Reel-E

Adds photos to a draft: photos attached in the chat (where the app supports it), the photos the listing link import found (useImportedPhotos), files uploaded with request_photo_upload (uploadIds, up to 12), and/or https image links (urls, up to 12 per call). JPG, PNG, WebP or HEIC, up to 25 MB each. Sending the same photo again adds nothing twice, so repeated calls continue until notYetAdded is 0. A listing needs at least 3 photos; the account's photo limit decides how many appear in the video, and the rest stay in the listing gallery.

Add photos the person attached (attachment storageIds) to the listing and, by default, to the video at a 1-based position (default: the end).

Put library photos (listing photos not in the video) into the video, at a 1-based position (default: the end). The plan's photo limit applies.

Add a badge like "Just listed" or "Open house". open_house, price_reduced and a few others take one short line (e.g. "Sat 1-4 PM").

Add a text overlay for a time range (seconds, 0.1s steps, at least 1s). At most 8.

Starts the song at the point that makes the video's first cut land exactly on the drop (checked with the Studio's own timing, within one frame); later cuts stay on the beat grid. It uses the drop read_music_structure reports, by trimming the song's start (the person's own uploaded or generated tracks only). Uses no credit; undoable. A track still being made or analysed reports that, and the alignment can be done once the track is ready.

A one-tap VFX on one shot: dusk (day to twilight), staging_fill (an empty room furnishes itself; interiors only), seasons (season_from current or a season, season_to summer, fall, winter or spring), shadows (a sun time lapse), stars (a starry night), sign_drop (3D letters drop onto the lawn: sign_text is a preset such as For Sale, Just Listed, Open House or Sold, or up to 2 words and 16 characters; default For Sale), helicopter_reveal (a helicopter lifts a cover off the home), sketch_to_real (a pencil sketch becomes the photo), house_drop (a construction time lapse) or people (a couple placed in the scene). The finished clip replaces the shot's clip, with Undo, after a few minutes. suggest_direction picks a VFX and shot for a wish; browse_options vfx lists them all. Uses 1 AI+ credit. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Puts the recorded narration into the video, or takes it out (enabled false). The person normally does this with Apply in the Voiceover panel after hearing it; this tool covers a request in words ('turn the narration on', 'put it in', 'take it out'). While a voice is still being recorded it is queued, and the recording goes in as soon as it finishes. Captions are separate (set_captions).

The product guide: every option the Studio has for one area, in the Studio's own words, generated from the same lists the editor uses (title designs, text, stickers, camera and timing, transitions, music styles and trimming, voices, photo edits and staging, branding, formats, plans) plus how_to_choose, which turns a wish like 'more modern', 'luxury feel' or 'calmer' into real settings. Read it before offering or describing options, and describe options only in its words.

Where the latest revision notes stand: each note's fix, whether it was made, and how the re-check judged it (fixed, better, not fixed). A note counts as fixed only when this says so.

Runs an action that is waiting for confirmation (it uses a credit, or publishes). It needs the credits (or publish) permission on this connection; without it the owner confirms the action in Keylight.

Create the private seller report page (visit counts only, no names) and its link. It replaces any earlier seller link, which stops working. Needs the person's yes. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Lays out the exact words the person typed or dictated ('say: Welcome to 12 Oak Lane...') over the shots as the listing's draft script, in the Voiceover panel, without spending anything. The person's words go in play order, one part per shot or pair of shots, as written; a part with more words than its shots leave time for is shortened to fit (a part over more shots, with toShot, keeps more words), and wording that could break fair-housing rules is replaced. The reply lists the exact saved words of every part and says which changed, so the person can read them before recording. A script written for the person comes from request_voiceover_script, the script writer that watches the video.

Changes the words of one part of the draft script (section: its 1-based number in the Voiceover panel, or its id), the same save as the Voiceover panel's section editor. Uses no credit; undoable. A recorded voice keeps the old words until that part is recorded again (rerecord_narration_section).

Turns on the hosted MLS video link: a public, unbranded 16:9 page for MLS video fields. Needs the person's yes. get_mls_kit shows the kit first. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Looks at shots with vision and works out a frame that keeps a subject in view for Wide, Square or Phone. A subject can be given ('the pool and the front door'), or left empty to frame each shot around its main subject. Returns a ready frame for each shot, which frame_shot accepts as frames; a shot where the subject can't be found, or that already shows it with the default framing, says so and gets no frame. Changes nothing. Suits any framing that depends on what is in the photo.

Chooses which part of a shot's photo each version shows: Wide (16:9), Square (1:1) or Phone (9:16). Uses no credit; undoable. A version's framing is its own: framing Phone leaves Wide and Square alone. There are four ways to frame, one per call: (a) focusX, focusY (0 to 1, where in the photo the frame centers; 0.5 is the middle) and zoom (1 to 3); (b) a nudge: zoomBy (zoom in is +0.25 for a bit), moveX / moveY (-0.1 moves the frame a little left or up, +0.1 right or down; the frame moves, so the picture appears to slide the other way); (c) frames: one {photoId, focusX, focusY, zoom} per shot, for when each shot needs its own frame (find_framing returns ready frames); or (d) reset true to go back to the default framing. aspect 'all' frames every version at once; a reset with 'all' clears the framing completely.

The state of a request that needed confirmation: pending (waiting for a yes), running, done or failed, with what happened.

The current edit as structured text: every shot in play order (photoId, room, what it shows, length, speed, movement, AI+), transitions, music (tempo, sections), opening title, text, stickers, branding, voiceover and credits left. Most edits start from the photoIds and positions it lists.

The listing website's analytics over a period: visits, visitors, video plays, how much of the video people watch, inquiries (leads) and top traffic sources. The full chart is in the listing's analytics in Reel-E. Owner visits are not counted.

Links for a listing: its page in Reel-E (watch and download the video there), the editor, and the public website once it is published (publish_website). Downloads: request_download.

Where a draft or a listing is: collecting, awaiting_owner, declined, processing, ready or needs_attention. Pass draftId (it reports the listing once started) or listingId. waitSeconds (up to 25) holds the answer until something changes. A video usually takes a few minutes.

Whether the hosted MLS video link is on and ready, the copy-ready MLS fields (video link, address, price, facts, description) and step-by-step posting guidance. It comes before enable_mls_link.

Everything the agent needs to update the home seller: the live links (website, video), where to download the video, and a ready-to-send message draft with the latest numbers. It sends nothing; the agent copies or sends the message themselves.

Every AI+ version saved on one shot (natural motion and VFX), newest first: its number, what made it (the prompt or the VFX), which one plays now, whether it fits the shot's current length, and how the person rated it. Switching to a saved version uses no credit (use_ai_version); a new one uses 1 AI+ credit (regenerate_ai_version).

The tracks generated for this listing, newest first, with each name, when it was made, tempo and which one is playing. Suits going back to an earlier generated track; set_music takes the id (or propose_music lets the person hear it first).

The listings this connection can edit, newest first, with their ids.

Existing tracks this listing can use (the royalty-free library and the person's own uploads and earlier generated tracks), with tempo (BPM) and length. It lists existing tracks only; a new original track is made with request_music_generation.

The songs a new listing video can use, with their musicId, tempo and energy.

Move one shot to a 1-based position in the video.

Shows one existing track (an id from list_music) for the person to play and approve in Reel-E's editor (set_music applies a track directly). Changes nothing until they approve. Several calls in one step offer several options, each with its own reason tied to this listing. Not for generating a new track.

Publish or update the listing website with this video. Needs the person's yes. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

The playing track's structure laid against the video's cuts: BPM and bar length, the sections (intro, verse, chorus, bridge, outro and so on) with song and video times and loudness, the drop (where the full sound returns), separately when the beat comes in, the part of the song the video plays, and every shot with its start, its length in bars, the bar its cut lands on and which transitions its gap allows. Read it before a music-aware request such as transitions on the beat or at the drop, faster paced or ending on the last chorus. Only name drop or section times read here. For more interesting transitions, a speed ramp or dissolve (set_transition) on one or two cuts at section changes or the drop is usually enough; AI Seamless needs request_ai_seamless and the person's yes.

The narration script as it stands, part by part: each part's shots and seconds, its exact words, how many words it holds against what its shots fit, whether a voice has recorded it (and whether the words changed since) and whether the narration is in the video. Read it before recording so the person sees the words. Changes nothing.

The latest edits on a listing (by the owner's agents), with changeIds for undo.

Makes a new AI+ version of one shot for 1 AI+ credit, once approved: a VFX repeats its effect and options; natural motion repeats its prompt unless a new one is passed. version is the saved version to repeat (a number from list_ai_versions; default: the one playing now). The new version lands on the shot, undoable, and the old ones stay saved. A credit-using result's one free redo (Make again free) is the person's own button under the version after a thumbs down; it is not a tool. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Take shots out of the video. They stay in the listing library and can come back. The video keeps at least 3 shots.

Remove a sticker.

Remove a text overlay.

Set the full play order. List every current shot exactly once.

Swap a shot for an attached photo (attachment storageId), in the same place with the same length and movement. The old photo stays in the library.

Generates cinematic AI motion for one shot (moving water, fire, people, trees). The camera on an AI+ shot is one gentle push-in, so there is no camera move to pick or describe; the prompt names what should move in the scene. Uses 1 AI+ credit once approved. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

AI Seamless is one continuous generated camera move from a shot into the shot right after it, steered by an optional prompt: walking through a doorway, turning from one view of a room to another (a 90 or 180 degree turn between two photos of the same space), flying from an aerial down to the front door. It joins two Standard shots next to each other anywhere from shot 2 to the second to last shot (the first cut and the last cut stay clean cuts), a shot can be in only one AI Seamless pair, and an AI+ shot or a Boomerang cannot be part of one. To join the opening photo, another shot has to open the video first. afterPhotoId is the first shot of the pair. Uses 1 AI+ credit once approved; a pair it cannot join is refused before anything runs, with the reason. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Where the person downloads the video in one format (16:9, 1:1 or 9:16): downloads are made in Reel-E, so this returns the listing page link (and the editor's) for the owner. Nothing renders from this call.

Generates a new original instrumental track for this listing, the same as Create in the Music panel. It suits a request for a new, custom, original or generated track, with the style that fits best (the person says yes before it runs) and a short direction: mood, instruments, pacing, tied to this listing, without an artist or song name. It takes a couple of minutes and counts as one of this listing's music generations. The finished track waits in the Music panel for the person to play and approve. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Edit one listing photo: how it looks or what is in it. enhance (brighter, sharper, better color), declutter, twilight (dusk or night look, lights on; interiors too), day (daytime, blue sky), stage (virtual staging in a style) or custom (describe the change in instructions: remove or add an object, swap a finish). Use it for the shot the person named or is on. The shot then shows the edited photo (kept in the photo's history) and its clip is remade from it by itself. Uses 1 photo edit once approved; on an AI+ shot the AI+ clip is remade too (1 AI+ credit), and the confirmation says so. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

For a photo file on the assistant's own machine: returns an uploadUrl. The file's bytes are PUT there with the returned headers within 10 minutes, and the uploadId then goes to add_listing_photos.

Records the current draft script with an AI voice (Miles, Maya, Grant or Nora). It follows request_voiceover_script or draft_narration, once the person has read the draft (read_narration_script shows its exact words) and chosen a voice. Uses 1 voice generation once approved, and the recording goes into the video only when they apply it. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Keylight's script writer for any narration or voiceover request ('add a voiceover', 'write narration', 'rewrite the script'), the same as Generate auto voiceover in the Voiceover panel: it watches the video and writes a script timed to the shots. Uses 1 script generation once approved (the reply says how many are left); the draft then appears in the Voiceover panel to read, edit, hear in any voice and record. For the person's own exact words, draft_narration is the tool. On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Record only one part of the narration again, keeping the other parts' audio, the same as re-recording one block in the Voiceover panel. Uses 1 voice generation once approved. The voice stays the recording's voice: to change the voice, record the whole script (request_voiceover_audio). On a Full control connection it starts at once; otherwise it returns a confirmationId and waits for the owner's yes.

Takes the person's notes that point at times in the video, like 'at 0:28 a weird blur with the plant and the blue chair' or 'at 0:33 remove the older gentleman', passed in their own words. Notes without a time work when the words name the shot ('drop the closet shot'). It finds the shot or transition at each time on the saved video (or the shot the words name), looks at that spot (the photo and the rendered clip), names the element and the cause, and returns one plan with the first fix for each note (tool and arguments) and the credit cost of any fix that uses one. Editing notes ('start with the kitchen', 'make the pool shot longer') come back as the edit itself (move_shot, remove_shots, set_shot_length). Free fixes can be made straight away; credit fixes are requested with the tools it names. In the Keylight editor it checks each spot again after the fix renders. Uses no credit.

Turn narration captions on or off. Needs a voiceover with audio.

The end contact card: on or off, the agent's details, and the line above the name (prefix: default presented_by, none removes the line).

The video's edit style, the same switch as the Music panel's: cinematic (the default, today's edit), flow (Flow: a push and an orbit each bar to the drop, then Boomerangs on the beat; AI+ is off in Flow and its clips come back in cinematic) or fast_cuts (Fast Cuts: a cut on every beat after the drop). Available when the edit summary has an EDIT STYLE line. A style writes each shot's length, Movement and transition (editable after) and sets the song start so the drop lands on a cut; Fast Cuts sets AI Seamless aside, Flow sets AI+ aside and the AI Seamless pairs that touch its Boomerangs or ramps. Uses no credit; undoable.

Turns the video's fades on or off: videoIn (fade in from black at the start), videoOut (fade out to black at the end), musicIn and musicOut (the music's fade in and out). Off leaves the video's length unchanged: the last frame or end card holds instead of going to black, and the music stops with the picture. Only the fades passed are changed. 'No fade at the end' means videoOut false (and musicOut false if the sound is meant too).

Use an uploaded headshot (attachment storageId) on the contact card, or null to remove it.

Optional choices before start_video: the song (a musicId from list_songs; default is Reel-E's first library song), branding ("saved" uses the account's saved brand kit, "none" has no contact card or logo; default is the saved kit when there is one), videoPhotoCount (how many photos the video uses, up to the account's photo limit), pace (cinematic, fast or flow), and the address or listing facts. Each choice is saved on its own: one that cannot be used is listed with the reason and the others still save.

Turn the logo end card on or off, or use an uploaded logo (attachment storageId).

Standard camera movement for shots. An AI+ shot plays one gentle push-in and has no move to choose, so it is skipped; if every named shot is AI+ the tool refuses. It does not remove AI+ from a shot.

Turns the music off (musicId null), or switches to a track the person has already heard and approved, including one of this listing's generated tracks (list_generated_music). A new track goes through propose_music so the person can hear it first. Cuts re-sync to the new beat, and a track the person trimmed earlier keeps its trim. Uses no credit; undoable.

How loud the music plays under the narration, as a percent (0 silent to 100 full), the same slider as the Voiceover panel. It changes the mix only while narration plays; without narration the music plays at full level. Uses no credit; undoable. Keylight has no fade-in or fade-out control for the music volume.

Makes the edit faster paced, slower and more cinematic, or back to normal, as one undoable change. By default (amount a_little) every other shot takes a half-length or double-length beat-aligned step; amount a_lot steps every shot, for much faster or much slower. A shot whose length or speed was already chosen keeps it unless named. The first and last shots and AI+ shots keep their fixed length. The video stays on the song's beat grid; read_music_structure shows where the cuts now land.

How long shots stay on screen. Cuts land on the music's beats, so length is shorter (half), normal or longer (double). The first and last shots have set lengths: a shorter first shot while the opening title is on, or a shorter ending, comes back as a choice for the person (remove the title or an end card, or keep it).

Playback speed of the camera movement: slow (0.5x), normal or fast (2x).

The opening title card over the first shots: whether it shows, its DESIGN (the same designs as the Text panel's design buttons; browse_options title_designs shows what each looks like), WHICH WORDS it shows (the address, or the listing name, like the Text panel's Address / Listing name switch), which listing facts it shows (price, bedrooms, bathrooms, square feet) and its background. The words come from Property details, as in the Studio: for a title like 'Just Listed in Austin', update_property_details sets the listing name and wording listing_name here shows it. The address itself is changed in Property details. The result says exactly what visibly changed, or why nothing did.

The transition after a shot: cut, speed_ramp or dissolve. Not available before the last shot or next to an AI+ shot. After the first shot it is available only while the opening title is off (the title plays over that whole shot); with the title on, it comes back as a choice for the person.

Starts a NEW listing video. Takes the address (or a name for a listing with no public address) and, when the person has one, their public listing page link (Zillow, Redfin or a brokerage page), whose photos and facts Reel-E imports. An idempotencyKey, reused on retry, returns the same draft every time. Next come add_listing_photos, optionally set_listing_basics, then start_video. Photos come from the person: attached in the chat where the app supports it, imported from the listing link, or added at addPhotosUrl from the result, a Reel-E page where they drop photos from their computer or phone; get_listing_status then shows the count.

Makes the video for a draft. It creates the listing and uses one of the account's listing credits, exactly like making it in the Reel-E app. A connection that may create listings starts it now and gets the listingId; any other connection gets an approveUrl for the account owner, who approves it in Reel-E, and get_listing_status shows progress. Calling it again does not make a second listing.

Takes a wish about how the video feels or looks rather than a named edit: slow, boring, flat, more energy, like TikTok or Reels, too fast, too busy, calmer, more luxury, make it pop, or a look missing from the photos (dusk, a night shot, a season, a Just Listed sign, a reveal). It reads the edit style, the song's BPM, energy and drop, how often the video cuts, which shots are exteriors and the credits left, and returns a plan: what can be done now at no credit (undoable), what needs the person's yes (a VFX or photo edit with its cost, or library tracks with their BPM to hear first), or a short question to ask. Uses no credit.

Chooses which part of the song plays: startSeconds and endSeconds in SONG time (read_music_structure lists the sections in song time), at least 0.5 seconds apart and with at least two downbeats between them; either one alone moves only that edge and keeps the other where it is. For 'end on the last chorus', endAtSection takes a section label such as chorus, with an optional occurrence (1 = first; default the last one): the song then ends where that section ends, keeping its current start. Suits 'start the music at the chorus', 'shorten the song' and 'end on the last chorus'. Only the person's own uploaded or generated tracks can be trimmed, as in the Music panel; a library track plays from its start, so its drop can't be moved onto a cut either (a generated track can be). Uses no credit; undoable. Cuts re-sync to the trimmed song.

Undo one change by its changeId. Only the parts nobody has edited since go back.

Changes the listing's facts in Property details: price, bedrooms, bathrooms, square feet, listing status (active, coming soon, pending, sold), the listing description or the listing name (what the opening title shows when its wording is listing_name, for example 'Just Listed in Austin'; it is also the listing's own name in Reel-E, so the editor header, the listings list and download file names change with it). It updates the listing itself, the same as the Property details dialog, and the website and the opening title read it. On a listing whose website is live it needs the publish permission (Full control), since the site shows the change at once; without it nothing changes. It puts no text on the video; add_text and set_title do that. Suits 'change the price to 1.2 million', 'it's 4 beds' or 'mark it pending'. Uses no credit; undoable. Addresses are changed in Property details in Keylight. When a number could mean two things (for example 'six-bedroom bathroom count'), the person clarifies first.

Change an existing text overlay (textId from the edit summary).

Add a listing photo, logo or headshot to the account from an https link (JPG, PNG, WebP or HEIC, up to 25 MB) or base64 (up to about 3 MB of file). Returns a storageId for add_photos, replace_photo, set_logo or set_headshot.

Switches an AI+ shot to one of its saved versions (a number from list_ai_versions). Uses no credit; undoable. A version made for a shorter spot can't fill a longer shot; a shot without AI+ gets it with request_ai_plus, which reuses a matching saved version without a new credit.

Looks at the actual shots with vision and answers a look-and-feel question: pacing, mood, what a shot shows, which shot is strongest. Slower than the summary, so it suits questions the summary cannot answer.