Connector tools reference

The creator's series, episodes, chapters, takes, characters, places and props, newest first, 20 at a time. Optionally narrow by words in the name or title, by kind (default: series and episodes), by series, by status (not rendered, rendering, done) or by changed since a date. Each item has its title, kind, where it sits, status, last changed, thumbnail, its ids, and for episodes the link to open it in the app.

A series' bible — audience, platform, target length, format rules, and the recurring cast, places and props with their default looks — plus every episode at a glance: number, title, one-line story, status, runtime, takes rendered, last changed, thumbnail and link. Long series come in pages of 25 episodes.

An episode's brief and look (lens, light, palette, era), its story plan (premise, emotional promise, world rules, character arcs, voiceover), the acts and sequences outline, every chapter with its clip count, runtime and render status, the finished film link (final_film_1080p_url too, only while its 1080p film matches the current film), the episode's settings (aspect ratio, resolution, clip length, language, and clip_limits: the lengths its video model renders, pictures and speaking voices per clip, the voice-sample seconds a clip may carry, and how many places one clip may be set in), and its link in the app. Read clip_limits before planning or writing clips. Says plainly what the episode does not have yet.

Every clip of a chapter in order: what it is meant to show, the screenplay with dialogue, narration, feeling, sound, the camera in plain words, length and render status, with a thumbnail picture of each rendered clip. A clip with place pictures that are out of date or the wrong shape carries place_pictures_out_of_date, with the newer_place_picture_ids to swap in; the other clips do not.

A clip in full: what it is meant to show, the screenplay with dialogue, narration, feeling, sound, the camera in plain words, length and render status; who and where is on it with the look each uses; every rendered take with its video link, the picked one marked; for a cut clip, what is kept and cut on the original render's clock; film_speed when not 1; and the pictures it is made from, each labelled with its role (character, place, or the previous clip's end). saved_frames: frames kept under it (id, second, label, picture). reference_images: pictures pinned to the take, in order, with type and url. moments: the stills painted for it, whether each is on the take and the pictures it was painted from; they come as images: look at each beside its pictures. ready_to_render: whether it can be filmed now, and waiting_on what blocks it. place_pictures_out_of_date: its place pictures that are out of date. opening: the clip before and both clips' places. fingerprint changes whenever anything about it changes. in_story: its chapter with its story, and the clips either side. include_pictures false skips the pictures. To change anything, ask the studio with talk_to_agent.

Every character, prop and place in one or more episodes or series: description, looks and outfits with their look ids, how each character sounds, lasting changes to a character and the story moment they start, and each one's card picture. An episode also lists its picture tags: the names its clips point at pictures by (#Maya, #AvenueNorth), each with what it means.

Every picture of one or more places, as images: for a filmed place, every view cut from its current place films. Each view has its place picture id, the place id, its time on the film, its description, its look, whether it is hidden, where it was taken from and which way it faces, and whether it is named yet. Give episode_id to also get each place's takes_on_older_pictures: that episode's clips still on out-of-date pictures of it. To change a place or its pictures, ask the studio with talk_to_agent.

Every picture of one or more characters or props at full size: face close-up, full body, turnaround, the glued sheet sent with each take, and each look's pictures, each with its picture id and kind, plus the asset id and look ids.

What one or more episodes say and play: the narrator's voiceover lines per clip, the music cues (what each is for, its mood and lyrics, the chapters it spans, whether it is composed, in the film and placed — a composed piece plays in the film placed or not — where it plays in episode seconds, its volume curve, whether and how far it dips under speech, and its hand-off), and on-screen text and subtitles with their times in seconds. With include_sound_map true (slower, it measures every clip's sound, so ask for it only on the episodes you need) it adds each episode's sound map: numbers only, in episode seconds — chapters with their places, end credits, each clip's slot and the cuts between clips, where people speak, the clips not yet scanned for speech, the quiet gaps and loud peaks of the clips' own sound with their dB levels, and each piece of music on the film with its lane, start and end, volume curve, dip under speech and hand-off. Read it before planning music for a whole film; it judges nothing. Speech in unscanned clips is unknown, not absent: listening for it is a small paid step only the studio takes, so once the creator agrees, ask the studio with talk_to_agent to listen for where people speak, then read the map again.

Watch one or more rendered clips (up to 5 per call). For each: the picked take (or the take you name) as grid images of up to 16 frames with the time on every tile, read left to right, top to bottom; the first and last frame at full size; the pictures the clip was made from, to compare with what came out; the screenplay, narration, sound and music, and timed spoken lines when the episode has subtitles; which take it is, its length, and a link to its video for a client that can play video. For a cut clip, the stretches kept and thrown away. film_speed when the film plays it faster or slower. A clip that is not rendered yet returns only its pictures; a clip that is rendering says to try again shortly. frames_per_second is your choice, anywhere up to 4; left out, it is 2, the least that shows motion: watch the whole take at 2 or more before calling a render good. Frames carry no sound, so say you could not judge voices, lip sync or music. Acting, camera moves, background life and natural sound you didn't write come from the studio's own direction and are intended. Flag them only when they break the story, such as an extra person speaking or a changed place. Flag a cut the clip didn't ask for: cuts are where continuity breaks. Watching is free.

Watch an episode 5 clips per page, in playing order. Each clip comes as one grid image of up to 16 timed frames (left to right, top to bottom), with its screenplay, narration, sound, music and timed lines when there are subtitles, and its cut points: where it starts and ends in the finished episode, plus a link to its video. Also returns the total number of pages. Clips not rendered yet, or rendering, say so. Use watch_clip for a closer look. Watching is free.

Every post the creator has made, newest first, 20 per page. Each post has its id, status, platforms, a short preview of its words, when it was made, scheduled and published, links to where it was published, and the chat it lives in: a video studio episode or a posts chat, with the link that opens it in Agent Social. A video post also carries its series and episode ids. Filter by status, platform, kind, episode, chat, dates, or words in the post. Pass next_cursor back as cursor for the next page. Read only: publishing happens in the app; to schedule a post, ask its agent (the studio or the posts agent) with talk_to_agent.

One or more post cards, by post id or by episode id. Each card has its ids, the platforms and accounts it goes to, the X post text, the YouTube title, description, tags and category, the Instagram caption and the Facebook title and description (video posts), the thumbnail or pictures, whether it tells sites it was made with AI, its state with links to where it was published, and the chat it lives in with its link. To change a post, ask the studio (a film's post) or the posts agent (a text or image post) with talk_to_agent; ask the same agent to schedule it. Publishing happens in the app.

Files the creator uploaded (photos, videos, audio), newest first, 10 per page, each with its id, name, kind and upload time, and a picture of each image. Search by words in the file name. Pass next_cursor back as cursor for the next page.

End credits are closing videos played as they are at the end of a finished film (a logo sting, a sign-off). Lists the creator's end credits with their id, name, shape (16:9, 9:16…), length and whether it is the shared Agent Social card. Give episode_id to also see what that episode ends on now. Free.

For one or more episodes: whether the studio is working on it right now, its latest progress lines, and which clips are rendering.

The creator's credit balance in dollars. Reading and watching are always free.

The complete, current guide to Agent Social: how films, clips, moments, place views and pictures work, how to work in the account step by step, and what the studio can do. Call it once at the start of a session, before writing or asking for any take or moment. Free.

Check on a turn started with talk_to_agent: whether it is still working, done, failed or stopped, what the agent has said, its progress lines, what changed, running_seconds (how long it has run) and seconds_since_progress (since it last moved). Answers at once with the turn as it is now; pass back the returned since value to get only what is new. Check again about once a minute until it is done, failed or stopped (a render or a painting takes minutes), never several calls in a row. next says what to do now: the moment a turn is done, failed, stopped or refused, tell the creator. gone_quiet means no progress for 5 minutes: tell the creator it has gone quiet rather than wait in silence. A finished painting comes back as pictures beside the pictures it was painted from: look at them before telling the creator, and if something is off, ask the studio to fix that one thing. changed names what changed; read what it touched (get_clip, watch_clip) before telling the creator. waiting_for_creator is true once a done turn's last reply asks the creator something the studio will not go on without, with choices: tell the creator its question and choices in plain words, and send their answer with talk_to_agent only once they give it; never answer for them.

The Agent Social link where the creator can watch an episode or post change live: the episode's video studio chat, or the chat a post lives in. Accepts episode, chapter, clip, post or chat ids; a chapter or clip gives its episode's chat.

The video and image models in use and the ones that could be chosen. Video: the default a new film starts on and, with episode_id, that episode's model and resolution; then every video model with model_id, whether this episode (or a new film) can use it and why not, its resolutions, the clip lengths it films, whether it makes sound, and a rough price per resolution. Image: the default model, size and quality, and every image model with its sizes and quality levels. Free. The creator changes a model in the app's Model picker or by asking the studio.

Post a chat message to the creator's own Agent Social studio, the AI collaborator that makes their films and series, exactly as if the creator typed it in the app. The message appears in the creator's chat marked as yours, and the studio answers there in its own turn. Pick one: an episode (its studio chat), a chat by session_id, or new_session: true for a brand-new chat. Find an existing episode or chat with list_my_work or list_my_posts. Write one plain ask, like "Generate chapter 1, clip 1", naming things as the creator sees them, never by id, and never text the creator wants kept private. It returns at once with a turn_id: check it with get_turn_result until the turn ends, and share the chat's link so the creator can watch. The studio's work uses the creator's credits, as it does in the app.

Stop what the creator's Agent Social agent is doing in one chat — the chat's own Stop button. The current step finishes and nothing new starts. Name the chat by episode_id or session_id. Stop only when the creator clearly says stop, never on a guess; and never stop a turn to patch a message you sent: gather everything first, then send once.

Create the next episode of a series after the given episode — the app's 'Create next episode' button. The series' core cast always carries over. Leave carry_asset_ids out to carry what the last episode put on screen; give it to carry exactly the characters, places or props it names plus those tagged in asset_ids. The new episode's studio starts with the opening message if you give one (asset_ids and file_ids are tagged with it). continues_without_cut picks up the exact moment the last episode ended. Each episode has at most one next episode; asking again returns the one already made. This uses the creator's credits for the studio's first turn. Returns the new episode and its chat link.