WideCast Playground

Read-only, free. Your account profile, remaining credits, and connected platforms.

Read-only, free. The social platforms currently connected to your account.

Read-only, free — but SLOW (aggregates across connected platforms). Pick a period.

Synchronous — Run returns video ideas derived from a product/service description (≥10 chars) immediately. Consumes credits.

Generate written content (blog / Facebook / X / LinkedIn post) from a URL, an idea/topic, or pasted text. Async: submit, then this card auto-polls `/v1/status` until `completed` — `result.review_url` is the public content viewer. Consumes credits.

Synchronous. Charges **1 credit per image generated** (count=4 → 4 credits). Returns an `image_set` with each image labeled by `number` so an AI agent can render a numbered thumbnail list for the user to pick by number.

Every section below is a working example you can run as-is. Pick an industry pill to load a domain-specific script (Real Estate market data, Insurance regulation, News structural analysis…). WideCast is built for videos with depth — substantive narration, story structure, B-roll that matches the subject — not shallow entertainment clips.

Synchronous, free. Manages a per-video in-memory edit session that makes an AI editing run safe under PARALLEL scene edits and gives instant read-after-write. `start` caches the whole video in memory (crash-safe via a write-ahead file) so every /v1/modify_scene write lands in the cache and ElasticSearch is written ONCE at commit; `commit` flushes to storage (REQUIRED to finish — idle sessions auto-commit after 45 min); `abort` discards staged edits; `status` reports {active, staged_writes}. If the response has cache_enabled=false the server is in legacy direct-write mode — proceed without a session.

Read-only, free. The curated FOUNDATION-VIDEO template library — proven starter angles for an industry that a user can adapt into a new video. A navigation / discovery aid: creates or modifies nothing. Re-promoted to the public surface 2026-07-13 (Round 30). `industry` falls back to the account's saved industry.

Synchronous for most branches, no credit charged until /v1/export_video re-renders the final MP4. Pick ONE scene by `voice_file` (the stable per-scene UID — preferred), `id` (display/order, UNSTABLE), or `text` (fuzzy). Then send exactly ONE field family per call — the only intentional multi-family call is `layout.batch`, which composes layout-only children. Branches: (A) mediaUrl/mediaType — background media swap (roll-aware) (B) remotion.upload_overlay — FREE agent-supplied image → Remotion spec (NOT Regenerate Overlay) (C) remotion.object.rect — PREFERRED overlay layout (read layout_id from /v1/scene_geometry first; uses 280×498 preview coords) (D) remotion.group.rect — low-level wrapper edit (canvas or preview) (E) overlay.narrator.rect — narrator box layout in 280×498 preview (F) overlay.caption.y — vertical-only caption placement in 280×498 (G) layout.batch — composes E/F/C/D atomically (one persist, one MQTT) (H) voice.upload — async; user-supplied narration audio (I) narrator.upload_video — async; user-supplied A-roll video (J) roll.active / roll.switch — sync A/B-roll data switch (K) segment.text — text correction (keeps audio timing) (L) metadata — pattern / type / visual / keyword / quote / talking_point segment.remotion_spec='none' = user-disabled overlay; do NOT auto-enable.

Synchronous, no credit. Push a notification to YOUR OWN account. EMAIL is the default channel (always sent when the account has an email on file); if you have also completed `Connect Telegram` at https://widecast.ai/#setup, the same notification is ALSO sent to Telegram. `subject` (required) is the email subject line and is prepended in bold before the Telegram body. No email + Telegram connected → Telegram only; neither → 400 no_delivery_channel.

Free. Save one platform's publish PREFERENCES (e.g. youtube privacy, reddit subreddit, facebook page id). ⚠ Running writes to your account.

Read-only, free. The saved per-platform publish preferences (privacy / page / subreddit) — not access controls.

Read-only, free. The weekly production plan (ideas + topics). Omit the week range for a pure read of the current plan.

Distribute content to the social platforms connected to your account. Provide **exactly one** of `topic_id` (an existing WideCast video/blog), `text` (arbitrary post), or `video_url` (an external video FILE url, needs `title`). `platforms` defaults to **all connected**. **⚠ Running posts PUBLICLY and charges 1 credit.** The response carries the upstream `request_ids` — poll `GET /v1/status/{request_id}` for per-platform post URLs.

Read-only, free. The account's content roadmap — weeks, slots, completion, streak.

Synchronous, free, read-only. Use AFTER /v1/video_data and BEFORE /v1/modify_scene or /v1/scene_inspector. Resolves one scene by stable voice_file (preferred), loads {voice_file}_spec.json if available, and returns narrator/caption/Remotion Storyboard display boxes in one 280×498 editor-preview coordinate space — plus dead zones, collision violations, and warnings. Designed for LLM-only agents that cannot see images: reason over JSON, then call /v1/modify_scene with a layout.batch (or single-field) edit. Never broadcasts MQTT, never renders screenshots, never modifies the video.

Synchronous, free. Use ONLY after data-first audit with /v1/video_data, when you specifically need BROWSER TRUTH (mounted DOM, computed boxes, preview play state, a small 280×498 screenshot) — not what's stored in ES. Server publishes a tiny MQTT probe to every open editor tab for this video, elects ONE healthy foreground/active browser within ~800ms, then sends the command there. If no editor is open the response is `status="unavailable"` with `code="no_live_editor"` (no presence) or `"no_active_editor"` (presence but unresponsive) — fall back to /v1/video_data + fetching `remotion_spec_url`.

Synchronous, free. Search stock B-roll for a keyword. Use **kind=video** for stock clips (Pexels/Pixabay/Shutterstock — same engine as the broll.js Stock tab) or **kind=image** for real photos (Google — broll.js Photos tab). Returns a list with each result labeled by `number` so an AI agent can render a numbered thumbnail list for the user to pick by number. **Special**: when `keyword` is the exact single word `"grid"` (case-insensitive) AND `kind="video"`, the server skips Pexels/Pixabay/Shutterstock and returns WideCast's curated internal grid-background video list in the SAME `results[]` shape. Phrases like `"grid background"` still hit normal stock search.

Load the WideCast AI-VIDEO-EDITOR playbook for refining/editing an EXISTING video. Different from /v1/skills/writing — that one is for AUTHORING new scripts; this one is for EDITING already-generated scenes. **Delivery mode**: download_zip (experiment, 2026-06-30). The response carries (a) a `preload_tools` hint block + (b) a 4-step instruction: mkdir + curl + verify + unzip, then Read `./.widecast-skill-video-editing/video-editing/SKILL.md` from local disk and follow its LOAD MAP. The skill content is identical to previous inline-content mode (master INDEX + 5 core modules + per-scene modules + style libs). **Preload tools first** (`preload_tools` field): before running the 4 setup commands, copy `preload_tools.one_shot_query` into your MCP host's `ToolSearch` `query` param. This loads EVERY WideCast tool (video_data, scene_geometry, scene_inspector, modify_scene, search_broll, upload_asset, etc.) in one round-trip. Some hosts (Codex, Claude Code) default to NARROW keyword search and drop critical tools like `widecast_modify_scene` — the edit flow then fails mid-run. The list is auto-derived from the live MCP registry, so adding a new tool later is zero-config. Fallback for hosts without `select:` syntax: `preload_tools.broad_query` (`widecast`) with `broad_max_results` (~30). **Always-returned `contract` field** (~1.5 KB cross-cutting rules) survives even if the agent cannot complete the download — it is the safety net. **Host requirements**: local shell + curl + unzip + a file-reading tool (`Read`/`cat`/view_file). If your host has none of these, you cannot follow the full skill in this mode. **Adding/editing modules**: drop `.md` files under `widecast/skills/video-editing/` → `bash deploy_widecast.sh` (no Flask restart needed for content updates). Server picks up new zip mtime automatically on next request. **Reverting**: server env `WIDECAST_EDITING_DELIVERY_MODE=inline_content` + Flask restart → falls back to multi-module inline-content mode (with `available_modules[]` + per-module `?module=` lazy-load).

Synchronous, no credit. Upload an audio / video / image / document (HTML / PDF / Markdown) file and get back a public S3 URL valid for 24 hours. Use the URL as `audio_url` / `video_url` on /v1/create_video, paste into `script_text` as an inline `` for images, or `curl ` to fetch a document body before passing it as blog_text / script_text. Server cap 500 MB.

Synchronous, free. First step for data-first scene audit/edit — call this BEFORE /v1/modify_scene or /v1/scene_inspector. URL-RESPONSE CONTRACT: the HTTP response is a small envelope; the full video_data tree is published to a temporary public JSON file on `origin.widecast.ai` and the envelope carries `data.url` + size + 15-min TTL + a 3-step `instructions` block. The agent fetches the JSON file with `curl ` and reads it locally — the full tree is NOT inlined. Bypasses MCP per-tool-call output caps that previously truncated 50–300 KB payloads mid-JSON. Every call publishes a NEW unique file so URLs are never reused (matches realtime semantics). The published JSON file content is the same as the legacy inline response: annotated segment dicts with stable-identity fields, Remotion spec metadata, and coordinate-space hints (720×1280 canvas vs 280×498 legacy editor preview). `voice_file` is the stable per-scene UID; `segment.id` is only current display/order.

Read-only, free. Lists the account's recent videos/scripts (20 per page).