Return the VIVID account linked to the API key: name, email, plan and remaining credits. Call this first to check the connection.

Get the link where the user buys a credit pack (500 / 1,500 / 5,000 credits) and can turn on auto top-up. Use it when a call fails for lack of credits or before a big batch that the balance (vivid_whoami) will not cover. This tool never charges anything: the user picks the pack and pays in their browser. Show them the link, wait for them to confirm, then continue.

Monthly usage summary for the account (analyses, images, videos vs plan limits) and current credits.

List the AI models available to this account with their cost in credits and capabilities. Use the returned slug as the model argument of vivid_generate_image / vivid_generate_video. Video models expose durations, resolutions, aspect ratios and what they support (startFrame, endFrame, reference, audio…).

Run one chat completion on VIVID's LLM routing (Grok, Claude, OpenAI GPT via Kie.ai; Kimi via Wavespeed) paid in credits: provider price + 50%, billed per token (list with vivid_list_models type="llm", prices in credits per 1M tokens). Non-streaming. Pass the full conversation each call (system/user/assistant); image parts as {type:"image_url", image_url:{url}} on vision models. Returns the reply text, token usage and the credits charged.

Generate one or more images on VIVID from a text prompt. Feature saved products and testimonials by NAME with products / testimonials (same as the app's #Product / @Testimonial mentions — list them with vivid_list_references), or pass ad-hoc reference image URLs (public URLs, e.g. from vivid_upload_file). Credits are charged per image according to the model. By default waits for completion and returns the asset ids and download URLs.

Start a video generation on VIVID. Pick a model with vivid_list_models (type=video) and respect its durations / resolutions / aspect ratios. Reference images, start/end frames and audio must be public URLs (use vivid_upload_file for local files). Credits are pre-charged (credits/second × duration). Videos take 1-5 minutes: by default this returns the jobId immediately — poll with vivid_job_status. Note: a model's start/end frame and reference images can be mutually exclusive (e.g. Seedance 2.5): the backend resolves it and returns a warning. Reference videos (referenceVideoUrls) work on models whose vivid_list_models entry has referenceVideos: Seedance 2.0/2.5 bill the uploaded seconds too, Gemini Omni takes one source clip.

Check a generation job (image or video). For video jobs this also triggers the provider status refresh, so call it to advance a processing job. Returns status, asset id and download URL when done.

List recent generation jobs, newest first. Filter by type (image_gen, video_ugc, product_analysis…) or status (pending, processing, completed, failed).

Browse the account gallery: generated images/videos, uploaded products, testimonials (type: image | video | audio; category: generated | ecommerce | model | context | video …).

Metadata for one asset (type, filename, favorite/public flags, share link if public, originating job).

Download an asset (image / video) to a local directory and return the saved path.

Make an asset public and get a shareable URL (no login required), or revoke sharing — revoking breaks the link for everyone who has it. Also toggles the favorite flag. Only change sharing when the user asked to. The reply carries disclosure.notice: relay it to the user when they share or publish AI-generated content.

Upload a local file (absolute path) or a remote URL — image, video or audio — to VIVID's temporary storage and get a public URL to use as startFrameUrl, referenceImageUrls, objectImageUrls or audioFileUrls. Temporary files expire after 7 days; generated results are stored permanently in the gallery.

The account's saved PRODUCTS (from product analysis: name, category, colors, material…) and TESTIMONIALS / characters (name, gender, age, look…) with their asset ids. Use the names in vivid_generate_image products / testimonials to feature them with locked look and identity — the same as typing #Product / @Testimonial in the app. Shared public talents are included (isPublic).

Analyze a product photo: vision analysis (name, category, colors, material, finish, style, keywords) + a clean e-commerce render, saved to the account as a PRODUCT you can then feature by name in vivid_generate_image products (see vivid_list_references). Costs 10 credits (refunded if the analysis fails) and counts toward the plan's monthly analysis quota. Takes 30–90 s; waits by default. For jewelry the geometry lock may ask the app to confirm the category; the name comes from the analysis (rename in the app if needed).

Create a reusable TESTIMONIAL on the account — a consistent person you can cast by name in vivid_generate_image testimonials (identity locked). Two ways: photos = three photos of a REAL person (left profile, front, right profile) → composite identity. Requires consent: true: set it ONLY after the user has confirmed that the person pictured is them or has consented to being turned into a virtual presenter; the consent is recorded with time and IP, and photos with nudity or sexual content are refused; or attributes = design the person from scratch (gender and age required; optional ethnicity, bodyType, faceShape, nose, eyeShape, eyeColor, hairColor, hairStyle, hairTexture, skinType, skinColor, expression, distinguishingMarks… values in English as in the app, e.g. gender "Female", age "25-35", ethnicity "Mediterranean"). Costs 50 credits. Takes 1–3 minutes; waits by default and returns the assigned name (personId) and the composite asset. The name is picked automatically from a curated pool.

List the text-to-speech providers and preset voices available on VIVID (Deepgram Aura-2, MiniMax HD presets, OmniVoice voice design, Gemini 3.1 Flash) with their credit cost. Use it before vivid_generate_voice.

Synthesize speech from text with a preset, designed or previously saved voice. This tool never clones a voice from an audio file or URL. Providers: "gemini" = Google Gemini 3.1 Flash presets like Kore/Puck/Zephyr/Charon/Aoede (1 credit/clip), "minimax" = MiniMax Speech 2.8 HD presets (ids from vivid_list_voices → minimax[]) or the id of a voice the account holder saved in the VIVID web app, with an optional emotion (1 credit/clip), "omnivoice" = voice designed from an English description (0.5), "deepgram" = Aura-2 presets (free). Saved voices can only be created in the VIVID web app, after the account holder confirms they own the voice, gives biometric consent and confirms they are 18+; this tool cannot create, import or modify a saved voice. Returns a public MP3 URL (7-day temp storage) and optionally downloads it.

List the account projects (brand containers) — use a project id to group generations.

Generate an original music track from a text brief (genre, mood, instruments, BPM, use). Credits are charged only when the track is ready. Providers: "suno-v6" (default, 6 credits for TWO takes of the same brief: url plus alternates[] — let the user pick) — the brief is the style; durationSec (10–360) is honoured within ~1–3 s for instrumentals and for vocals with lyrics; vocals WITHOUT lyrics let Suno write the words and choose the length (~2–3 min). Usually ready in 20–90 s. "minimax-music-3.0" (14 credits, one take) — 44.1 kHz/256 kbps MP3, instrumental or with vocals; it has NO exact length control: durationSec is a strong hint (30 s asked → ~2 min delivered), so trim in the editor. Lyrics accept [Verse]/[Chorus]/[Bridge] tags on both. Rendering can take up to 3 minutes: the tool waits up to waitSec, then returns status: "processing" with a taskId — call vivid_music_status with it to get the track. Returns a public URL (7-day temp storage) usable as an editor audio clip or video soundtrack; set outputDir to also download it.

Get the track of a vivid_generate_music call that returned status "processing". Waits up to waitSec; returns the public URL when ready (status "completed"), or "processing" again. Credits are charged once, when the track is ready; failed tracks are free.

Music generation providers with credits, capabilities (exact duration, vocals, max length) and whether each is configured on the server.

Search the Freesound.org library for existing music, loops and sound effects (free, no credits). By default only licenses usable in commercial content are returned: CC0 (no attribution) and CC BY (credit the author — attribution is the line to put in the caption or credits). Results carry an MP3 preview URL (~128 kbps, fine for social video): pass it as a media source to vivid_create_editor_project / vivid_edit_timeline, or to vivid_analyze_audio for beats. For a track written to a brief, use vivid_generate_music instead.

Speech-to-text with Deepgram Nova-3 (free). Input: a VIVID asset id, a public URL, or a local audio/video file (uploaded for you). Returns JSON with words[] (raw per-word startMs/endMs — use these for voice↔subtitle alignment), readable cues[] (3–8 words, timing stretched for legibility) and the full transcript; or a ready-to-use SRT / VTT file (format=srt|vtt, granularity=cue|word), optionally written to outputPath. Languages: it, en, es. Max 100 MB.

Edit ONLY one object/area of an image and leave every other pixel untouched — e.g. "the ring on the left hand" → "make it a plain yellow gold band". Say WHERE with target (plain text: the tool locates it with a two-pass labelled-grid vision step, ~40–60 s), or region {x,y,w,h} as fractions of the frame, or a mask image (white = edit). A square crop around the target goes to the edit model at full resolution and the original bytes are composited back outside a feathered region, so the rest of the photo is byte-identical. Models: nano-banana (default, 6 credits), seedream (4), grok (3), z-image (2, true mask inpaint — needs region or mask). Returns the new asset id + download URL; chain with vivid_compare_product to verify against the SKU and feed its fixPrompt back here.

Visual quality control: checks that the product in a generated/retouched image matches the real product photo (SKU card). Vision LLM compares shape, proportions, materials/finish, colours, stones/elements, logos, hardware — product only, background ignored. Returns match 0–1, verdict pass|review|fail, a list of differences with severity, a summary and a fixPrompt you can pass straight to vivid_retouch. 1 credit per call (a retry is charged again), ~15–25 s.

The reference sheet of a saved product is a 2×2 grid (front, two category-specific angles, worn/in use) drawn by Nano Banana 2.1 from the product photos; every panel is checked against the real photo with the same vision QC as vivid_compare_product. This tool only reads: it returns the current sheet (four panels and the grid with download links, the QC scores, or the failure message) and what a new one would cost, and waits up to waitSec while one is being made. Free. To make a new sheet use vivid_create_product_sheet.

The reference sheet of a saved product is a 2×2 grid (front, two category-specific angles, worn/in use) drawn by Nano Banana 2.1 from the product photos; every panel is checked against the real photo with the same vision QC as vivid_compare_product. Starts a new sheet: 13 credits, or free for the first sheet of a free account (vivid_product_sheet shows the price first, for free). One failed check is regenerated for free, a second failure refunds everything. Takes 1–3 minutes: waits up to waitSec, then poll with vivid_product_sheet. If a sheet for the product is already being made, it waits for that one and charges nothing. Angles the photos do not show are reconstructed: pass an extra photo of the side/back in extraPhotos for a faithful sheet.

List the saved video-editor projects (timelines) of the account. Use the id with vivid_render_project.

Tempo and transient analysis of a music / voice file, computed on the VIVID server (free, cached per asset). Returns bpm, firstBeatMs, the beat grid (beats[] / downbeats[] in ms from the start of the file, constant tempo) and peaks[] — the detected hits/accents/drops ({ms, strength 0–1}). Use beats for metronomic cuts and peaks for cuts on what the ear hears (dynamic montage, "tagli sui picchi"): pass them as clip boundaries (UPDATE_CLIP startMs/durationMs) or let CUT_TO_BEAT do it (grid beats|peaks). Input: VIVID asset id, public URL or local file (uploaded for you). MP3 or WAV only. First 90 s analysed by default.

Load a saved editor project and return its timeline: canvas preset, tracks, media library (assets with local ids), clips (id, track, start/duration/source offset, speed, transitions, animations), canvas objects (position/size/rotation, effects, masks) and text overlays. Use the ids in vivid_edit_timeline. Pass raw=true to get the full project JSON instead.

Create a new editor project (timeline) on the account: pick a canvas preset, import media (VIVID assets, URLs or local files), optionally apply an initial batch of commands, and save. Returns the project id (use it with vivid_edit_timeline / vivid_render_project) and the timeline summary. Media added here is available to ADD_CLIP by its local id (= asset id unless localId is given). Commands are the same editing operations as the VIVID editor: [{ "type", "payload" }], one typed payload per type. Timeline: ADD_CLIP {assetId, canvasObject?{x,y,w,h,rotation,opacity,blendMode}} (places the whole asset at the first free gap of a matching track; then trim it with UPDATE_CLIP) · UPDATE_CLIP {id, startMs?, durationMs?, sourceOffsetMs?, playbackRate?, volume?, opacity?} · REMOVE_CLIP {id} · SPLIT_CLIP {clipId, splitAtMs} · DUPLICATE_CLIP {id} · DETACH_AUDIO {clipId} · ADD_TRACK {type:'visual'|'audio'} · CUT_TO_BEAT {audioClipId (an AUDIO clip id in the timeline), beatInterval (1–32: beats or peaks between cuts; 4 = one cut per bar), grid?: 'beats' (constant BPM grid, default) | 'peaks' (detected hits/accents — right for cuts "sui picchi", drops, voice, music without a steady tempo), minStrength? (grid peaks: 0–1, keep only hits ≥ it), mode?: 'redistribute' (default: re-slice all visual clips in order to fill the music) | 'snap' (only nudge existing edges), syncClips?: 'all' | [clipIds]} — the audio is analysed on the server automatically (free); to plan cuts by hand call vivid_analyze_audio and use its beats/peaks (ms) as UPDATE_CLIP boundaries. Look: SET_TRANSITION {clipId, edge:'in'|'out', type:'dissolve'|'fade-black' (visual) | 'crossfade'|'fade-in'|'fade-out' (audio), durationMs} · SET_CLIP_ANIMATION {clipId, phase:'in'|'out', type:'fade'|'slide-up'|'slide-down'|'zoom-in'|'zoom-out'|'zoom-pan'|'spin'|'blur-reveal', durationMs} · SET_CINEMATIC_LOOK {look} · SET_CANVAS_PRESET {id,label,width,height} (crop/reframe: landscape 1280x720, landscape-fhd 1920x1080, portrait 720x1280, portrait-fhd 1080x1920, square 1080x1080, social 864x1080) · UPDATE_CANVAS_OBJECT {id, x?, y?, w?, h?, rotation?, blendMode?}. Text: ADD_TEXT {text, startMs, endMs, x (0-1), y (0-1), fontFamily?, fontSize?, color?, animation?, fill?} · UPDATE_TEXT {id, ...} · REMOVE_TEXT {id}. fill = animated gradient on the glyphs: {type:'gradient', colors:['#ff6ec4','#7873f5','#4ade80','#facc15'], angle: 100, animate:'shift'|'none', speed?: periods/s (0.35)} — presets: iridescent, sunset, ocean, gold, candy (same colors as the in-app Fill picker). Subtitles (CapCut-style word-level captions): SET_SUBTITLES {cues:[{text, startMs, endMs, words?:[{word,startMs,endMs}]}] | null, templateId?:'outline-reveal'|'ugc-pop'|'karaoke-marker'|'cinematic-fade'|'neon-cyberpunk'|'playful-wiggle', position? (0 top – 100 bottom), styleOverrides?:{fontSize,color,highlightColor,…}, layout?:{x,y,scale,maxWidthPct,rotation}}. Get cues from vivid_transcribe (its cues[] carry per-word timing) on the voice/video asset; words are synthesized evenly when omitted. Keyframes & effects (v2): SET_ANIMATIONS {objectId, objectType:'canvas'|'text', animations:{x|y|scale|rotation|opacity|blur|...: {keyframes:[{t (ms), v, easing?}]}}} · SET_OBJECT_ANIMATION · ADD_EFFECT {objectId, objectType, effect:{type:'blur'|'vignette'|'colorGrade'|'grain'|'glitch'|'chromaticAberration'|'pixelate', ...}, replace?} · REMOVE_EFFECT {objectId, objectType, effectType} · SET_EFFECT_KEYFRAMES {objectId, objectType, effectType, param, keyframes:[{t (ms FROM THE CLIP START — not the timeline), v, easing? 'linear'|'easeInCubic'|'easeOutCubic'|'easeInOutCubic' (default)|'easeOutBack'|'easeOutElastic'|'easeOutBounce'|[x1,y1,x2,y2] bezier}] | null} — animates one value of an effect already on the object (ADD_EFFECT first, same batch is fine): cameraShake.intensity|speed, motionBlur.intensity, flash.intensity|decayMs, zoomBlur.amount|centerX|centerY, lightLeak.intensity|speed|hue, blur.amount, vignette.intensity|size, colorGrade.brightness|contrast|saturation|hue, grain/glitch/chromaticAberration/pixelate values. Use it for anything that changes over time (a shake that builds up, a leak that fades in). Motion & light effects (canvas objects only): {type:'motionBlur', intensity 0–1 (shutter = intensity × 2 frames; blurs the clip's OWN movement — keyframes, intro/outro, shake, beat pump — not motion inside the footage, so pair it with SET_ANIMATIONS or SET_CLIP_ANIMATION), samples? 2–32} · {type:'cameraShake', intensity 0–1 (1 = ±3% of the short side, ±2°), speed? Hz 0.5–20 (6), seed?} · {type:'flash', intensity 0–1, color? '#rrggbb' (white), everyBeats? 1–16, bpm? (used only when the project has no analysed music), decayMs? 30–2000 (150)} · {type:'zoomBlur', amount 0–1, centerX? centerY? 0–1} · {type:'lightLeak', intensity 0–1, speed? 0.1–4, hue? 0–360 (25 = warm)}. Effect recipes (one style across the edit, preferred over hand-written effects/keyframes): APPLY_EFFECT_RECIPE {recipe:'beat-flash' (flash pinned to the music's real beats — the audio is analysed on the server automatically, like CUT_TO_BEAT) | 'punch-in' (scale pop + motion blur on every cut) | 'whip' (fast slide-in from alternating sides + motion blur on every cut) | 'handheld' (subtle shake) | 'light-leak' | 'clear' (removes what recipes added), intensity? 0–1 (0.6), clipIds? (default all visual clips), everyBeats? (beat-flash), color? (beat-flash), audioClipId? (beat-flash)} — its keyframes sit on the timeline, so send it AFTER CUT_TO_BEAT or anything that moves clips. · SET_MASK {objectId, mask:{type:'rect'|'circle'|'reveal'|'clipPath'|'none', ...}} · ADD_AUDIO_REACTIVE. Speed: UPDATE_CLIP.playbackRate is constant per clip; SET_SPEED_RAMP {clipId, preset:'speed-up'|'slow-down'|'slow-mo-hit'|'punch-in'|'ease-in-out'|'none'} or {clipId, keyframes:[{t (ms from clip start on the timeline), v (rate 0.1–8)}]} gives a variable speed (linear between keyframes; the clip keeps its timeline duration unless the source runs out, then it is shortened). Ids: read them from vivid_get_editor_project or from "created" in the previous result; a batch is applied one command at a time, so a later command can use a clip created earlier in the same batch.

Apply editing commands to a saved editor project — clips, trims, speed and speed ramps, transitions, canvas preset (16:9 ↔ 9:16), texts (incl. animated gradient fills), word-level subtitles, masks, keyframes, audio — through the same engine as the VIVID web editor, then save it back under the same id. Optionally import media first. Returns applied/errors (plus warnings for things applied but likely not visible, e.g. motion blur on a still clip), the ids created, and the updated timeline. Render the result with vivid_render_project. Commands are the same editing operations as the VIVID editor: [{ "type", "payload" }], one typed payload per type. Timeline: ADD_CLIP {assetId, canvasObject?{x,y,w,h,rotation,opacity,blendMode}} (places the whole asset at the first free gap of a matching track; then trim it with UPDATE_CLIP) · UPDATE_CLIP {id, startMs?, durationMs?, sourceOffsetMs?, playbackRate?, volume?, opacity?} · REMOVE_CLIP {id} · SPLIT_CLIP {clipId, splitAtMs} · DUPLICATE_CLIP {id} · DETACH_AUDIO {clipId} · ADD_TRACK {type:'visual'|'audio'} · CUT_TO_BEAT {audioClipId (an AUDIO clip id in the timeline), beatInterval (1–32: beats or peaks between cuts; 4 = one cut per bar), grid?: 'beats' (constant BPM grid, default) | 'peaks' (detected hits/accents — right for cuts "sui picchi", drops, voice, music without a steady tempo), minStrength? (grid peaks: 0–1, keep only hits ≥ it), mode?: 'redistribute' (default: re-slice all visual clips in order to fill the music) | 'snap' (only nudge existing edges), syncClips?: 'all' | [clipIds]} — the audio is analysed on the server automatically (free); to plan cuts by hand call vivid_analyze_audio and use its beats/peaks (ms) as UPDATE_CLIP boundaries. Look: SET_TRANSITION {clipId, edge:'in'|'out', type:'dissolve'|'fade-black' (visual) | 'crossfade'|'fade-in'|'fade-out' (audio), durationMs} · SET_CLIP_ANIMATION {clipId, phase:'in'|'out', type:'fade'|'slide-up'|'slide-down'|'zoom-in'|'zoom-out'|'zoom-pan'|'spin'|'blur-reveal', durationMs} · SET_CINEMATIC_LOOK {look} · SET_CANVAS_PRESET {id,label,width,height} (crop/reframe: landscape 1280x720, landscape-fhd 1920x1080, portrait 720x1280, portrait-fhd 1080x1920, square 1080x1080, social 864x1080) · UPDATE_CANVAS_OBJECT {id, x?, y?, w?, h?, rotation?, blendMode?}. Text: ADD_TEXT {text, startMs, endMs, x (0-1), y (0-1), fontFamily?, fontSize?, color?, animation?, fill?} · UPDATE_TEXT {id, ...} · REMOVE_TEXT {id}. fill = animated gradient on the glyphs: {type:'gradient', colors:['#ff6ec4','#7873f5','#4ade80','#facc15'], angle: 100, animate:'shift'|'none', speed?: periods/s (0.35)} — presets: iridescent, sunset, ocean, gold, candy (same colors as the in-app Fill picker). Subtitles (CapCut-style word-level captions): SET_SUBTITLES {cues:[{text, startMs, endMs, words?:[{word,startMs,endMs}]}] | null, templateId?:'outline-reveal'|'ugc-pop'|'karaoke-marker'|'cinematic-fade'|'neon-cyberpunk'|'playful-wiggle', position? (0 top – 100 bottom), styleOverrides?:{fontSize,color,highlightColor,…}, layout?:{x,y,scale,maxWidthPct,rotation}}. Get cues from vivid_transcribe (its cues[] carry per-word timing) on the voice/video asset; words are synthesized evenly when omitted. Keyframes & effects (v2): SET_ANIMATIONS {objectId, objectType:'canvas'|'text', animations:{x|y|scale|rotation|opacity|blur|...: {keyframes:[{t (ms), v, easing?}]}}} · SET_OBJECT_ANIMATION · ADD_EFFECT {objectId, objectType, effect:{type:'blur'|'vignette'|'colorGrade'|'grain'|'glitch'|'chromaticAberration'|'pixelate', ...}, replace?} · REMOVE_EFFECT {objectId, objectType, effectType} · SET_EFFECT_KEYFRAMES {objectId, objectType, effectType, param, keyframes:[{t (ms FROM THE CLIP START — not the timeline), v, easing? 'linear'|'easeInCubic'|'easeOutCubic'|'easeInOutCubic' (default)|'easeOutBack'|'easeOutElastic'|'easeOutBounce'|[x1,y1,x2,y2] bezier}] | null} — animates one value of an effect already on the object (ADD_EFFECT first, same batch is fine): cameraShake.intensity|speed, motionBlur.intensity, flash.intensity|decayMs, zoomBlur.amount|centerX|centerY, lightLeak.intensity|speed|hue, blur.amount, vignette.intensity|size, colorGrade.brightness|contrast|saturation|hue, grain/glitch/chromaticAberration/pixelate values. Use it for anything that changes over time (a shake that builds up, a leak that fades in). Motion & light effects (canvas objects only): {type:'motionBlur', intensity 0–1 (shutter = intensity × 2 frames; blurs the clip's OWN movement — keyframes, intro/outro, shake, beat pump — not motion inside the footage, so pair it with SET_ANIMATIONS or SET_CLIP_ANIMATION), samples? 2–32} · {type:'cameraShake', intensity 0–1 (1 = ±3% of the short side, ±2°), speed? Hz 0.5–20 (6), seed?} · {type:'flash', intensity 0–1, color? '#rrggbb' (white), everyBeats? 1–16, bpm? (used only when the project has no analysed music), decayMs? 30–2000 (150)} · {type:'zoomBlur', amount 0–1, centerX? centerY? 0–1} · {type:'lightLeak', intensity 0–1, speed? 0.1–4, hue? 0–360 (25 = warm)}. Effect recipes (one style across the edit, preferred over hand-written effects/keyframes): APPLY_EFFECT_RECIPE {recipe:'beat-flash' (flash pinned to the music's real beats — the audio is analysed on the server automatically, like CUT_TO_BEAT) | 'punch-in' (scale pop + motion blur on every cut) | 'whip' (fast slide-in from alternating sides + motion blur on every cut) | 'handheld' (subtle shake) | 'light-leak' | 'clear' (removes what recipes added), intensity? 0–1 (0.6), clipIds? (default all visual clips), everyBeats? (beat-flash), color? (beat-flash), audioClipId? (beat-flash)} — its keyframes sit on the timeline, so send it AFTER CUT_TO_BEAT or anything that moves clips. · SET_MASK {objectId, mask:{type:'rect'|'circle'|'reveal'|'clipPath'|'none', ...}} · ADD_AUDIO_REACTIVE. Speed: UPDATE_CLIP.playbackRate is constant per clip; SET_SPEED_RAMP {clipId, preset:'speed-up'|'slow-down'|'slow-mo-hit'|'punch-in'|'ease-in-out'|'none'} or {clipId, keyframes:[{t (ms from clip start on the timeline), v (rate 0.1–8)}]} gives a variable speed (linear between keyframes; the clip keeps its timeline duration unless the source runs out, then it is shortened). Ids: read them from vivid_get_editor_project or from "created" in the previous result; a batch is applied one command at a time, so a later command can use a clip created earlier in the same batch.

Queue a render (MP4 export) of a saved video-editor project. VIVID renders in the browser: the job returns an openUrl — open it (or pass openBrowser=true to open it automatically on this machine) and the editor renders, uploads the MP4 to the gallery and marks the job done. Poll with vivid_render_status. Requires being logged in on vividai.tv in that browser, preferably on a computer: phone browsers may stop a long render when the tab goes to the background.

Status of a render job created with vivid_render_project: queued, rendering (with progress), completed (with the output asset and download URL) or failed.

Record a screen-capture video of vividai.tv (or any site) by driving a local Chromium with Playwright: a scripted list of steps (navigate, move, click, type, scroll, wait) is executed with a visible cursor, smooth mouse moves and click ripples, and the result is uploaded to VIVID as a video asset ready for vivid_create_editor_project / vivid_edit_timeline (pass the returned durationMs as media[].durationMs). On VIVID sites the browser is logged into the account of VIVID_API_KEY automatically (login=true), so app pages open directly; other sites never receive the session. Runs on THIS machine: needs Google Chrome (or npx playwright install chromium) and, for mp4 output, ffmpeg (otherwise webm, which the editor handles). Default site: https://vividai.tv. Tips: onboarding on /create can be skipped with click "[data-testid=onboarding-skip]"; keep clips short (one feature per recording) and slow the pace with waitMs. Costs no credits.