stands ("anchor", "weather board"). A mark has a facing too.
  • Cameras -- a position, a height, a direction, a tilt and a field of view. A set can hold as many as you like ("wide from the front", "single at the desk").
  • Pictures on surfaces -- any picture in the location's media can be put on a plane, stretched or cropped to cover it. The default backdrop is one. A shot can swap it (the night storyboard instead of the day picture).

You edit all of it on a top-down floor plan beside a 3D view: drag things on the plan, drag the round handle to turn them, and type exact numbers (sizes, heights, seat heights, camera angles) in the fields below. The 3D view shows what a camera sees, or a free view you can orbit and pan to look around. When the free view shows a spot you like, Make a camera here turns it into a new camera. To match a camera to the location's picture, select it and choose Fit to the picture: draw lines along the floor and walls of the picture, set the camera's height and field of view, and save. The camera's tilt is set from the horizon the lines imply, and the backdrop is re-sized and re-placed so the picture fills that camera's view again -- so the camera and the picture always agree.

Lengths can be typed in metres (1.68) or in feet and inches (5'6.3"), and either is understood when pasted. The same length editor is used on a character's Height and Width.

Fitting a camera to the picture

A camera's height, tilt and field of view decide how the room picture lines up with the plan. Fit with lines works them out from the picture itself: draw two lines along the floor or wall edges that run one way, two along the edges that run the other way (the two ways should be at right angles in the room), and one vertical line on something whose real height you know -- "the desk top is 0.8 m". Tap the two ends of each line and drag the dots to fit them exactly. The tool gives the field of view and the tilt from the edges, and the camera height from the known size; nothing in a picture gives scale, which is why one height is asked for. Apply to the camera writes all three and re-sizes the backdrop so it fills the camera again. Fit to the picture (sliders) stays as the manual alternative.

Blocking a shot -- who is where

Open a shot and choose Block this shot. The room is the location's Blocking Set; what is new is the people.

  • Stand-ins are the characters: a plain capsule body with a head ball and a painted face, so you can see which way each is turned. Each is sized from its character's Height and Width (or from the height class and body type when those are empty). A stand-in sits on a mark when dropped near one, and you can turn its body and its head.
  • Seated is worked out for you. Put a stand-in over a chair, stool or armchair and it is drawn sitting; move it off and it stands again. You never set a "sitting" switch.
  • The camera for the shot is chosen from the set's cameras. You can add a camera or a prop that exists only in this shot.
  • Changing the set for one shot needs no special step: just drag the chair. The shot then holds only the changed values and shows changed in this shot beside it, with Reset to the set to undo. Everything you did not change keeps following the set -- later you resize the chair in the set and every shot gets the new size, while the position stays this shot's own. If the set object a change refers to is later deleted, the change is dropped and the tool tells you.

Blocking the scene -- who is where when it starts

Open a scene's media and choose Block this scene. This is the same editor, one layer up: the stand-ins you place here are where everyone stands when the scene begins, and every shot of the scene inherits them with no edits. A shot then changes only what moves.

Blocking is built in layers, each standing on the one below:

Layer Holds
The location's Blocking Set The room: furniture, marks, cameras
The scene Who is where when the scene starts
Each shot Only what changed in that shot, on top of the layer it is Based on
  • Based on. A new shot is based on the previous shot of its scene (the first shot, on the scene). You can pick a different one from Based on in the shot's editor. The choice is stored, so reordering shots never changes what an approved shot inherits. A shot can only build on the scene or on another shot of the same scene and the same location, and never on a shot that already builds on it.
  • A move reaches the shots after it, not the ones before. If Sally walks to the desk in shot 1.4, shots 1.5 and 1.6 keep her there and shots 1.1 to 1.3 do not change.
  • Where a value comes from is shown beside it: from the set, from the scene, changed in shot 1.4, or changed in this shot; Reset returns the value the layer below gives.
  • Editing a shot makes the captures of that shot and of the shots built on it older than their layout, and no others.
  • Show cameras / Hide cameras hides the camera bodies on the plan and in the free view, so they do not crowd a busy set.

Capture -- the guide pictures

Capture renders the shot from its camera and puts three pictures in the shot's media:

  • the Blocking View (named in the gallery with the camera's name when it is not the default camera) -- the room as the camera sees it, backdrop picture included, with the stand-ins in grey;
  • the Blocking Depth Map -- how far away everything is, near bright and far dark;
  • the Blocking Room -- the same camera with no stand-ins: the part of the location picture the shot sees. A close-up is given this as its location reference in the recipe instead of the whole room. (Unlike the other two it is an ordinary picture, offered as a reference.)

Both are tagged so they are easy to find, the depth map is linked to the view it came from, and a small snapshot of the whole layout is stored with them so you can always tell what a picture was made from. If you change the layout afterwards, the tool tells you the captures are older than the layout.

The size comes from the recipe that will use them; when that is not known it is the size H3 takes as a control picture (864 x 480, or the same shape).

Blocking words

At capture the tool writes the blocking words: who is facing which way as the camera sees it ("in a three-quarter view, facing the right side of the frame"), who is sitting on what, who turns to face whom. They are always in terms of the frame -- left and right of the picture, never degrees. They are shown under the capture and you can edit them. The shot's Character blocking text is rewritten from the stand-ins and marks so the prompts and the assistant see the blocking without opening the tool.

From blocking to storyboard

The recipe References to Video (MiniMax H3, Depth Control) takes the Blocking View as both the guide for where everyone stands and the source of the depth, the character sheets as reference pictures, and the blocking words in the prompt. In the Recipe panel for a shot that has a capture, the Blocking View fills its own slot for you -- and only that slot: a capture is a guide, never offered as a reference picture, because the model would draw it as one. It runs on a local ComfyUI only. Run it inside an experiment for the shot, look at the frames, and pick the one you want as the storyboard.

A walkthrough

You do You see
Open the newsroom location and choose Blocking Set A plan and a 3D view: one camera facing the newsroom picture
Add a desk, put a plane for the weather screen, add marks "anchor" and "weather board" Named things on the plan; the 3D view matches
Add a second camera for the single A second wedge on the plan; it appears in the shot's camera list
Open the scene and choose Block this scene, put the anchor on "anchor" and the presenter on "weather board" Two grey stand-ins; the anchor sits on the chair; every shot of the scene starts with them there
Open the two-shot and choose Block this shot Based on shows the scene; both stand-ins are already there, marked from the scene
In a later shot, drag the presenter to the desk changed in this shot beside her; the shots after it keep her at the desk, the shots before it do not
Turn the presenter to face the anchor The words under the capture update to "facing ... " in frame terms
Choose Capture A Blocking View, a Blocking Depth Map and a Blocking Room in the shot's media
On the set's camera choose Fit with lines and draw the edges and the desk height Field of view, tilt and camera height filled in; the backdrop fills the camera again
Open the single and pick the second camera; nudge the chair "changed in this shot" beside the chair; the two-shot still has the chair where the set has it
Run the depth-control recipe from the Recipe panel inside an experiment Both characters on their marks at the right relative size, the newsroom kept

What it does not do (yet)

Pictures cannot yet be projected from a camera onto the room (a picture set to project is shown as cover); there is no movement over time; stand-ins are not 3D character models; and the assistant does not yet propose a blocking. A rear three-quarter view is often drawn as a profile, and the tool warns when you pick one.