Valmera — AI Video Editor

Adds short visual punctuation at explicit program positions or measured speech, cut or supplied-track landmarks. Supports full-frame or regional treatments and configurable density. It adds neither sound nor zoom; flash-safety limits apply.

Adds an eased change of the visible picture's aspect ratio at a program-time position. The file retains one output resolution; the visible frame changes without moving audio or program timing.

Adds a visual breakdown of the actual edit timeline with a live playhead, using the edit's real cut, graphics, camera, world and sound lanes.

Creates burned-in captions from the word-timed transcript or explicitly supplied text items with source timestamps, or disables captions. Styling supports named presets and detailed presentation parameters.

Creates a mosaic, scrolling thumbnail wall or flatlay from specified project footage and images. Supports up to 64 tiles, including playing video and timed transitions toward a hero tile.

Inserts a procedural solid-color or gradient screen at a program-time position. The insertion shifts later material by its duration and may include a simple camera drift; it does not use an image-generation model.

Inserts a locally synthesized digital-glitch, VHS or static screen at a program-time position. Later material shifts by the inserted duration. No generative image or video model is involved.

Applies a validated single-stream FFmpeg video-filter chain to the program picture. The chain must preserve dimensions and frame rate; file access, labeled graphs and multi-chain separators are prohibited. Validation includes a dry run.

Adds a depth-based visual device such as parallax, camera drift, depth of field, rack focus, fog, light or point-cloud treatment. Source images or frames and program windows determine the affected material.

Places a screen recording or source frame inside a drawn laptop, phone, tablet or monitor stage. An explicit screen region or recorded interaction element can be lifted out as a breakout.

Adds a procedural light, particle, drawing, shape, pictogram or sprite element with configurable timing and placement. Elements can be attached to measured world anchors.

PROJECT-SCOPED: this call acts only on the explicit project_id and returns the project identity with its result. A filmstrip over a range of the program: the upload's pictures run past horizontally, two or more in view, each its own source frame, whole, in its original colour and aspect, on a soft blurred backdrop (or black). One call solves every picture in the range, baked into one file or not; length and sound never move. start/end: program seconds. direction: left|right. height: 0.3-1 of the frame. backdrop: blur|black. remove_filmstrip takes it off.

Adds a windowed picture finish such as display rephotography, halftone, xerox, film, thermal, line-art or other procedural treatment. Options control matte weighting and whether the finish also affects type; flash checks apply to luminance changes.

Adds a held program frame with configurable duration, audio treatment, optional text, blur, darkening and still-image motion.

Adds configurable frame furniture such as corner labels, a recording indicator, timecode, brackets, crosshairs or a grid over a program window.

Creates transcript-driven typography across a program window, with phrases timed to speech and positioned around the speaker. The default scope is the whole program.

Copies and places an explicitly user-selected library track into the project with its license record. A valid selection token and the user's choice identify the track. Placement uses program time and supports looping, gain and voice ducking.

Places a browser-rendered motion graphic on the program timeline using a template or supplied composition. Graphics are silent by default; optional owned sound cues use measured timing and level checks. Caption clearance and subject-depth treatment are configurable.

Adds an existing project audio asset as a music track over a program window. Defaults adapt level and ducking to whether speech is present; explicit gain and ducking settings override them. The storage key must identify available media.

Places an image or clip over the program picture during a program-time window. Supports picture-in-picture, logos, full-frame cover and picture-region fitting while retaining the underlying program audio and timing.

Applies a visual effect inside a defined region during a program-time window. Effects include color transformations, blur, pixelation and glass-like distortions; picture outside the region remains unchanged.

Creates a continuous move into a screen visible in footage, with the supplied clip corner-pinned onto the screen and then handed off to full-frame playback. The effect combines the tracked screen plane, camera push and inserted clip.

Places a one-shot sound effect at a program moment with configurable level and timing. Library recordings have measured hit offsets and placement checks against speech and visual events.

Downloads a saved media-search result belonging to the current project and registers it as an asset. The result includes frames from the downloaded rendition. Downloaded video is silent and is not automatically placed on the timeline.

Adds a frame-exact sequence of project images, source frames, program picture or colors over a fixed program window. Holding patterns and negation are configurable. Audio and program timing remain unchanged; high-frequency flashing requires the explicit flash-safety override.

Adds a windowed finishing effect to the program picture, such as grain, vignette, glow, halation, chromatic fringing, diffusion, tape treatment, shake, sharpening or denoising.

Adds an effect constrained by a measured subject matte, such as fill, poster treatment, glow, outline, negation, isolation or echo. Timing can use explicit windows or measured events. Without a measured matte, previews use a stand-in; the operation adds no sound or zoom.

Adds a designed text element over a program-time window using a title, subtitle, lower-third, callout, number, quote or chapter template. Position, size, palette and typography are configurable independently of spoken captions.

Composites text behind a moving subject using a temporally maintained person matte, so the subject occludes the letters.

Adds picture-time effects such as reverse, echo, frame stepping, stutter, interpolation or remapping. Most modes preserve audio and program length; ramp mode changes picture and sound speed and re-anchors later material. Speech-sensitive modes require an explicit override over spoken words.

Inserts a standalone full-frame text card at a program-time position, then returns to the footage. Later material shifts by the card's duration, and spoken-word captions are absent from the inserted card.

Adds phrase-scale typography solved for the current frame, with words aligned to spoken onsets and layout around the subject. It replaces ordinary captions for those words and supports behind-subject depth.

Adds large glyphs filled by footage, a neighboring shot or color, with configurable field, fill mode and reveal. The composition is stored as one motion item.

Adds interface-based explainer scenes with timed layouts and identified objects that can morph across scenes. Metrics bind to supplied data or spoken values, and media slots reference real project recordings. Supports solid, dimmed or transparent backdrops.

Adds a renderer-native rectangle, ellipse, line, arrow, ring or progress graphic over a program-time window. Geometry uses frame fractions; palette, stroke, rounding and element-local motion are configurable.

Places an existing audio asset over the program from a specified output-time position, with optional source offset, duration and ducking of other audio. Main-source dialogue can be referenced by its source offset.

Adds an eased camera move over a program-time interval with configurable framing, strength and motion style.

Adds a continuous traveling camera move with at least two program-time keyframes. Each keyframe supplies a target rectangle or center and zoom strength, enabling holds and travel between subjects.

Aligns supplied lyric text to detected singing timing on the selected track.

Returns measured structure of a user-supplied track, including tempo, beats, bars, sections, drops, gaps and instrument-event timing mapped to the program. It requires an available track.

Analyzes the sequence of visual-device classes in an uploaded reference and maps those classes to moments in the current edit. It does not copy the reference's wording, logos or footage into the program.

Adds or updates camera movement for a still-image overlay or insert, using a face, image geometry or explicit motion treatment to determine the movement.

Atomically saves a group of edit operations against the current base version, with a stable retry operation ID. Operations set complete layers, upsert or remove identified items, or reorder a canvas sequence. Times refer to the resulting program timeline. Invalid fields or media reject the batch; saving does not render or certify the result.

Applies an explicitly selected format pack's craft brief and chosen proposals in one batch. Depending on the pack and arguments, it may return proposals for selection instead. Existing framing is preserved unless aspect changes are explicitly enabled.

Applies a coordinated visual treatment across captions, grading, texture and transitions, with optional explicitly selected music.

Applies one identified visual recipe to the current cut or a specified spoken moment, returning the written item IDs. Recipe frequency and flash-safety constraints apply.

Saves caller-authored edit batches for up to 30 child shorts belonging to an explicit parent project. Each child uses its current base version and a unique retry operation ID. Each child batch is atomic, partial failures are reported, and no model execution, preview render or final export is started.

Presents a specific question and options to the user and waits for a reply. The response can include the user's selection and a selection token for an account-bound choice.

Attaches an existing motion graphic to a measured face, hand, joint, point, object or plane anchor. Configurable pin, following behavior, occlusion and lost-track handling determine placement; clear detaches the graphic.

Audits the current edit's audio roles, files, windows, offsets, gain, ducking and mastering. Includes available provenance and rights signals, identifies duplicate playback or likely misclassified tracks, and reports sound-effect placement checks.

Compiles and audits the current caption artifact used by rendering. Reports coverage, lateness, overlaps, density, line wrapping and other measured caption issues, including speech represented by graphics.

Audits the current visual-device ledger against intensity, flash-safety, expected device levels and spacing constraints. Returns findings without rendering.

Measures subject location and retained visual detail across the kept footage, then sets the output frame. Automatic mode chooses a subject-focused crop or a fitted treatment according to the measured composition.

Moves internal edit boundaries toward nearby beats within a tolerance while avoiding word interiors. Uses the program's music by default, with options for source audio or an explicitly supplied tempo. The first start and final end remain fixed.

Records a provenance binding between an existing renderer-visible movement and a motion-motif ID from the active blueprint. It does not create an effect. Static targets and the reserved stillness motif are rejected; a clear value removes a binding.

Adds a visible blur, mosaic or black-bar censor over a fixed source-frame rectangle and time window. Coordinates are source-frame fractions. The result covers the selected region rather than reconstructing its background.

Analyzes speech audio across a program window of up to 60 seconds, including cuts between source spans. Reports audible words, untranscribed sounds, clipping, clicks and sound-safe boundary evidence.

Returns labeled visual comparison pages for specified uploaded clips and images, including representative frames, clip timestamps and sentence or shot evidence IDs. It does not rank candidates by upload order or place them in the edit.

Renders two or three independent video columns side by side on a black 16:9 canvas. Each column may contain a sequence of clips. The result is a project video asset, not an automatically inserted program segment.

Reports same-shot jump cuts and their existing visual coverage without modifying the edit in report mode. The scale-step mode can apply hard, nonanimated framing steps to selected jumps.

Creates an empty editing project or podcast-shorts intake project and selects it for navigation. Returns the new project ID. Creating a shorts project does not automatically select story arcs or start editing.

Evaluates the newest render or a specified asset against visual and editorial criteria, including rhythm, typography, framing, flash safety and continuity. Returns a score and ranked findings with suggested edits; it does not write changes.

Removes one interval on the assembled program clock, including intervals that cross inserted clips. Affected source ranges and inserts are split or removed as appropriate, with dependent timing re-anchored in one versioned write.

Removes one source-time interval from the main video's current retained ranges and saves a new edit version. The remaining edit is preserved. Sound-safe boundary adjustments and optional word snapping are reported in the result.

Removes detected pauses above a duration threshold while retaining breathing-room padding and aligning speech boundaries. When many detected gaps contain audible material and the pass would be substantially destructive, it cuts only the quiet gaps and keeps the sounded ones, listed, unless the explicit nonquiet override applies.

Describes adaptations for alternative aspect ratios or, with an explicit apply value, rewrites the edit for one selected aspect. Layouts, overlays and text are re-solved while preserving source-space crop aim.

Returns a temporary viewing or download URL and durable receipt for an existing project render. A specific asset ID can recover a historical render. Preview files are low-resolution review drafts; final files are completed full-quality exports. This operation does not create an export.

Detects a screen recording's pointer, repaints its original position and redraws it with configurable scale and path smoothing. Optional click ripples use explicit source-time click events; clicks are not inferred from pointer appearance alone.

Applies picture sharpening and denoising over a program-time window or the whole video. It does not add source resolution or recover detail absent from the recording.

Detects matching burned-in text regions and repaints those pixels to reconstruct the surrounding picture. The operation preserves cuts, timeline positions, transcript and separate caption settings.

Repaints one or more explicitly defined source-frame regions during specified windows, reconstructing surrounding picture content. Multiple regions can be processed in one request; processing cost depends on the affected windows.

Returns current tool domains and schemas, including capabilities absent from a client's cached catalog. describe=[names] retrieves named schemas. call={name, arguments} dispatches an explicitly named operation through the same authorization and execution path as a direct request. Listing schemas does not edit media; dispatch inherits the selected operation's effects.

Creates or recovers a full-resolution MP4 export of an explicitly identified, reviewed edit version. Ownership, upload, timeline, quality and branding checks apply. Repeated requests reuse an existing export or running job; a queued result identifies pending work rather than a completed file.

Extracts the audio from an uploaded video into a separate project audio asset. The result includes its storage key; extraction alone does not place the audio or the video's picture in the edit.

Downloads media from a supplied direct-file or supported media-page URL into the project. Media kind is detected automatically or explicitly selected, including audio-only extraction. Downloading an asset does not place it on the timeline.

Measures picture motion and sharp sound transients over every frame of the main video and returns active and idle source-time stretches with a padded keep list. Measurements are relative to the footage and do not identify what the activity is.

Measures burned-in text regions in source frames and returns their rectangles and visibility windows. Scope can distinguish captions, persistent marks, other text or all detected regions.

Finds candidate web links to footage of a named person, company, product, event or other concrete topic. Search results do not download media, establish reuse rights or insert shots into the project.

Finds shots in the main recording and uploaded clips by visual similarity, framing, lighting, shapes, face scale or storyboard text query.

Returns detected silence intervals of at least the requested duration within a source-time window, including midpoints and neighboring words.

Searches for candidate web links to a specifically named song or recording. Results are discovery candidates and do not establish permission to reuse the music or place it in the edit.

Searches the indexed main-video filmstrip for a concrete visible event and returns coarse source timestamps with visual evidence. Results are candidate locations rather than frame-exact edit boundaries.

Returns cached measured audio features, including tempo and confidence, beats, energy changes and stressed words. Timestamps use the selected source asset's clock.

Returns a compact source timeline joining sentences or shots with overlapping scene changes, energy, beats, vocal stress, pauses and measured visual-layout evidence. Supports indexed uploaded clips and focused views such as story, faces, interface regions or energy peaks.

Returns the current edit decision list and version. Large timelines support compact indexes, selected sections and pagination, preserving valid JSON rather than returning a truncated document.

Returns the transcript retained by the current edit, with program timestamps, corresponding source spans and repeated-phrase detection.

Returns indexed shot boundaries and scene changes within a source-time range.

Returns measured syllable onsets for words spoken within a program-time window.

Returns a sentence-level transcript for a source-time range of the main recording or a specified uploaded media asset. Timestamps belong to that source asset rather than the edited program.

Returns video metadata, indexing information and an edit-decision summary for the explicit project.

Returns word-level timestamps for a source-time range of the main recording or a specified uploaded clip. Responses contain at most 400 words with pagination information. Asset-specific timestamps belong to that clip.

Validates an uploaded TTF or OTF brand-font asset for embedding and prepares a sanitized subset. The license statement and explicit rights attestation accompany the import.

Validates and applies a CUBE lookup table from a project asset or URL over a program-time window, with configurable strength and a maximum supported grid size.

Returns the progress and result of the explicit project's media analysis. Transcript, shot and silence data become available when indexing completes.

Splices an uploaded video, original-source excerpt or image at a program-time position. Positions inside a take split it at an appropriate boundary, and later timeline elements are re-anchored. Asset windows, duration and framing control the inserted content.

Keeps a selected hue, the subject or both in color while rendering the remaining picture monochrome. Supports optional skin preservation and a program-time window.

Records an evidence-backed justification for an eligible verification false positive on an immutable edit version. Missing previews, caption proof, corrupt glyphs and invalid music timing are not eligible for justification.

Replaces the retained source ranges in the given playback order and saves a new edit version. Repeated ranges repeat footage. Cut boundaries may be adjusted for word, sound and shot safety, with adjustments reported in the result; the word-snapping option controls additional alignment.

Lists project files, including timeline media, unused uploads and historical renders, filtered by kind. Results include exact storage keys and availability information.

Lists available motion-graphics templates with their purposes, parameters and optional sound cues.

Lists the built-in music library by mood with exact track identifiers, license information and indicators of prior use by the account.

Lists the authenticated account's video projects, newest first, with video availability, project kind, shorts progress and parent relationships. Navigation-pointer information does not determine the target of subsequent project-scoped requests.

Lists available recorded sound effects with identifiers, descriptive usage information and measured hit levels.

Returns timestamp-labeled sampled frames from the main source or assembled program, with a coordinate grid. Program geometry includes crops and inserts; rendered mode reads the current preview's actual pixels, including burned text and grading. Optional overlays show measured mattes, free space or timing landmarks.

Returns labeled frames from a specified uploaded image, clip or historical render at asset-local timestamps. Video responses also include a cached sparse temporal profile of motion, static intervals and abrupt changes; this is sampled analysis rather than continuous playback.

Returns source or program frames with measured face, hand, pose, point, object and plane anchors drawn for inspection.

Make Valmera's automatic first cut of an explicit project's analyzed main video: a finished vertical film (the strongest stretch, tightened, with a hook title, captions and the picture reframed for phones), made by the same chain the Studio runs after a first upload. Use it when the user wants a finished video and has not asked for anything more specific, or as the first draft you then refine with the editing tools. Needs index_status 'done'. Returns a job: call wait_for_job, then download_url(kind='preview') to watch it and export_final(edl_version) for the MP4. One first cut per project; calling it again returns the existing one.

Creates locked child projects from explicit caller-authored source ranges and story context in a long recording. Over MCP, story arcs are supplied by the caller; the pipeline seeds those windows without selecting creative styling or rendering finished edits.

Creates up to six registered, procedurally transformed still-image variants from one main-video frame at a program timestamp. Saves the images as project assets without placing them in the edit.

Builds a shot-matching color lookup table for a program window against a specified hero frame. Reports measured color differences before and after matching.

Reorders an existing inserted scene in place relative to another insert or the beginning of its boundary. Scene-anchored elements move with the scene.

Updates the position, size or time window of an existing overlay. Omitted fields remain unchanged.

Retimes an existing sound effect while preserving its recording and level. The requested time is the measured hit position for library sounds and the file-start position for other audio.

Selects a project for navigation and returns its video, transcript, shots, edit decision list and available assets. Project-targeted requests retain their own explicit project IDs.

Opens a generated short by child project ID or by its one-based card number and explicit parent project ID. Returns the child identity and state for direct editing; it does not start or delegate to another editing agent.

Returns labeled pages from the main footage's persisted visual storyboard. With no explicit page selection, it returns pages not yet delivered in the current logical turn.

Returns a suitable still-frame candidate from a program window using sharpness and clearance for the specified text-lockup box.

Places an existing graphic or new text on a tracked plane or an explicitly supplied four-corner surface. Supports occluders, shadow interaction and blending with the surface.

Places a graphic or new text relative to a described world location, with explicit geometry where needed. It writes the necessary placement and anchor definitions in one batch.

Returns a compact preproduction shot list for a brief, including framing, camera, lighting, action and the visual device each shot supports. Also describes possible approaches using newly filmed, existing or stand-in footage.

Proposes a loop ending whose picture relates to the opening frame. With write enabled, it changes the program ending before the end card.

Builds or revises a montage over a program window using the supplied track's beats, or speech stresses when no track is present. A dry-run mode returns proposed slots without writing the montage.

Returns the current video, transcript, shots, edit decision list and assets for the explicitly identified project.

Renders small evidence windows with and without selected devices and returns frame strips, changed-pixel overlays and timing or layout measurements. Metrics include landing error, legibility, matte jitter, flash risk, junction integrity and face collisions.

Adds timed emphasis camera moves at measured stressed words that remain in the current edit, with detected face targets when available. Count, strength and scope are configurable.

Retrieves topic-specific editing reference documentation, optionally limited to a section such as evidence, patterns, verification or repair. It does not modify project media or the edit decision list.

Changes a selected tracked object's specified source color to a requested color while preserving shading, detail, other hues and the background. The operation is limited to the object's tracked window of at most six seconds.

Captures a supplied live webpage as a project video asset, with optional smooth scrolling and start/end holds. Duration and orientation are configurable. The operation records page footage rather than an interactive multi-step demonstration.

Captures a supplied website while a visible cursor performs explicitly authored interaction steps, such as clicks, typing and scrolling. The resulting recording includes interaction timing for subsequent editing.

Removes an identified mid-program aspect-ratio change.

Removes an identified censor region, or all censor regions when no ID is supplied.

Removes identified B-roll cutaways, restoring the underlying picture.

Removes pointer enhancement and restores the original cursor appearance from the untouched source.

Removes an identified custom video-filter chain.

Removes identified depth or time effects.

Removes an identified editorial composition and its owned caption suppression.

Restores original pixels for one identified erased region or all erased regions. Window patches can be removed directly; legacy whole-file erases rebuild from the original.

Removes timestamped filler tokens from speech using detected word boundaries. The default targets non-word hesitations; an explicit word list can select other tokens.

PROJECT-SCOPED: this call acts only on the explicit project_id and returns the project identity with its result. Remove a filmstrip by id (get_edl fx), or every one ('all', the default).

Removes identified grade windows and finishing effects.

Removes one inserted media item and restores the surrounding timeline timing.

Removes an identified motion graphic and its owned sound cues.

Removes an identified background-music item from the versioned edit.

Removes an identified visual overlay.

Removes an identified footage picture card.

Removes identified accent, strobe, junction or matte-flash items, or all punctuation items.

Removes identified region effects.

Removes the inset screen frame and returns the picture to full bleed.

Removes an identified screen takeover, including its corner pin, camera push and handed-off clip. The optional keep-clip setting retains the clip as a plain insert.

Removes an identified sound-effect item from the versioned edit.

Removes an identified speed span, restores normal playback speed and re-anchors program-time items.

Restores the original mixed soundtrack by removing stem-level mixing.

Removes an identified stylization effect.

Removes identified subject effects and background treatments.

Removes an identified text element, including text composited behind the subject.

Removes identified typeset items and returns their words to the ordinary caption track.

Removes an identified native typography scene and its owned caption suppression.

Removes identified interface scenes and the footage in their owned media slots.

Removes an identified renderer-native vector graphic.

Removes an identified voiceover item.

Removes identified world definitions and detaches graphics that refer to them, or detaches selected motion graphics directly.

Removes an identified ordinary zoom effect.

Removes an identified keyframed traveling zoom.

Renders evidence for the current immutable edit version. A changed-section proof is distinct from a complete preview. Complete and approval modes produce a stored full-program review video; pending work returns job information rather than a completed render.

Renders three typography treatments of an existing lockup as still images on the actual frame. Returns comparison images without changing the edit.

Researches a set of B-roll moments with concrete visual queries, editorial purposes, intended program positions and durations. Returns candidates grouped by story purpose with comparison evidence.

Creates a new edit version based on the full original source, removing the current cuts, captions, tracks and effects. Earlier edit versions remain recoverable.

Restores a previously removed source-time interval to the retained ranges and saves a new edit version. Boundary adjustments for sound safety and optional word alignment are reported.

Applies supplied translations to captions and on-screen graphics while retaining their timing. Accepts phrase translations, subtitle text or text mappings, and can list source phrases. It does not perform machine translation.

Searches configured stock-media and open-photo providers and returns a visual candidate grid with provider and license information. Results may cover generic visuals or topical photos; no asset is downloaded or placed by the search itself.

Finds sentence locations in the transcript using substring and fuzzy matching.

Segments and tracks an object within a window of at most six seconds, seeded by a box or points at a program timestamp. Stores an object definition for placement and occlusion.

Separates the original soundtrack into speech or vocals and other audio, then applies independent gains to those stems. Separation is cached per video. Separately added music tracks are unaffected.

Changes the gain in decibels of an existing music, sound-effect or voiceover item without replacing the item.

Replaces the area behind the subject over a program window with a solid, gradient, blur, source image or frame, or strobe sequence. The subject can optionally be held at a specified time. Flash-safety rules apply; sound and zoom remain unchanged.

Sets the project's brand palette from an uploaded logo or explicit accent color and can place a supplied logo sting before the platform end card.

Replaces, appends, lists or clears exact displayed-caption substitutions, optionally scoped to output-time windows. Supports changed word counts, punctuation, capitalization and deletion by empty replacement. Audio and cuts remain unchanged; the response includes a compiled caption preview.

Replaces the complete set of program-time intervals where burned captions are hidden. An empty interval list restores captions everywhere. Audio and cuts remain unchanged.

Changes the visual styling of existing captions while preserving their wording and timing. Only supplied style fields are updated.

Sets or clears a whole-video color-grade preset. The preset affects footage rather than caption styling.

Updates how an existing motion graphic blends with the picture, including blend modes, footage-filled glyphs and ghost treatment for occluded text. Supplied settings merge with the existing composite.

Creates or replaces an identified editorial composition using editable native text and vector layers. Supported forms include statements, comparisons, metrics, quotes, chapters, labels and headlines, with explicit text and layout geometry.

Sets opening and closing fades for picture and audio. Durations are in seconds; zero clears the corresponding fade.

Sets output aspect ratio and crop, padding or blurred-padding treatment. Crop focus uses source-frame fractions, with the top-left at zero. A centered focus is the default; output canvas dimensions do not imply additional detail in the source footage.

Adjusts continuous footage color controls after the preset grade. Exposure, temperature and other controls have defined ranges; contrast and saturation accept small signed deltas or explicit multipliers, with neutral values documented by the schema.

Sets or replaces a color treatment over a program-time window, with frame-based entrance and exit ramps and optional regional or skin/brand protection. Picture outside the window remains unchanged.

Changes an existing inserted clip's source offset, duration or playback rate in place. Rate-only changes preserve the selected source window while changing its program duration, with dependent timeline positions adjusted.

Sets the project's calm, designed or charged treatment level. Lowering the level removes excess low-ranked devices and flashes in one reported write; raising it does not automatically add effects.

Sets or removes a duration-preserving cut device at specified program junctions. Each cut has one device, replacing the previous selection; sound and program timing remain unchanged.

Enables or configures social loudness mastering for previews and final exports. Dialogue leveling precedes the combined mix, which uses a final loudness target and true-peak protection; format-dependent defaults may already enable mastering.

Updates an existing motion graphic's window, parameters, template, layer, HTML or caption suppression. Owned sound cues and measured clearance are updated with the graphic; parameters merge with existing values.

Updates the timing, looping, fades or ducking of an existing music item. Omitted settings remain unchanged. Ducking supports a smooth voice-following envelope or the legacy stepped mode.

Sets scalar or element-local keyframe curves for an existing overlay's position, scale, rotation and opacity. Keyframe time zero is the overlay's start. Full-frame covers and tracked screen takeovers have separate geometry constraints.

PROJECT-SCOPED: this call acts only on the explicit project_id and returns the project identity with its result. Blurred fill: every picture with bars, whether baked into the upload (a tall photo in a wide slideshow) or made by the frame (a wide photo in a 9:16 frame), sits fitted and whole over a blurred, enlarged copy of itself. One call: the whole film by default, or start/end (program seconds), or at= one picture. blur: 0.01-0.2 of the frame. dim: 0-0.9. mode='none' removes it, whole or in a range. Pictures with no bars stay as they are.

Composes footage as a rounded card during a program-time window while keeping captions and branding independent. An existing ID updates that card; omitted settings retain their current values.

Places the finished picture inside a rounded, inset window with configurable shadow and solid or gradient background. Geometry is expressed as fractions of the frame.

Changes playback speed over a main-video source-time interval, preserving audio pitch and re-anchoring program-time items. Overlapping speed spans are replaced and reported. Slow motion may duplicate frames when the source lacks sufficient frame rate.

Adds an eased speed ramp within one shot or across an identified video insert. Source frame rate constrains duplicate-free slow motion; explicit options govern duplicate frames and speech-bearing windows.

Composes two to four independent footage panels with optional labels and a panel expansion. Returns panel rectangles for subsequent layout operations.

Sets or replaces element-local position, scale, rotation and opacity curves on an existing ordinary text item. Empty or omitted motion clears explicit curves. Subject-matted text does not support this movement.

Sets duration-preserving effects at scene changes and inserted-media junctions. Footage does not overlap and the program's timing remains unchanged.

Updates an existing typeset item's voice, layout, motion verb, depth, palette, wording or time window and re-solves its layout for the frame.

Creates or replaces a speech-cued mixed-font phrase as editable native text in one revision. Runs use absolute program-time cues and reserve their final positions; build and still modes control their appearance over the shared window.

Updates an existing vector graphic's geometry, palette, kind, program window, progress value or motion. An empty motion object clears its keyframes.

Applies volume automation to the original recording's audio over a source-time interval. It does not address separately added music or voiceover tracks.

Adjusts a supplied track's offset, window, fades, filtering and selected stutters to align a drop or other musical event with a program moment. Optional holds add a preceding silent interval.

Returns a project's podcast-shorts progress, planner job, generated child project IDs, edit versions and final-render states. For a child short, the response includes its parent run and preserves the requested child identity.

Places a screen-recording clip into the program and applies camera movement around positioned clicks. Recorded event data or caller-supplied click times determine the interaction moments; click sounds are optional.

Searches configured media providers for explicitly described story moments, evaluates candidate files and places selected B-roll. Candidate checks include applicable reuse-license signals, crop survival, resolution, motion and visible marks.

Replaces the current program with a new sequence beginning with one uploaded clip or image in a single valid edit-version write. Video windows support source offsets and duration; previous versions remain recoverable.

Returns ranked visual-device candidates for the current program, bound to measured story moments and checked against intensity and flash-safety constraints. Results include explicit candidate edit operations; the request does not apply them.

Returns candidate emphasis words from the actual transcript using measured vocal stress, digits and word distinctiveness. Returned words preserve their transcript spelling.

Proposes a sequence linked by face alignment, shape or similar framing. With write enabled, it inserts the selected cropped, muted shots at the requested program position.

Returns timed graphic candidates for the current program, using transcript-backed numbers, enumerations, contrasts and emphasis. Camera and sound-effect suggestions are optional and disabled by default.

Returns ranked, story-timed visual recipes with rationale, reference information and explicit edit operations. Ranking considers story fit and reuse across related shorts. Suggestions do not write the edit.

Returns story-bound visual proposals for hooks, turns, payoffs, callbacks and loops, with timing, rationale and novelty information relative to the project.

Replaces an existing music item's source track while preserving its timeline position, level and fit settings. The replacement identifies an available project audio asset.

Tracks a specified point through its shot and records a world-track definition. The seed uses program time and normalized picture coordinates.

Completes a previously prepared storage upload and registers its media in the explicit project. A completed main-video upload starts media analysis, including transcript, shot and silence indexing.

Prepares a local-file upload for an explicit project and returns presigned storage URLs and the upload-completion contract. Supported kinds include main video, supplemental clip, music and image; clips may have a reference-only shorts role.

Waits for a bounded period on a background job and returns its result or current running state. Repeated requests observe the same job without restarting it.

Returns the project's assembled timeline, original source or specified media asset as a playable MP4, optionally restricted to a time window. Responses can include sampled frames and window audio. The timeline mode renders the current edit if no suitable render exists.