Assemble and finish a film

Start with completed clips you have watched and approved. This recipe produces one finished downloadable film. Generation of missing clips, music, narration or subtitles can add cost; joining the clips remains a separate $0 operation.

Join the clips in the intended order

Connect approved clips to Montage (Монтаж) in story order: video 1, video 2 and following inputs. Inspect the connections instead of assuming that node positions define the sequence.

When using scene expansion, expand in ascending scene order and read warnings about ordering. Montage follows its connected input order. Run it — Montage joins the clips on the server, Montage itself costs $0 — and watch the entire result, including each cut and the final frame.

Montage processing runs in the cloud while the Studio keeps its job and progress. You can reload the page or reopen the project and continue following the same job; a submitted cloud render can also be recovered after the Studio server restarts. Wait for its completed video instead of submitting another build because the connection was interrupted. Progress describes the current work, rather than a guaranteed completion time.

Land scenes on the song timecodes

When the Scriptwriter wrote the scenes against song timecodes, each scene carries its plan: where it starts on the song and how long it stays on screen. Clips are shot in whole seconds, and many models shoot at least five, so a clip usually runs longer than its plan. Montage’s Trim to plan (обрезка по плану) is on by default: for every clip that comes straight from a Video node with a planned scene, Montage cuts the extra tail so the scene ends on its timecode. A clip without a plan, and a clip passed through Clip trim, Lip-sync or a pinned file, goes in whole. Set it to Do not trim to keep every clip at full length. The field shows only when at least one connected clip carries a plan (or when it is already set to Do not trim), so a Montage of plain clips has nothing to switch.

After the run the Montage card lists what actually happened, from the measured clips: how many clips were trimmed and the film length, each trim (for example clip 1 (sc. 1): 5.17 s → 3.4 s), a clip shorter than its plan and how far the following scenes start early, a break in the timecodes (a clip without a plan, a skipped scene, a clip out of order) and the shift after it, and whether the song and the film differ in length. A clip shorter than its plan is never stretched; read those lines before exporting, and fix the Video length or model if a scene must land exactly.

Arrange, trim and mute on the timeline

Press Montage studio (Студия монтажа) under the Montage card, above Assemble, or Open the timeline (✎ Открыть шкалу) in its panel. Both appear once at least one clip is connected, and not in the tutorial sandbox. Montage Studio (Студия монтажа) shows a preview of the whole film above the lanes: clips back to back, the song (the music input) and the voice. A clip whose wire is connected but which has not been shot yet stands as a placeholder with its plan length (a nominal 5 s without a plan), marked “not shot yet”; it gets its place in the film when it is shot.

Order. Drag a clip along the lane. The studio rewires the video N inputs on the canvas, so the input order is still the film order; unshot inputs move with the rest, empty inputs keep their place. Ctrl/Cmd+Z undoes it, wires included.
Edges. Drag a clip’s left or right handle to start it later or end it earlier, in steps of one 1/30 s frame; the neighbours close up behind it. By default the end sits on the scene plan (marked “to plan”) or on the end of a clip without a plan. Your edges win over the plan, and ↺ returns the clip to the plan. Edges belong to that exact take: if the clip is re-shot, or the file becomes shorter than the start you chose, the edges are dropped and a line says so. A trimmed clip breaks the timecode chain, and the line names how far the following scenes move.
Mute a clip. 🔇 replaces that clip’s sound with silence. Music and voice stay as they were, and the Clip audio choice does not change; when every clip is muted, a line suggests Clip audio: Turn off, which plays the song at full level.
Song. Drag the song along its lane to start it later in the film, drag its left handle to begin from a later second of the song, and drag its right handle to end it earlier — the song then stops at that second of the song instead of playing on to the end of the film. The inspector takes the same three numbers: From which second of the video, From which second of the song and Until which second of the song; an end at or past the song’s own end removes the trim. The original file is never changed: the trim only says which part of it plays. The lane shows the result, including the part that runs past the end of the film. If you replace the song, the shift and the trim are dropped with a line; when the scenes follow song timecodes, the line names how far the song now sits from them.
Song pieces. One song can play as several pieces on its lane — for example the chorus twice — without a second wire or a media library. Put the playhead on the song and press ✂ Cut in the timeline header (or S): the piece under the playhead becomes two pieces of the same file, end to end. Click a piece to select it; it then moves, trims with its own left and right handles and keeps its own fades, and the inspector shows its numbers. ⧉ on the selected piece repeats it right after itself; the pieces that follow move to the right only as far as they must to make room. ✕ or Delete removes the selected piece and leaves silence in its place; the last remaining piece cannot be removed — to drop the song, disconnect the music wire. Pieces on the lane never overlap: a piece dragged or trimmed towards its neighbour stops at it. Ctrl/Cmd+Z undoes a cut, a repeat or a removal in one step. The file itself is never changed: each piece only says which part of it plays and from which second of the film. The card lists the pieces in one line, and the server builds them exactly as the lane shows.
Fades and volume. Drag the song’s top corners: the left one lengthens the fade-in, the right one the fade-out (0 to 5 seconds, in whole frames); the inspector takes the same numbers. The line on the lane is the song’s volume, the same Music volume as on the card, and Ctrl/Cmd+Z returns it with the rest. The fade-out ends where the song stops sounding: on the last frame of the film when the song is longer, or on the song’s trimmed end. The preview and the server use one calculation, so what you hear is what is built, give or take one frame at the very end.
Voice. The voice sits on its own lane; drag it to start it later in the film (the voice never starts before the film does, and a tail past the end is cut and hatched). Its left and right handles trim the recording — the start and the end of what is heard — and the inspector takes the same numbers (From which second of the recording, Until which second of the recording). When the clips’ own sound stays on and there is no voice, the clips play louder once the song stops — also when it stops at a trimmed end; the preview plays it the same way.
Zoom, snapping and keys. Zoom goes down to one frame (12 px); the ruler reads minutes:seconds:frames at 30 fps. Edges, the playhead, the song and the voice snap to scene timecodes and to each other within 8 screen pixels: switch snapping with N or the button, hold Alt during a drag to skip it. ? lists the keys: Space plays, ←/→ move by a frame, Shift+←/→ by a second, +/− zoom, S cuts the song at the playhead, Delete removes the selected song piece.
Re-shoot a clip. ↻ on a clip shoots its source again; the price is shown before the click and is the one the confirmation shows. The new take takes the clip’s place, and the clip’s edges are dropped with a line saying so.

Build ▸ in the studio does the same as Build on the card: unshot and stale clips upstream are shot at the current estimate, with the same confirmation before any charge, so a re-shot clip loses its edges (its plate warns about this beforehand). The server builds the film exactly as the timeline shows it: every piece has the frame count the studio displayed. One difference you only hear in the result: the music ducking under the voice is not previewed.

Every edit stays visible on the Montage card as a warning line (clips trimmed by hand, muted clips, the song shift); after the run the card adds what the server measured. Edits are stored in the Montage node itself and travel with the project, not with a published recipe: the recipe author is told that the edges, 🔇, the song shift, the audio trims and the song pieces stay home, while the clip order travels as wires. The studio opens read-only while viewing as another user or outside a live project. Vini sees a short summary of the edits (including the song fades and the voice shift).

Let Vini edit the timeline

Ask Vini in the chat to arrange the clips, trim an edge, mute a clip, place the song, end it at a given second of the song, set its fades and volume, or move the voice. Trimming the voice and cutting the song into pieces are done on the timeline by hand; when the song already has several pieces, Vini’s song step changes the first piece and leaves the others as they are. Vini proposes the whole change as one card whose lines read in plain words (for example “video 3 → position 1; song: fade-out 2 s”); Apply does all of it at once, and if any step is impossible nothing is changed and the card says which step failed. Vini names a clip by the name of its input (“video 3”), not by its number on the card, so a disconnected input does not shift the addresses. Edges can only be set on a clip that has already been shot.

When you open the studio after Vini’s change, a line above the lanes says what Vini changed, and one Ctrl/Cmd+Z returns the order, the edits and the volume as they were. If you edited by hand after Vini, Ctrl/Cmd+Z first undoes only your edits; if you changed things in a way that makes a one-step undo impossible, the line says so. Your own gestures are never blocked.

The old Transition (переход) field on Montage is gone: it only ever offered a plain cut.

Listen before adding more audio

Your generated clips may already contain sound. Listen first. Add music or voice only when the film needs it, then review their balance against the original soundtrack. Generating a new Music or Voiceover result can charge.

When the film goes under your own song and the clips’ sound should not be heard, set Montage’s Clip audio (звук клипов) to Turn off. Some models always generate sound — MiniMax H3 has no switch for it — so this is the one place to drop it for every model. The film then carries only music and voice; without a voice the music plays at its own level, and Music volume is hidden because there is nothing for it to sit under. With a voice, the music is still lowered under speech. Listen to the whole result: the song is not trimmed to scene boundaries; by default it plays from its beginning and stops when the film ends, and the timeline can shift it.

Voiceover creates audio; Dubbing creates an alternate-language audio track; Lip-sync aligns supplied audio with a visible speaker. Choose the operation matching the problem. Voice isolation produces speech and a best-effort residual background, not a guaranteed clean replacement soundtrack.

Make optional finishing edits

Trim, reorder and mute clips in Montage Studio. Clip trim (Обрезка клипа) and Export (Экспорт) are retired: new ones are not created, and nodes already in older projects still open and run. Review the rendered result after changing duration, sound or framing.

Connect the finished movie to Decoration (Оформление). V2 can recognise its speech and design the episode from the actual movie, or reuse a connected Design system. To review the transcript separately, run Subtitles (Субтитры), correct every recognised segment and pass its VTT to Decoration. Supplied lines are preserved. New V2 nodes export on the server by default; choose Plan + decoration studio to inspect and edit the saved composition first. Its later Render · $0 rebuild uses that composition without another authoring call.

The previous decoration mode retains its browser studio and Subtitle look (Card or TikTok); switching only that look re-renders the saved plan at $0. See subtitles and decoration for the modes, pricing and recovery boundaries. Check readability, timing, spelling and the end card in the complete rendered video.

Download and verify

Select the finished Montage and use Download result (Скачать результат) on its toolbar. Open the downloaded file and check playback, resolution, sound and duration. A successful save is not a visual or audio quality check.

Through MCP, assemble_video can join explicit clip URLs (optional plans — one {start, seconds} or null per clip — trims each planned clip to its timecode and returns trims, short, breaks and music facts), and individual tools cover audio and subtitle operations. run_node currently refuses Montage, Clip trim, Subtitles and Decoration nodes; do not attempt to run the full finishing graph through it. The retained edit_video operation is a server trim route, not the browser Clip trim editor.

An agent can prepare the next step with list_social_accounts and put_publish_draft: caption with hashtags, the account and an applicable cover frame. Open its returned link to review the filled form. Publication is confirmed by the person in the browser; the agent’s draft sends no social post. See publish a finished video, individual media operations and runs and billing.

Last checked: 22 October 2026. Review each output; this recipe does not establish a completion-time or queue guarantee.