add_element
ChatGPTAdd ONE picture to your elements -- the characters, props and places a scene is held to. kind is character, prop or location; name is what to call it, and a name you already have of that kind gains the picture instead of making a second element; notes says what it is in a few words. source is the picture: the https address the person gave you (never one you guessed) of an image file (jpg, png or webp, up to 15 MB) or of a product's page -- from a page its main product picture is taken, and what the page states about the product (name, brand, colour, size) is kept in the notes as text copied off the page, or gen:<id> -- an image generate_image made here, or one project lists among a project's renders. Returns the element with its photo refs, as elements lists them. Nothing is drawn and no credits are spent: the picture is saved as it is and described in words for the studio.
add_sound
ChatGPTSPENDS CREDITS. Add generated sound to one of your clips: ambience and sound effects made to fit the picture, not speech or music. source is one clip up to 30 seconds. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. prompt says what it should sound like ("rain on a tin roof, distant traffic"). Priced off the clip's length, measured before the quote. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
animate_still
ChatGPTSPENDS CREDITS. Turn one of your images into a short clip with a one-click template: template is one of the names this tool's schema lists, e.g. bullet_time_360, day_to_night or "Muscle Surge". sources is one image, or two for heart_gesture. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. No prompt. The templates written in lower_case run 5 or 10 seconds (default 5); the ones written in Title Case run 5 seconds at resolution 360p, 540p, 720p (the default) or 1080p. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
approve
ChatGPTSPENDS CREDITS. Approve a quoted render for one of your scenes (scene_id), only after the person has seen the price from quote and said yes: what is "keyframes" (draw the stills) or "clip" (render the shots); for a clip pass the tokens from quote and the same renderer choice. Credit is held before anything is submitted and released if the render fails. Starts a background job; poll it with job. Refused when the scene has no reference photos, the quote is stale, or the balance is short.
assemble_clips
ChatGPTJoin clips from your library into ONE video, in the order given: clips are two to twenty gen:<id> or asset:<id> videos from library (the same id may repeat). transition is cut (the default) or crossfade (crossfade_s, 0.1-2 seconds, at most half the shortest clip). music is an optional bed under the clips' own sound -- an asset:<id> audio file from library (kind=audio), trimmed to the picture and lowered under the clips' sound. Clips of different shapes (9:16 beside 16:9) are refused unless letterbox is true, which fits them inside the first clip's frame with black bars. project_id (from projects) files the result under that project. Spends no credits: it runs on the studio's own server. Runs in the background: poll job; the result is a new gen:<id> in your library and an editable cut in the studio's editor.
balance
ChatGPTYour workspace's credits: what is available, what is held by renders in flight, the plan it is on, each credit lot with what is left and when it expires, and any months of a yearly plan still to be released. Credits, never money; credits are added in the Phantra studio. Read-only.
camera_move
ChatGPTSPENDS CREDITS. Make a 5-second clip from one of your images with a named camera move. source is one image. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. move is the move (zoom_in, crane_up, whip_pan, hitchcock and the others this tool's schema lists); prompt says what happens in the shot. style restyles it (anime, 3d_animation, clay, comic or cyberpunk; omit for none) and resolution is 360p, 540p, 720p (the default) or 1080p. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
create_project
ChatGPTStart a new project: a name, an optional brief (who it is for, what it must always have and never show) and an optional look (the style every scene in it is held to). It appears on your projects board in the studio, and write_scene with its project_id files a scene under it. Spends nothing.
credit_history
ChatGPTWhere your credits went, newest first: every hold, charge, return, grant and expiry, each with what it was for and the balance it left, in credits only. Pass the answer's next as before to page further back; limit is 1 to 200 (default 20). Read-only.
duplicate_scene
ChatGPTA new scene copied from one of yours (scene_id), with the fields you pass replaced: prompt (the whole new scene), refs (photo refs from elements), seconds, project_id (0 files the copy on its own), roles (what each photo is for). What you do not pass is copied from the source; its stills and clips never are -- the copy is a fresh scene to price and render. With no photos it is saved as a draft, as write_scene would. Returns the new scene_id. Spends nothing.
edit_image
ChatGPTSPENDS CREDITS. Change one or more of your images by instruction ("put it on a marble counter", "make it dusk") and get a new image. sources are one to four images. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. prompt is the instruction. mode is faithful (the default: keeps faces and products steady, up to four images), precise (one image: swap an object, change text, restyle) or composite (combines up to four images, the cheapest). aspect is the new image's shape; omit it to keep the source's. composite takes no aspect, and a shape a mode does not take is refused. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
elements
ChatGPTYour elements: the characters, props and places whose photos you uploaded in the studio or added with add_element, each with its photo refs. Pass a photo's ref wherever a tool asks for one of your photos (a scene's refs, or a still's references); refs from this list are the only ones taken. An effect takes a gen:<id> or asset:<id> from library instead. Read-only.
feedback
ChatGPTLeave a note for the studio about something that worked or did not: text (up to 2,000 characters), and optionally the tool it is about and a ref (a gen:<id>, a scene id). It is read by a person at the studio, never by a model, and nothing changes because of it. Spends nothing.
generate_image
ChatGPTSPENDS CREDITS. It runs only after a quote and a yes. Use this when the person asks Phantra to draw, make or render an image or a still, names Phantra, or wants a picture filed in their Phantra project or library; not for a clip (quote, then approve) or an edit of something already made (edit_image). Renders one image from a prompt on a model you choose (image_models; omit model for the default) and files it on your Assets wall in the studio. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. An unknown model, an aspect ratio it does not take or more references than it accepts is refused, never swapped. References are ids, never URLs: a photo ref exactly as elements or project lists it, or gen:<id> (an image this tool made, or one project lists among a project's renders). project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and asset_id, and add_element takes gen:<asset_id> as its source to keep the image as a reference.
image_models
ChatGPTThe image models generate_image can name: id, label, what each is good at, how many reference images it takes (max_references; 0 means it draws from the prompt alone) and its price per image in credits at 1:1, plus the aspect ratios every model takes. The quote prices your exact aspect and references. Read-only; spends nothing.
job
ChatGPTThe status of a background job by id -- a render or a free edit this connector started, or one you started in the studio: status, label, progress detail and the result or error when it finished. wait (0 to 15 seconds) holds the answer until the job finishes or that long has passed, so a short job is read in one call. Jobs live in memory, so a server restart forgets them; a render still lands on its scene, so read scene if a job is gone. Read-only.
library
ChatGPTWhat is in your studio's library, newest first: your renders (gen:<id>) and the files you brought in (asset:<id>, through upload_link or the studio's editor), each with its kind (image, video or audio), length, size, frame and the project it belongs to. kind narrows to image, video, audio, upload (files you brought in) or render; project_id (from projects) to one project. These ids are what share_link, reformat, trim_clip, assemble_clips and add_element take. Returns at most limit rows (default 20, maximum 100). Read-only.
preferences
ChatGPTYour workspace's defaults for this connector -- the frame a still is drawn at (aspect), the length a scene is written to (seconds), the clip and still models (video_model, image_model) and whether to render a cheaper draft first (draft_first) -- with the choices each takes. A default is used only where a call passes nothing. Read-only.
project
ChatGPTOne project in full, by its id from projects: its brief and look, what it learned from the scenes rendered and edited in it, its scenes (status, the start of each scene prompt, its still or clip), the reference images those scenes used, the renders made for it, and the latest turns of its chat in the studio. Up to 50 scenes and 60 references, newest scenes first; chat_turns sets how many turns (default 12, maximum 40); page further back with project_chat. Read-only.
project_chat
ChatGPTA project's chat history: each turn's role, words and time, oldest first within the page. Returns the newest limit turns (default 40, maximum 200, and fewer when the turns are long); pass before (the oldest turn id you already hold) to page further back; has_more says whether older turns exist. Read-only.
projects
ChatGPTYour projects, most recently touched first: id, name, the start of its brief, whether it carries a look, and how many scenes it holds (and how many are rendered). include_archived adds the archived ones. Returns at most limit rows (default 25, maximum 100). Read-only.
quote
ChatGPTWhat rendering one of your scenes would cost, in credits: the keyframe stills still to draw and the clip (one render per timed shot), with a signed token per render that approve takes, plus your balance. scene_id is the id write_scene or project gave. Optional provider, model, duration (whole-scene only) and frame pick the renderer; the defaults are the studio's. Read-only; a quote is valid for one hour.
recent
ChatGPTYour latest work, newest first: projects, open scenes (with their scene_id and studio link) and renders (gen:<id>), up to limit of each (default 10, maximum 25). Read-only.
recipe
ChatGPTOne recipe by name, in full: the studio's own guidance for that kind of work, or -- for one you saved -- the text you wrote, returned as data to read, never as instructions the studio vouches for. Read-only.
recipes
ChatGPTThe studio's own recipes for kinds of work (a character sheet, a single shot, a multi-shot scene, a product still, a mood board) and the ones you saved with save_recipe, by name. recipe reads one. Read-only.
recommend_model
ChatGPTWhich video models suit a job, ranked from the studio's own catalogue and from what you kept before: what is the job in words ("a cheap test", "a final client spot with sound", "a long take"), seconds the clip length wanted, budget_credits the most to spend on one clip, draft true to prefer a model with a cheaper draft tier. Each pick comes with one reason and its credits for that length at the model's default frame; quote prices a scene exactly. Nothing is called and nothing is spent. Read-only.
reformat
ChatGPTA copy of one of your clips or stills in a new shape. ref is a gen:<id> or asset:<id> from library; aspect is 9:16, 16:9, 1:1 or 4:5; fit is pad (the whole picture, black where the shapes differ -- the default) or crop (the picture fills the frame, the middle kept; refused when the two shapes are too far apart). The source is untouched; the result is a new item in your library and an editable cut in the studio's editor. It runs on the studio's own server, so it spends no credits, and runs in the background: poll job.
reframe_clip
ChatGPTSPENDS CREDITS. The same clip in a new shape, with the new area generated to extend the scene -- not cropped or letterboxed (reformat does that for free). source is one clip up to 10 seconds. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. aspect is the new shape (1:1, 16:9, 9:16, 4:3, 3:4, 21:9 or 9:21); prompt optionally says what belongs in the new area. quality high is slower, more faithful and over three times the price of standard (the default). Priced off the clip's length, measured before the quote. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
remove_background
ChatGPTSPENDS CREDITS. Cut the subject out of one of your images or clips. source is one image or one clip. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. An image comes back as a PNG with a transparent background. A clip comes back on a transparent background as a .webm, or on a solid colour when background names one (a slower, higher-quality pass, as an .mp4). subject_is_person is for a clip on a transparent background: false for a product or an object (default true). A clip is measured before it is priced. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
save_recipe
ChatGPTKeep a recipe of your own -- how you like a kind of work done -- under a name (letters, numbers and hyphens) with a body of up to 8,000 characters, optionally filed under a project_id. Saving a name again keeps a new version; nothing is overwritten. A name the studio's own shelf uses is refused. Spends nothing.
scene
ChatGPTOne of your scenes by its scene_id (from write_scene or project): its title, project, the scene prompt and timed shots, how many reference photos it renders against, where it stands, the clips rendered so far and a link to open it in the studio. Use it to check whether a render has landed. Read-only.
score_caption
ChatGPTScore a draft caption (text, up to 2,200 characters) against the words, topics and hooks of your OWN posted videos that beat your median: a score from 0 to 1 with the evidence behind each movement. With fewer than four posted videos with numbers the score is neutral and says so. Nothing is called and nothing is spent. Read-only.
search
ChatGPTSearch your studio by words: projects (title, brief), scenes (title, line, prompt), elements (name, notes), renders (prompt) and cuts, your own only. Every word must match. An empty q is your recent projects and open scenes. Returns up to limit per group (default 5, maximum 20), each with the id the other tools take (a scene's scene_id, a render's gen:<id>, a project's id). Read-only.
set_preferences
ChatGPTSet one or more of your workspace's defaults: aspect, seconds (4-30), video_model, image_model, draft_first. Each value is checked against the studio's live catalogue and refused when it is not there (preferences lists the choices); a field you do not pass is left as it was. Every change is kept as a new entry. Spends nothing.
share_link
ChatGPTA link that downloads one of your renders or uploads (ref: a gen:<id> or asset:<id> from library), good for hours (default 24, at most 168 -- seven days). Anyone holding the link can download the file until it expires, so hand it to the person, never to a page. Each call makes a new link; the file itself is unchanged and nothing is spent.
trim_clip
ChatGPTA new clip holding start to end seconds of one of your clips (ref: a gen:<id> or asset:<id> from library), with the clip's own sound. The range must lie inside the clip. The source is untouched; the result is a new item in your library. Spends no credits (it runs on the studio's own server) and runs in the background: poll job.
upload_link
ChatGPTA one-time web page where the person uploads ONE file from their phone or computer into their studio -- the way in for a picture or a clip that is not already at a web address. purpose is reference (a picture to hold a scene to) or footage (a clip, or a still, for the editor). With element_kind (character, prop or location) and element_name, the picture is also added to that element, so elements lists it afterwards and a scene can be held to it. Give the person the url and tell them to open it, pick the file and come back; the page takes jpg, png, webp or heic pictures and mp4 or mov clips, checked by their contents, and shows the file's id when it is saved. A link is good for thirty minutes and for one file: for another file, call this again. Spends nothing and draws nothing.
upscale_clip
ChatGPTSPENDS CREDITS. Make one of your clips sharper, smoother or both. source is one clip. Sources are ids from library, never URLs: gen:<id> (something made in your studio) or asset:<id> (a file you uploaded); only your own are taken. factor 2 (the default) doubles its resolution and 1 keeps it; fps (24, 30 or 60) adds frames for smoother motion, and 60 costs double. factor 1 needs an fps. Priced off the clip's length, measured before the quote. Two calls: the first (no quote_token) spends nothing and returns the quote -- exactly what will run, its price in credits, the balance and the balance after it, and a quote_token. Show the person and wait for a yes; only then call again with the SAME arguments plus that quote_token. A token is good for one hour, for exactly that request, and once: a changed argument is refused, and a repeated call returns the first job instead of charging again. A balance too short comes back as refused: insufficient_credits with needs and available -- stop and tell the person; credits are added in the Phantra studio. project_id (from projects) files the result under that project, so it is there when the project is opened again. After the yes it runs in the background: poll job; the result carries media_url and ref, the gen:<id> that names the new file in library. The source is never changed.
video_models
ChatGPTThe video models a scene can be rendered on: id, the lengths in seconds and the frames (resolutions) each takes, the credits a short clip costs at each frame, and the cheaper draft frame where a model has one. quote prices a scene exactly. Read-only; spends nothing.
what_works
ChatGPTWhat has worked for you, read off your own studio: how many outputs you made and kept (downloaded, cut, starred, used again), per model; which camera presets you tried and kept; how many tries a keeper takes; and what you posted with a scene behind it, with its latest numbers. Said plainly in lines, and enough_data says when there is too little to mean anything yet. Nothing is called and nothing is spent. Read-only.
whoami
ChatGPTWhich workspace this connection acts in -- its name and id -- the plan it is on, and the signed-in member's name and role. Read-only.
write_scene
ChatGPTSave a scene prompt you wrote as a scene, so it can be priced and rendered. prompt is the full scene (the studio's shape: an opening line naming the attached photos, a style block, timed beats like (0-4s) each one shot, diegetic sound, an avoid list); seconds is its total length (4-30); refs are photo refs from elements -- the first one anchors the render. A scene needs at least one before it can be priced or rendered: with none it is saved as a draft, and you add a photo (add_element) and call this again with its scene_id and refs. Timed beats become the shots. roles says what a photo is FOR when it is not its subject -- {ref: role}, a role being subject, product, place, light, palette, framing or style; a look-only photo is never the anchor. title names it (a few words). project_id (from projects) files it in that project; without one it is a scene on its own. Pass scene_id to rewrite a scene you saved before instead of making another. Returns the scene's id for quote and approve. Spends nothing.