Scenario Video Generation and Editing

Overview

Scenario exposes a large video catalog through one MCP loop: text-to-video and image-to-video generators plus video-to-video editors, lipsync, upscalers, and deterministic cut/split/concat tools. Per-family contracts: scenario-kling, scenario-veo, scenario-seedance, scenario-gemini-omni, scenario-luma-video, scenario-runway, scenario-grok-imagine-video, scenario-wan, scenario-minimax-video, scenario-vidu.

Connection and the core generation loop: see the scenario skill. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.

Quick reference

Step	Tool	Notes
Find a model	recommend or search	recommend with the need in the user's own words plus a capability from its enum (img2video, video2video, audio2video for a song; upscaling footage and syncing it to a track are both video2video); search for a name
Inspect inputs	model_schema_get	Always before model_run; video schemas differ widely (duration, aspect ratio, frame anchors)
Upload source	upload_asset	A local still or clip becomes an asset_id
Generate	model_run	wait=false for video; dry_run=true to estimate cost first
Wait	jobs_wait	Re-call with the returned pending_job_ids until done; its ~180s timeout is not an error
Review	asset_display / asset_download	Display inline; asset_download returns the file URL, save it with curl -L

Worked example: animate a key art still into a short ad clip

1. recommend with the user's own words as prompt (the need is a capability, img2video); handle next_step as the scenario skill directs.
2. model_schema_get on the pick. Note the image field, duration options, and any last-frame anchor.
3. upload_asset the still; it returns asset_id="asset_abc".
4. model_run with parameters={"image": "asset_abc", "prompt": "Close-up on the mug; slow dolly-in as steam curls through window light, shallow focus on the rim. Room tone, no music, no subtitles."} and wait=false. Returns a job_id.
5. jobs_wait with job_ids=["job_xyz"], re-called with pending_job_ids while any remain.
6. asset_display the output (asset_id), then asset_download (no format).

The source image already fixes the look, so prompt only motion, camera, and timing ("orbit left", "hold on the final pose"), changing one clause per retry. A first-frame field pins how the clip opens (a last-frame field, where offered, how it lands, and only beside a first frame: alone it is a 400 A first frame must be provided if a last frame is set); a reference-image field carries identity and look without fixing any moment. So an exact opening composition goes in as a rendered still and a recurring character or product as references; whether one run takes both is per schema, and several forbid the pair. A reference-audio field is guidance the prompt has to claim: what it conditions (timing and energy on one family, a voice on another) and whether it rides alone or needs an image or video reference beside it are the family skill's contract, but on every member a track passed silently beside a character reference is easily ignored, so the prompt says what the track is for ("the reference audio is her own voice, lip-synced" for speech; the beat the cuts land on for music). Trained LoRAs do not ride along: at authoring time no video member took loras, so a custom style reaches video through a still generated with the LoRA and passed as the first frame or a reference. Iterate at the cheapest resolution tier and shortest useful duration, then upscale the keeper: a re-run at a higher tier is a new generation, since many video schemas expose no seed and none promises one reproduces a shot across tiers.

Prompting a shot

Write the prompt as a shot direction in sentences, not tags: shot size, setting, subject and one action, one named slow camera move, light source, style, and an audio line, in the order the family skill gives. Named moves and concrete light survive generation; mood words do not.

Many generators return sound with the picture (dialogue, effects, ambience, sometimes a score), switched by generateAudio or audio where the schema has one, each with its own default; with none, the search hit says whether sound comes back, and the prompt is the only lever over what does. Effects and ambience go in plain words, dialogue in quotes, plus a generic "no music", never a named genre or instrument (the score is laid once in assembly; switch the audio off when assembly supplies the whole soundtrack) and "no subtitles, no on-screen text", since a caption tool adds captions and removes none. Exclusions ride the prompt unless the schema has negativePrompt. Type the viewer must read is composited in assembly (scenario-video-assembly): generated type drifts.

Editing existing footage

All editing is model_run on a video-input model found with recommend:

• Prompt-driven edits (restyle, swap objects, characters, backgrounds, or reframe to another aspect ratio). A reframe outpaints past the frame, unlike a deterministic resize. When a prompt-driven editor misses a directed change (a recolor, a prop swap), edit the still: asset_get returns the clip's firstFrame as a free asset id; edit it with an image model and re-render from it as the first frame, the source clip beside it where the schema takes that pair (few do).
• Lipsync and dubbing: see the next section.
• Upscaling up to 4K.
• Deterministic utilities (trim, split, resize, effects, grading, frame extraction, background removal): see scenario-video-editing. Assembling a finished cut: see scenario-video-assembly.
• Extending a clip from its last frame, on a dedicated extend member or by chaining first-frame runs past any member's duration cap: asset_get the rendered clip and pass its lastFrame asset id as the next run's first frame, writing that prompt from what the frame shows (positions, facing, the camera's side, so screen direction survives the seam) rather than from the plan, since renders drift from it; join the runs with model_scenario-video-concat (a fixed id, for the same reason as the cut and split tools; hard cuts by default, per scenario-video-assembly).

Dubbing is not lipsync

Dubbing translates the speech and keeps each speaker's own voice, tone, and timing. It does not move the mouth, so a dubbed talking head still has lips forming the original language. Three steps, after any trim the limits below require:

1. Dub. Takes the clip as file and a required targetLang from the schema's allowed values. Omit sourceLang to auto-detect, since the value that means auto differs between models. When a brand or name must survive translation, pick a hit whose schema carries keyterms, as not all do; where it is array: true, pass ["Scenario"] even for one term.
2. Extract. Dubbing returns a dubbed video, not a bare track, and lipsync wants an audio asset. Pull the speech out with model_scenario-audio-extract, a fixed first-party id: Scenario's single deterministic tool for the operation, so discovery would only re-derive it.
3. Lipsync. Pass the dubbed video together with its extracted track. No schema says whether the input's own audio survives, so listen to the output before shipping. The clip and the track are separate fields, video/audio on most hits and videoUrl/audioFile on others. A duration-mismatch control is not universal: where present it is syncMode or lipsyncMode (cut_off, loop, bounce, silence, remap) with a per-model default, and there loop and bounce extend the shorter stream while cut_off ends at it; elsewhere a loop boolean loops the audio instead.

Judge a localized talking head on a stylized character before promising it on a photoreal one.

Duration limits

Where a model bounds input length it rejects rather than trims: a 30.08 second reference against a 30.0 second limit fails the whole run, with the error naming both numbers. A clean dry_run prices the payload and is not proof it clears a ceiling: read the input's duration off asset_get and compare it to the cap yourself. A ceiling can be a typed max_duration on the file field, prose in that field's description, or absent, so check both, on the audio input as readily as the video. When one applies, trim with the deterministic cut or split tools (model_scenario-video-cut, model_scenario-video-split, fixed ids for the same reason as the extractor above) before the run that enforces it, and before any step whose output must match the trimmed footage, such as a dub. Land inside the stated range, not on its edge.

Floors bind as hard as caps. The shortest clip a member renders is in its schema, as a min on a numeric duration, as the smallest of a duration enum's allowed_values, or as a fixed length when the field is absent (5 seconds on one reference-conditioned member, 1 or 2 on others at authoring time), so a 1 to 2 second VFX element is not a duration: 1 request on the first pick: recommend with the clip length in the user's words, read the floor off the schema, and when nothing goes low enough render the shortest clip and trim it with model_scenario-video-cut, or ask for several events in one clip ("three explosions, one after another") and split them with model_scenario-video-split. Extending the other way, past a max, is the last-frame chain under Editing existing footage.

Lipsync: footage or a still

Two lanes, told apart by the input in hand. Footage plus a new track is recommend with capability="video2video" and the ask in the user's words: these members redraw the mouth region and keep the rest of the frame (video and audio, or videoUrl and audioFile; some take typed text with a voice instead of audio, never both). A still plus a track is capability="img2video": the talking-avatar members animate the whole portrait and invent its motion, so they serve a portrait with no footage, or footage whose motion is disposable, and never rescue a clip whose lipsync failed, since the result is a different performance of a different shot. Naming lipsync as the capability gets re-read as one of these two, so name the input.

Preflight is free and decides the run. Sync members redraw pixels around the mouth they detect; a face that is small in frame, turned away, occluded, or blurred by motion gives them nothing to redraw, and the run then completes and bills with the mouth still under the new track, the failure users report as "no lipsync at all". asset_display the clip's firstFrame (free, from asset_get): the lips should read at display size. When they do not, crop onto the face first (model_scenario-compose-video, the compositor scenario-video-assembly teaches and a fixed first-party id like the cut tools above, or a generative reframe prompted to tighten on the speaker; an outpainting reframe widens the frame and shrinks the face) rather than retrying members. A crop lowers resolution, so read the cropped clip's width and height off asset_get against the member's floor before pricing the sync run, and upscale it or pick a member without a floor when it falls short. With several faces in frame, pick a member exposing speaker selection (activeSpeaker on one at authoring time) or crop to one speaker. Per-member limits sit in the schema, not the description: one member takes clips of 2 to 10 seconds at 720p to 1080p, others carry none; the video field carries cost_impact: true, so trim to the shipped range before the run (Duration limits above).

Plan gating is common here: recommend flags such members on their ranked entry with requires_plan_upgrade: true and names the plan in required_plan (unless its response says plan gating is _degraded), and running one anyway returns a 403 naming the model and the plan it needs, which no retry clears; pick an available member or surface the upgrade, per the scenario skill's plan row.

A completed job proves the model ran, not that the mouth moved. Before shipping, download the output and the source and compare frames at the same timestamps (sweep both into contact sheets locally, ffmpeg -vf "fps=2", and read the mouth region): unchanged mouth pixels across the sweep mean the face was not found, and the same payload reproduces it, so change the framing or the lane rather than the seed. Listen as well: whether the input's own track survives is undocumented (Dubbing above).

A whole song in one run

recommend with capability="audio2video" and the user's words finds members that turn a finished song into a music video in one call. The whole-song member live at authoring time took 10 seconds to 6 minutes of audio and returned a video as long as the track (the same ranking also lists segment-length members capped near 20 seconds), so:

• Trim first. audio carries cost_impact, as does resolution: cut to the section that ships with model_scenario-audio-cut, then dry_run the real payload.
• Style reference only under the custom style. A preset style ignores styleImage; set the custom value to use one. The character image is a separate, optional field. It takes no prompt: style, musicStyle, and the reference images carry the direction.
• Lyrics become burned-in subtitles. Leave lyrics empty when captions are added in assembly (scenario-video-assembly), or the two sets collide.

Pick this lane for one pass over the whole track; beat-cut shots under per-shot direction are scenario-seedance-music-video. Launch with wait=false and jobs_wait, since the job scales with the song.

Common mistakes

• Shipping a lipsync run on the strength of its completed status: the failure mode is a still mouth under the new track, so compare frames against the source first.
• Calling model_run without model_schema_get: a payload that worked on Kling will not fit Veo.
• Treating search hits as stable: catalogs evolve, so re-run search and prefer non-deprecated hits (a deprecated:<replacement_id> tag names the successor). Families retire whole generations on notice, and a retired member's per-member controls (a keyframe strength, an fps enum) do not carry over: re-read the successor's schema rather than porting the old payload, and when a saved id returns 404 Model not found, search the family name instead of retrying the id.
• Excluding music by name in an audio-enabled prompt: the soundtrack is moderated on its own, and a named genre or instrument trips it even inside an exclusion. Describe diegetic sound positively ("room tone, footsteps, one voice"), or turn the audio field off.