Shots

Shot editing -- properties, media, and sequencer.

A shot is the most granular unit of production -- a single image or video clip with a specific camera angle, characters, dialogue lines, and generated media.

Shots live within scenes, and all of an episode's scenes and shots form a single table of contents. There are three ways to reach a particular shot: Shots in Scene, at the top of the Shot group in the sidebar, which lists the shots of the scene you have open; the episode's Storyboard or Shot List, which cover every shot in the episode; or the prev/next control in the page header, which steps through the episode in order. See Navigation.

Shots in Scene

Shots in Scene, the first row of the Shot group, is the episode's Shot List narrowed to the scene you have open -- it answers "what else is in this scene". It works exactly like the Shot List: the same List, Detailed and Reel (the shots' clips, big, one line of caption each) views, and, in List and Detailed, the same dragging, inserting and shelving and the same AI controls on the scene heading (drafting the shots and reviewing their camera direction). Click a shot to open it.

If no scene is open yet, Shots in Scene takes you to Scenes in Episode first, and picking a scene there brings you straight back to its shots.

A scene with no shots yet shows its heading with the control that drafts them -- see drafting a scene's shots.

Shots are reviewed and reordered across the episode's shot views: the Storyboard (a frame grid), and the Shot List, which presents the same scenes and shots three ways -- List (pacing and running time), Detailed (the same rows carrying a thumbnail and the inputs going into each shot before generation) and Reel (every shot's picture from the sequencer, large, under its scene heading, with its number, name and length underneath; tap a frame to play the shot's clip or dialogue, tap the name to open the shot) -- chosen with the view buttons beside its heading. Shots can be reordered by dragging their handle, inserted between others with the + beside it -- identically in List and Detailed (Reel is for watching and opening, not editing) -- and shelved, duplicated, or deleted from the shot's menu -- the order determines playback order in the sequencer.

A shot that has no name is shown by its number: its position in the scene, counted from 1. Numbers follow the order, so inserting or moving a shot renumbers the ones after it. The number is how you refer to a shot when talking to the assistant too: "make shot 3 a close-up" means the third shot of the scene you are on.

Every one of a shot's pages is titled with the shot's scene and position and then its name -- "1.12 Eyes power on" is the twelfth shot of scene 1 -- the same two numbers the Shot List shows against that scene and shot, so a name like "Eyes power on" never leaves you guessing where in the episode you are.

The shot has three pages, listed under Shot in the sidebar. They open on whichever shot you last had open -- the breadcrumb at the top of the page names it:

Properties -- everything that defines the shot, in one place.

Media -- all images and video generated for this shot, and where you run operations to make more.

Sequencer -- the episode sequencer opened on this shot: the same editor, the same tracks, with playback and render controls.

Shelving a shot

Shelve, on a shot's menu in any of the episode's shot views, sets the shot aside without deleting it. It keeps its place in the running order and everything it holds -- its media, its properties, its length -- but it leaves the cut: it shows no picture, adds nothing to the running time, and the episode plan stops asking you to finish work on it. The shot's page titles read "(shelved)" while it is set aside, and the plan step that would have covered it says "1 shot shelved" so nothing disappears without a trace. Unshelve puts it straight back, exactly as it was.

The first shot of a scene cannot be shelved -- there is no earlier shot to take on its time -- so the menu item is offered but disabled there, with the reason.

Shelving is a fact about the shot. Approving a criterion in the episode plan is a different thing: your decision about one step of the work. See The production pipeline.

Planning a scene's shots

Each scene heading in the shot views carries one AI control, and what it does depends on where that scene has got to.

Deciding what a scene's shots are and deciding how each one is framed is one decision -- a shot is its framing -- so one button does both. It reads the scene's screenplay text, the scene's emotional tone, key story beats, pacing and energy curve, and whatever its shots already hold, and works across the whole scene at once.

  • A scene with no shots yet offers Plan shots. It builds the shot list from the screenplay in screenplay order, and then fills in as much of each new shot as it can -- name, description, framing, camera and blocking. A scene that had nothing a moment ago has no work of yours to protect, so a first pass goes as far as it can in one click.
  • A scene that already has shots offers Replan shots. Same action; the different word only reminds you that you have done this before. It protects the work you have already done:
    • a property you left empty is filled in straight away;
    • a property that already has a value is only ever proposed, never overwritten. Your value and its suggestion sit side by side on that shot with Accept and Reject, and every suggestion carries the reason it is making it. That is also how the framing gets reviewed -- see below;
    • a shot's length is only ever proposed, never changed for you;
    • a detail you typed that the screenplay does not happen to mention is simply left alone.

It never adds a shot to a scene that has some. Replanning does not grow the scene, however many times you run it. If the AI thinks the screenplay covers a beat none of your shots does, it says so on the result line -- Missing shot: Close up on Sally -- with what it would frame and an Add. Nothing appears until you press it. It also never removes a shot, never reorders the shots you have, and leaves a shot the screenplay no longer mentions exactly where it is.

Replanning does not duplicate what you have, either. The AI is told which shots the scene already holds before it re-reads the screenplay, so a shot it describes differently -- or that you have since renamed -- is still recognised as the same shot and reviewed in place. Run it twice on a scene you have not touched and it offers nothing new at all.

Either way it runs on one scene at a time and only when you ask it to, and it reports progress as it works. Afterwards, step through the shots to review what it wrote -- and to accept or reject whatever it proposed.

Dialogue that has been voiced but never placed

Generating a line's audio does not decide when it is heard. If you have generated a scene's dialogue but never put it on the timeline, Plan shots does that for you: it tells the AI how long each line runs, asks which shot holds it and where inside it, gives each shot enough length to contain its own dialogue, and lays the lines down in screenplay order with a gap between them.

Replan shots does the same thing for a scene that already has shots, so you get the same result whichever of the two the scene offers you. The one difference is that it works with the shot lengths you already have rather than re-timing them: it will not stretch a shot you have already cut, so a long line can run past the cut it starts in. Drag it, or lengthen the shot, if that is not what you want.

Two things it will never do:

  • It never moves dialogue you have already placed. Only a line with no position gets one. A line you dragged stays exactly where you dragged it, however many times you replan.
  • It is a starting arrangement, not a rule. A line belongs to the scene's timeline, not to a shot, so you can drag it across a cut or outside every shot and it stays there. Nothing snaps it back.

A line the AI does not assign to any shot is left unplaced rather than guessed at.

When the screenplay names a character, location, or prop you have not created yet, the Shot List's Detailed presentation offers to create it.

How the framing gets reviewed

Settling how each shot is framed is the step between having a shot list and drawing storyboards: a storyboard composes whatever the camera direction says, so framing nobody has looked at means boards you draw twice.

There is no separate control for it. Replan shots is the review -- on a scene that already has shots it reads the whole scene at once (the scene's emotional tone, key story beats, pacing and energy curve, plus every shot's current camera fields and how long each shot is on screen) and comes back with suggestions per shot. What it looks for is the craft that only shows up across a sequence:

  • Shot sizes that vary. Two neighbouring shots at the same size read as one shot with a jump in it, and a scene of all wides is a scene the audience watches from one seat.
  • Establish, then go closer. Once we know where we are, the shots can work inwards as the scene narrows.
  • The scene's energy curve. Where it builds, the framing tightens and the shots shorten; where it drops, they open out and breathe.
  • Holding on an emotional beat, cutting on an action. A realisation needs a frame it can sit in; a door slam wants the cut on the movement.

Every change to something you have written is a suggestion, and it comes with a reason. Each one waits on its shot as a proposed value with the reason under the field name, so you can see why it wants a close-up here rather than just that it does. Accept or reject them one field and one shot at a time. A field you left empty is simply filled in -- there is nothing of yours to replace, and a gap is an unfinished plan rather than a decision waiting to be made.

It may also propose a different length for a shot where the rhythm is wrong -- a held beat cut short, a cut that lands late. Lengths are only ever proposed. Accepting one lengthens the scene to fit it; accepting a shorter one never shortens the scene, because only you reduce a scene's length. See Scenes.

You can also ask the assistant to plan a scene's shots, and its suggestions arrive in the same place for you to accept or reject.

Media

The Media view is the shot's Media Gallery -- all images and video generated for this shot. Star an item to like it (a bookmark that floats it to the top); choose Set as storyboard or Set as rough clip on an item to make it the shot's storyboard or rough clip (and Set as finished clip for the full-resolution version, below). This is also where you generate: choose an operation in the Recipe panel, fill in its inputs, and run. Results land back in the gallery.

The Recipe panel follows the shot's stage

A shot goes through four stages -- Storyboard, Rough clip, Choose the rough clip, Finish -- and the Recipe panel loads the recipe for the shot's current stage. Choose another recipe lists the recipes for every stage, grouped by stage: the current stage's group first, the later stages after it under their names, and the stages you have signed off last. Nothing is hidden because of where the shot is.

Setting media in the gallery and running a recipe never approve anything: a run may be an experiment. You approve the shot's storyboard, rough clip and finished clip yourself, on the shot's Properties page, on the Shot media card (see below). If a recipe needs something the shot lacks, such as a storyboard or a voice sample, the run says what is missing and nothing is run.

The shot's picture

A shot shows one picture wherever the episode is drawn -- the Storyboard view, the Shot List, and the Sequencer. Which one it shows comes from three levels, and you set the top two yourself from this gallery:

  1. The finished clip. Once The Finish has upscaled the rough clip, the finished clip plays in its place. You can also set it yourself with Set as finished clip.
  2. The rough clip. Set as rough clip on an item makes it the chosen clip for this shot. It wins over the storyboard and everything below.
  3. The storyboard. Set as storyboard on an item makes it the shot's picture whenever there is no rough clip -- the frame that carries the shot through boarding, pacing and a rough cut.
  4. What gets borrowed. With neither set, nothing goes black: the episode borrows a picture of whoever speaks over the shot -- all of them, sharing the frame, when more than one does -- and falls back to a picture of the shot's location. If there is nothing to borrow, the shot is drawn as a labelled placeholder telling you what to add.

These actions are on every item in the gallery, on all of its surfaces: the card menu in grid view, the row menu in list view, the right-click menu in the graph view, and the buttons in the fullscreen viewer. Setting one again removes it -- Remove as storyboard, Remove as rough clip, Remove as finished clip -- and the shot drops back to the next level down.

You can also choose or clear them on the shot's Properties page: the Shot media card shows Storyboard and Rough clip as the chosen picture with an x in its corner to clear it. Tap the picture to see it full screen; when nothing is chosen, tap the empty slot to pick from this shot's own gallery in the Media Browser.

Neither level is restricted by file type. A storyboard is usually a drawn or generated frame, because that is the cheap way to get the blocking right -- but if the best thing you have is a quick unfinished video, set that as the storyboard and it will play as video. Equally, a rough clip does not have to be a video: set a still and the shot simply holds it for its whole length, which is how you hold on a face or a sign. The two levels are about how settled the picture is, not about what kind of file it is.

The two are independent: setting a rough clip does not discard the storyboard, it just covers it, and removing the clip brings the storyboard straight back.

The gallery shows you which item is which. On a shot's gallery the item that is the rough clip carries a strong green border and the item that is the storyboard still a paler green one -- the same colour, weaker, so the two read as a pair. Hover either for a label: Rough clip, Storyboard still. An item that is both is marked as the rough clip, because that is what actually plays. The cues are the same in grid, list and graph view, and they matter most when you arrive here by double-clicking a card on the episode Storyboard: you land on a gallery of twenty outputs and can see at once which two are in use. Galleries that belong to a character, location, prop, scene, episode or project have no such choices and show no green.

The finished clip

The Finished clip criterion in The Finish makes the full-resolution version of the rough clip you chose. The criterion row shows the rough clip and the finished clip side by side, each with its size, so you can compare them before you accept the result. It only works from a chosen rough clip: until you Set as rough clip, the row says no rough clip is chosen.

On the shot's Media page the recipe panel loads the Finish recipe with your rough clip in place. If the rough clip is already at or above the episode's Finished size (see Episode Properties) the guide bar says so and offers Use the rough clip as is, which fills the row without a run.

The finished clip is added to the shot's gallery and carries a strong green border and the hover label Finished clip. It outranks the rough clip, which stays chosen beneath it. The Sequencer, the episode preview and the export play the finished clip wherever one exists. Choosing a different rough clip, or removing the choice, clears the finished clip -- it was made from the old one -- and the row re-opens.

The Finish only improves the picture. It never trims. The finished clip is exactly as long as the rough clip it was made from, and it keeps the rough clip's own sound. If a rough clip is longer than its shot, the Sequencer shows the footage to spare when you select the clip, and you can slide which part plays (see Previewing the episode); the Finish does not change any of that.

To finish a whole scene, ask the assistant: "finish the clips for scene 1". Each shot with a chosen rough clip is upscaled with the graph its look recommends; a rough clip already at the episode's Finished size is used as it is and reported as such; shots without a chosen rough clip are listed and left alone.

Block this shot

Block this shot (at the top of the Media view, and beside Character blocking on the Properties page) opens the shot's location as a Blocking Set with this shot's own layer on top, Based on the previous shot of the scene (or the scene's own blocking) unless you pick another: the stand-ins it inherits, the camera the shot uses, the camera the shot uses, anything added for this shot only, and any set object you moved (marked changed in this shot, with Reset to the set). Capture puts a Blocking View, a Blocking Depth Map and a Blocking Room (the part of the location picture the camera sees, with no stand-ins, which the recipe uses as the shot's location reference) in the shot's gallery, and writes the blocking words under them; the shot's Character blocking text is rewritten from the stand-ins and marks. The captures are tagged as guides and are never offered as reference pictures. If the layout changes after a capture, the tool says the captures are older. See Blocking.

Start frame from the camera angle and the cast's poses

Choose Start frame from camera angle and poses in Choose another recipe (listed with the rough-clip recipes) to build the shot's start picture from what the shot already knows, then animate it. The recipe opens with its first step, Layer the start frame, already filled: the shot's camera angle picture behind, and one frame of each cast member's pose clip on top, taken from the look the shot uses for that character. The second step turns that picture into a clip.

  • Which pose -- the pose picked for the character if there is one, otherwise the first pose on the look. A character whose look has no pose clip is named in a message and left out; the same sentence is kept in the step.
  • Which way they face -- a pose clip is a camera orbit that starts facing the camera. A character faces the camera unless the shot's blocking says otherwise; in a shot with dialogue and two or more characters, the outer ones turn in toward each other. The frame is worked out from the clip's measured facing (desktop only) when it has one.
  • Where they stand -- left to right in the order the blocking puts them, else the order they first speak. Their height follows the shot size (wide, medium, close). This is a first guess: open the step and drag them.
  • At a marker -- if the camera angle picture has markers ("left seat behind desk") and the character's pick names one, they stand at that marker, at that size. A name the picture does not have is ignored and said in the start frame notes.
  • Behind a mask -- if the pick names a mask shape on the picture ("desk"), the character is drawn behind it. With no "behind" the character is in front of every named mask. A mask shape with no name is a foreground item: it is drawn on top of everyone. Each picture and its masks are independent, so the same table can be behind the characters in one camera angle and in front in another.
  • Your changes are kept -- once you have moved a layer in the layering editor, the step is never refilled.

Lighting or blending the characters into the background is not part of the recipe: the layered picture is an ordinary picture, so add a step after it if you want one. For an end frame (rare), add a second layering step and switch the video step to the first-and-last-frame workflow.

Render dialogue audio

A shot whose span contains dialogue gets one extra button above its gallery: Render dialogue audio. Press it and the shot's dialogue is rendered as one audio file per character, timed exactly as the Sequencer has it -- each line where it starts, the gaps you paced, and nothing that is not that character speaking. No music, no effects, no other voice. The files land in this shot's gallery, one per speaker, each labelled with whose voice it is, so a two-hander gives you two.

Each file is exactly as long as the shot. A character who falls silent halfway through gets silence to the end, and a character who does not speak until late gets silence before their first word. That matters because the file is fed to a lip-sync video model alongside the shot's start image: the audio has to be the length of the clip it is driving. You do not have to pair them by hand -- see Animate this shot below.

When the button turns orange, what is in the gallery no longer matches the Sequencer -- the shot's length changed, a line was dragged, a take was re-recorded, the line was reworded, or someone who used to speak here does not any more. Press it again and it goes back to normal. It never renders on its own: a shot's length moves every time you drag a cut, and re-rendering on each drag would bury the gallery in near-identical clips. The colour is there so you can decide when it is worth doing.

Pressing it only ever adds audio. A track for someone who no longer speaks here is left where it is, because you may already have generated a clip from it -- the colour tells you it is stale, and clearing it out is your call.

Before the first render the button is not orange -- there is nothing to be out of date yet.

If the scene's dialogue clips are not yet linked to the script's lines, you get a note saying so in place of the button. Opening a Sequencer links them for you, so the button appears once you have done that.

Operations

Operations are picked from Choose an operation in the shot's Media Gallery. These are the ones you reach for most on a shot; the gallery offers the full set.

Animating a shot

Once the shot has a still, press Choose an operation and pick an image-to-video operation to turn it into a clip. Describe the motion in the prompt box -- Write prompt can draft it from the shot's properties -- and run it.

A shot with spoken lines is better animated with a Lipsync operation, where the character's own dialogue drives the mouth. Give it the shot's still as the character image and that character's audio file from Render dialogue audio (above) as the audio. These models lip-sync one character at a time, so in a shot where more than one character speaks, run it once per speaker's file. See AI Models for the lipsync choices and what each is best at.

Rendering the dialogue audio stays your call -- nothing renders it, or runs anything, on its own.

Reference pictures fill themselves

Pick an operation that takes reference pictures and the shot's cast is already in the slots: the characters first, in cast order and in the look the shot uses, then the location, then the props. Each picture is the asset's reference image, or else its starred image. A tag under each slot says who it is and what it supplies -- "George -- character", "Forest edge -- location", "Lantern -- prop".

If an asset has no reference image yet, its slot stays empty and a line under the slots names it, followed by Choose 's reference. That opens Browse Media Files on that asset, so you pick from its gallery -- or go anywhere else in the project: one of its looks, another asset, even a picture made for this very shot. Only images can be picked; everything else is dimmed. Tap a picture, leave Also make this 's reference ticked, and press Confirm. The picture drops into its slot with the asset's name tag, and it becomes the asset's reference image everywhere. Untick the box for a picture meant for this one shot only: the slot is filled and named, and the asset's own reference stays as it was. Setting the reference on the asset's own page works too -- come back and the empty slot fills.

If the shot has more cast than the model takes, a line says who was left out. Swap or clear any picture you like. Only empty slots are ever filled, so a picture you placed yourself is never replaced, and a slot you cleared stays clear.

Pictures you add with Browse Media Files get the same name tags. In the browser, an asset's reference image carries the reference marker, and a tap selects a picture rather than opening it full screen.

Properties

The Properties view gathers everything that defines the shot, in two parts.

Cast & Setting Assignments

What is in the shot, in four sub-sections. These save immediately as you change them, and they inform the AI Assistant and feed operation inputs automatically.

  • Characters -- which characters appear. For each character, pick the look to use and set their performance for this shot: Emotion / mood, Facial expression, Body language, and Gesture detail. Leave the list empty and the shot inherits the scene's cast, which the section names for you.
  • Location -- where the shot takes place, and which of its looks to use. Inherited from the parent scene unless the shot sets its own, so the background can change part-way through a scene without splitting it. The location's starred image becomes an input to operations that generate the shot's background. See one location or two, and changing location mid-scene. When the location has camera angles, a Location angle picker appears under it, each angle with its picture and name (an angle with no picture for this look is marked). The shot-planning AI suggests one when it plans the scene's shots, choosing only from the angles of the location (or look) the shot is at -- it shows as a proposed value, with its reason, on a shot that already has one, and your own pick is never replaced -- and you can pick another or clear it (the first angle is then used). The chosen angle's picture is linked into the shot's gallery. The AI also chooses a pose for each character in the shot, from the poses of the look that character uses here; it is kept as the pose the shot's start frame uses, and a pose you picked is never replaced. When none of the angles or poses fits, the AI asks for a new one instead: a card at the top of the shot shows the name and the reason, with Make it (a pose loads the turn-around recipe on that look with the Pose filled in; an angle opens the location look's gallery and says the name to give a picture) and Dismiss. A different pose than the one you chose waits there with Accept and Reject. Nothing is made or run until you press a button.
  • Visual style -- what this shot means visually: the recipes its style recommends for each kind of media, in order, and the art style each uses. It reads Inherit from scene (Dream scene) until you set one, so you can always see what it is getting; if the scene has none it is blank, and no look is added and you pick the model. Setting it to another style is an override -- that is how one shot inside a dream sequence goes back to the episode's ordinary look. Scene Planning proposes a style for a shot that differs from its scene, with a reason, and you can change or clear it. See Visual styles.
  • Props -- which props are present.
  • Dialogue sequence -- the dialogue lines that fall in this shot, in the order they play, each with how far into the shot it starts. Read-only here. See below.

Which dialogue a shot contains

You do not assign dialogue to a shot. A line lives on the Sequencer, as a clip you can drag; a shot is a span between two cuts. Whichever lines fall inside that span are the shot's dialogue, so moving a clip past a cut moves the line to the next shot, and moving a cut changes what the shots either side of it say. Nothing has to be kept in step by hand.

The Dialogue sequence section lists what currently falls in the shot and offers Open the sequencer, which is where you change it. The screenplay's lines are placed on the Sequencer for you when you open it -- every line that does not have a clip yet, and never moving one you have already positioned.

A line is allowed to run across a cut. It plays over both shots, and it belongs -- for audio -- to the shot its start lands in. See Audio & Voice Pipeline.

Attributes

The shot's descriptive fields, in cards down the page:

Card What it holds
Core properties The shot's name, first, because it is what you look for
Frame The single "how to shoot this" surface: location and action directions, the beat label, the shot's purpose and beat description, character blocking, and the camera fields -- shot type, subject, framing / composition, camera angle, camera motion, depth of field, background treatment. The camera fields are what a storyboard is composed from and what a replan works on first, so they are the ones worth filling in
Shot media The shot's Storyboard, Rough clip and Finished clip, after the Frame card -- you approve the framing first, then the pictures made from it. Each row has its own checkbox and the card has the usual Approve. This is where you approve them: nothing is approved by setting media in the gallery or by running a recipe
Performance Drive this shot from a recorded performance -- Not decided, Yes or No. Yes means you will record or upload your own acting, face or voice into the shot's media (mark it as a performance recording); until you do the shot is on the episode's Performances to record report and its clip waits for the recording. No means the shot is made from text and pictures alone
Look Lighting intent. These are the ones a shot usually shares with its scene
Audio properties Delivery style, SFX / ambient, music cue
Editorial properties Transition in / out, and timing notes

Text fields save when you click Save on each row.

Proposed values

When Replan shots on a scene has something to suggest for a field you have already filled in, it does not write it. The shot list marks which shots are waiting: a shot holding suggestions shows a "N proposals" badge beside its other badges, and clicking it opens that shot. Without it you would be opening shots one at a time to find the handful that changed.

The suggestion waits at the top of the shot's Properties view as a proposed value: your current text on one side, the suggestion on the other, with Accept and Reject. You can edit the suggestion before accepting it. Accepting writes it onto the shot and clears the proposal; rejecting clears the proposal and leaves your value exactly as it was. Empty fields are filled in directly, so a proposal always means "this would have replaced something of yours".

Where the AI has a craft reason for the change it shows it under the field name -- "medium" instead of "wide shot" is not a decision anyone can make, but "go closer, the beat is hers" is. A replan is asked for one on everything it proposes.

A proposed shot length appears as a row like any other, and it is the one accept with a consequence beyond the field: see how the framing gets reviewed above.

Note: This protection belongs to the scene's Replan shots control, which never overwrites a value you typed. The Screenplay page has a Regenerate shots action in its Edit menu, and it is a different thing: it re-reads the whole screenplay and rebuilds every scene's shot list from it. Shot edits can be lost that way, so leave that one until the shot structure is stable.

Sequencer

The Shot Sequencer is the episode sequencer opened on this shot -- the same editor, the same tracks, windowed to the shot's span. It is not a read-only preview: dialogue and sound are placed here just as they are on the scene and episode Sequencers, and zooming out reaches the rest of the episode. It also carries playback controls, an output resolution selector, and a video fit mode selector (Cover, Contain, Fit90, Constrain, Constrain90).

The Render Shot button (orange when the shot has changed since its last render, green when it is up to date) produces a video clip using server-side rendering. Its dropdown holds the rest:

Action What it does
Render Shot Renders this shot to a video clip
Render Scene Renders every shot in the scene and joins them
Render Episode Renders every scene in the episode
Play Rendered Shot Video Opens the last rendered clip (Close Rendered Shot Video puts it away again)
Download Rendered Shot Video Downloads the rendered clip
Delete Rendered Shot Video Removes the rendered clip

During a render: still images loop for the shot duration; video clips are trimmed from a configurable start offset; if a video is shorter than the shot duration the last frame is held; per-character audio tracks are mixed together.

When a shot render finishes, the finished clip is added to that shot's Media gallery, labelled "Sequencer render". If the shot did not already have a chosen clip, the render also becomes the shot's clip, so it shows up on the sequencer straight away. If you had already chosen a clip for the shot, that choice is kept -- the render just sits in the gallery until you pick it. Re-rendering the same shot refreshes the same gallery item rather than adding another one.