lupa_check_connection
ChatGPTChecks that the Lupa connection works and returns the workspace it bills to and the granted scopes. Use it when the user asks to check the connection or which workspace is connected. Read-only. Generation tools do not need it first.
lupa_create_upload_url
ChatGPTReturns a short-lived presigned PUT URL for uploading one media file to Lupa storage, and the public URL the file will have after the upload. The Lupa upload panel uses it; it can also be used when the user attached a file in the chat and a code-execution environment is available: upload the attachment with curl -sS -X PUT -H "Content-Type: <mime>" --data-binary @<path> "<uploadUrl>", then use publicUrl as the image or video input. If the upload cannot be made from that environment, use lupa_open_upload_panel instead. This call stores nothing by itself.
lupa_crop_expand_image
ChatGPTCrops, expands (outpaints) or reframes an image and saves the result to the connected Lupa workspace — the image editor's Crop & Expand. The frame works as in the Lupa image editor's Crop & Expand, in the source image's pixels: aspect_ratio picks one of the editor's presets (21:9, 16:9, 3:2, 4:3, 1:1, 3:4, 2:3, 9:16) and grows the canvas to it without cropping; or set each side by hand with expand_top / expand_right / expand_bottom / expand_left (positive adds canvas on that side, negative crops into the image; -4096 to 4096). pan_x / pan_y move the image inside the frame (px; positive = right / down), scale resizes it (0.05–20, 1 = as is) and rotation turns it (degrees, clockwise, -180 to 180). If the image still covers the whole frame the result is a crop: free, finished at once. If any of the frame is left uncovered, Lupa fills it in (an expand) for a flat 20 credits. With a prompt, the new area is drawn from it; without one, it continues the picture. lupa_estimate_cost (kind "crop_expand") says whether a frame is a free crop or a paid expand before spending. To let the user set the frame by hand, open lupa_open_crop_expand instead. The result joins the source image's versions in the Lupa image editor. The call returns at once with the pending job; the Lupa card shows its progress and the image when it is ready, so no follow-up call is needed. A failed expand is refunded automatically. idempotency_key is optional: the server makes one when it is omitted; sending the same key again returns the same job without a second charge, so a retry with it is safe. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_estimate_cost
ChatGPTPreview what a job will cost in tokens before spending, plus whether the connected workspace has enough. Covers image generation (kind "image", the default), video generation (kind "video" — the price moves with model, duration, quality and sound), the 1080p finalize of a Seedance 2.5 draft (kind "video_finalize" — duration_seconds of the draft and has_reference_video), Video Presets (kind "video_preset" — model is the preset slug, quality 480p/720p/1080p, duration_seconds for ugc-tour), 3D objects (kind "object_3d" — tripo-h3-1 by texture none/2k/4k/8k and face_limit; rodin-v2-5 by texture none/2k/4k) and image upscales (kind "image_upscale" — precision, creative, flux and classic price off the OUTPUT size, so pass image_url or the source dimensions; the answer also reports what the output would measure; texture-enhancer quotes 0 when the run would be the user's Free Generation, and refuses what lupa_upscale_image would refuse), Remove Background (kind "remove_background" — a flat 5) and Crop & Expand (kind "crop_expand" — the same frame arguments as lupa_crop_expand_image plus image_url or original_width/original_height: 0 when the frame is a crop, 20 when it leaves canvas to fill). The number is the number the matching lupa_generate_ / lupa_upscale_ / edit call would charge, resolved with the same defaults. Read-only — spends nothing. For the balance on its own use lupa_get_credits, and for the model lists and their prices use lupa_list_models.
lupa_finalize_video_draft
ChatGPTRenders a finished seedance-2-5-draft video (made with lupa_generate_video) at 1080p, as the "Generate in 1080p" button does in Lupa. It keeps the draft's seed, aspect ratio, duration and audio; nothing else can change (a different prompt or inputs needs a new draft). Spends credits from the connected workspace at the finalize price the draft reported (draft.finalizeTokenCost; lupa_estimate_cost kind "video_finalize" also quotes it). One finalize per draft: calling it again returns the existing 1080p job without a new charge. Drafts can be finalized for 7 days. Returns the 1080p jobId immediately; it usually finishes in about a minute. Use it when the user asks for the 1080p version of a draft. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_generate_3d
ChatGPTTurn photos of an object — or a text description — into a textured 3D model (GLB), the same as Lupa's 3D Objects on the web. Spends credits from the connected workspace. Two models: tripo-h3-1 (default) — sharp geometry, 4K textures, comes back at real-world scale, works from photos OR a text prompt, quality low/medium/high (100/115/150 credits); rodin-v2-5 — best on tricky shapes (flat, thin or sculptural objects such as wall mirrors), photos only (no text-to-3D), one setting at 115 credits, and it reads prompt ALONGSIDE the photos (e.g. "a thin flat wall mirror, no stand"); tripo-h3-1 ignores a prompt when photos are given. Photos: front_image_url is the main view and is required when sending photos; tripo-h3-1 takes the front only, rodin-v2-5 also takes side_image_url and back_image_url (a true 90° side view, not a slightly turned front). Photos must be Lupa media URLs — call lupa_upload_media first for local files. Background removal is not needed. Returns a jobId immediately — generation takes 1–4 minutes. The card shows a 3D viewer whose Export saves GLB, USDZ, OBJ, STL or SKP (SketchUp). When the user asks YOU for the file later, call lupa_download with this jobId and the format they want, and give them the link. When no card is tracking it, poll with lupa_get_job (outputModelUrl carries the GLB). Failed jobs are automatically refunded. idempotency_key is optional: omit it and the server makes one; pass your own only to make a retry of the same request safe. Clients that support MCP Apps (Claude, ChatGPT) show the result in an interactive Lupa card: the image or video with Download, Upscale, Open in Lupa and Copy link. When the result says the user sees it in the card, the reply ends there — at most one short sentence, no embedded image, no URL, no restated model, size or cost: the card already shows all of it. When the result says the card is tracking a job that is still running, the card polls on its own and shows the result when it lands: if that job was the whole request, reply briefly and stop — do not wait for it or check on it. If the request has a next step on that result (upscale it, make angles of it, animate it, upscale the video), call that tool right away with the job id as its source_job_id / first_frame_job_id: the server waits for the result itself. Only without the card, embed each image as a markdown image () and give the same URL as a plain "full resolution" link (videos: a plain link). Result URLs are the user's own media on cdn.lupa.art and are safe to embed. When the user later asks you for the file or a download link, call lupa_download with the jobId and give them the link it returns.
lupa_generate_camera_angles
ChatGPTRenders an input image from a new camera angle by rotating, tilting or zooming a virtual camera around the subject, and saves the results to the connected Lupa workspace. Spends credits from the workspace: one angle (rotation/tilt/zoom) costs 15 credits, or all_angles=true renders 8 preset angles in one call for 45 credits. The call returns at once with the pending job; the Lupa card shows its progress and the image when it is ready, so no follow-up call is needed. Failed jobs are refunded automatically. idempotency_key is optional: the server makes one when it is omitted; sending the same key again returns the same job without a second charge, so a retry with it is safe. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_generate_image
ChatGPTGenerates 1–4 images from a text prompt, optionally with reference images or a style preset, and saves them to the connected Lupa workspace. Spends credits from the workspace (prices per model and resolution are in lupa_list_models; lupa_estimate_cost quotes a request). Model "aros" is text-to-image only and can feature an Aros character through character: the user's trained characters and the example "Kiara", listed by lupa_list_models (characters are trained in the web app). The call returns at once with the pending job; the Lupa card shows its progress and the image when it is ready, so no follow-up call is needed. A failed job is refunded automatically. idempotency_key is optional: the server makes one when it is omitted; sending the same key again returns the same job without a second charge, so a retry with it is safe. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_generate_video
ChatGPTGenerate a video with Lupa (minimax-h3-max, seedance-2-5, seedance-2-5-draft, happyhorse-1-1, gemini-omni-video, seedance-2, seedance-2-fast, seedance-2-mini, kling-3.0). Spends credits from the connected workspace; requires a paid plan (every model except kling-3.0 needs Pro+). Text-to-video, image-to-video (first_frame_url / last_frame_url or reference_image_urls) and reference video/audio where the model supports them. HappyHorse picks its endpoint from the inputs: reference images → reference-to-video, first_frame_url → image-to-video, neither → text-to-video (it takes no end frame and no reference video/audio). Gemini Omni takes up to 7 images total, or one reference video plus up to 5 images (only its first 10s are used, and the model then chooses the output length, so duration is ignored); a first_frame_url spends one of those image slots. MiniMax H3 Max finishes in seconds rather than minutes at the lowest per-second price, and takes the fewest inputs: a prompt plus optionally a first_frame_url and last_frame_url, and nothing else — no reference images, videos, audio or multi_shots. Its only qualities are 480p and 768p (it has no 720p), its duration floor is 5s, and with a first_frame_url the output takes that image's aspect ratio, so aspect_ratio is ignored. Seedance 2.5 (seedance-2-5) is the long-form Seedance: 4–30s in a single shot at 480p/720p/1080p, up to 30 reference_image_urls, 10 reference_video_urls (30s of reference video in total, each frame at most ~1284×722 — it is NOT downscaled here, so re-encode bigger clips before uploading) and 10 reference_audio_urls, with a 5,000-char prompt cap. On 2.5 a first_frame_url/last_frame_url cannot be combined with ANY reference (images, video or audio), and a frame task ignores aspect_ratio (the frame decides it). With a reference video and an editing-style prompt ("replace the sky…") the model may reclassify the job as video editing, in which case the output takes the clip's length and ratio; Lupa retries that automatically and the original charge stands. Seedance 2.5 Draft (seedance-2-5-draft) is Seedance 2.5 with every rule above, rendered as a cheap 480p draft (quality is always 480p, priced as 2.5 at 480p) to check framing and motion. If the user likes it, lupa_finalize_video turns THAT draft into the 1080p video (same seed, ratio, duration and audio) for the finalize price in the result (draft.finalizeTokenCost), until draft.finalizeExpiresAt (7 days). Seedance 2.5 Draft can't run prompts Lupa's moderation flags — those are refunded and need seedance-2-5. Kling supports multi_shots with per-shot prompts and uses the first 2 reference_image_urls as image_urls. Cost depends on model, quality, duration and whether a reference video is attached. This returns a jobId immediately — generation takes 1–5 minutes; when no card is tracking it, poll with lupa_get_job (its outputVideoUrl carries the result). Failed jobs are automatically refunded. If the user only has local files, call lupa_upload_media first (kind:'video' for reference videos) and use the returned URLs. idempotency_key is optional: omit it and the server makes one; pass your own only to make a retry of the same request safe. Clients that support MCP Apps (Claude, ChatGPT) show the result in an interactive Lupa card: the image or video with Download, Upscale, Open in Lupa and Copy link. When the result says the user sees it in the card, the reply ends there — at most one short sentence, no embedded image, no URL, no restated model, size or cost: the card already shows all of it. When the result says the card is tracking a job that is still running, the card polls on its own and shows the result when it lands: if that job was the whole request, reply briefly and stop — do not wait for it or check on it. If the request has a next step on that result (upscale it, make angles of it, animate it, upscale the video), call that tool right away with the job id as its source_job_id / first_frame_job_id: the server waits for the result itself. Only without the card, embed each image as a markdown image (![Generate…
lupa_get_card_image_jobs
ChatGPTReturns the image generation job(s) a Lupa result card is displaying while the generating call is still running (job id, status, start time, model and the size settings), looked up by that call's idempotency key. Called by the card only and hidden from the model; lupa_get_job checks a job.
lupa_get_credits
ChatGPTReturns the connected workspace's spendable Lupa balance, the same number the web app shows ("tokens" in the app, "credits" in these tools: the same unit). Use it when the user asks how many tokens or credits they have left. The result splits the balance into plan, token packs and granted credits, and reports separately any credits held by a failed renewal or a cancelled plan, which cannot be spent. Read-only. For the price of a specific job, use lupa_estimate_cost.
lupa_get_job
ChatGPTReturns the status of a Lupa job (image, video, upscale, camera angles or 3D) and, once it has completed, its output: the image inline plus its full-resolution URL, or the video or 3D model URL. Waits server-side for the job to finish (default 45s), so one call is normally enough. A failed job has already been refunded. Read-only. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_list_models
ChatGPTLists what Lupa can generate and what each option costs: the image models (resolutions, credits per image, reference-image limits, aspect ratios) for lupa_generate_image, the video models (durations, qualities, aspect ratios, credits per clip) for lupa_generate_video, the upscalers for lupa_upscale_image, the style presets, the Video Presets for lupa_video_preset_run (credits per quality), the 3D object models for lupa_generate_3d (Texture Quality, face limits, credits and plan requirements), and the Aros characters the user can feature (their trained characters and the example "Kiara"). Prices match what the generation tools charge (Texture Enhancer's Free Generation aside: lupa_estimate_cost says when a run would be free). Two upscalers are priced by output size and have no fixed price here; lupa_estimate_cost quotes them along with the output dimensions. Read-only. Generation tools have defaults, so this is not required before a generation the user has already described.
lupa_list_page_images
ChatGPTLists the images on a web page the user provides (a property listing, a portfolio, an article): their URLs in page order (1 = first), alt text and the page title. Read-only. Use it when the user gives a URL to take images from, for example "make a Space Motion video from this listing" or "use the 3rd and 5th photos from this page"; then choose the images the user asked for (all, specific positions or certain rooms; logos, maps, floor plans and repeated shots are usually not wanted) and pass them on (lupa_space_motion_quote_video for a video). Some sites block automated access: the result then says so, and the user can upload the photos instead (lupa_open_upload_panel). The tool does not bypass such blocks.
lupa_open_crop_expand
ChatGPTOpens the Lupa image editor's Crop & Expand on an image, in a card where the user sets the frame by hand: the aspect presets, dragging the frame's sides to crop or add canvas, moving, scaling and rotating the image, and an optional prompt for the new area. The card shows whether the frame is a free crop or a 20-credit expand and runs it (lupa_crop_expand_image) when the user presses Crop or Expand; the card then shows the result like a generation card, and you are told its job id. Opening it creates and spends nothing. Use it when the user wants to crop or expand an image without saying exactly how; when they did say (e.g. "make it 16:9"), call lupa_crop_expand_image directly. In clients without cards, call lupa_crop_expand_image instead.
lupa_open_upload_panel
ChatGPTOpens a Lupa upload panel where the user adds local image or video files to use in a generation, edit, camera-angle or upscale request. Use it when the request needs a file the user has not given as a public URL (for example "use this logo" or "upscale this video"). If the user attached the file in the chat and a code-execution environment is available, lupa_create_upload_url can upload that attachment instead; when that is not possible, open this panel and tell the user that chat attachments do not reach Lupa directly, so the file needs to be added here. kind: "video" for lupa_upscale_video sources (MP4/MOV, up to 50MB, 1s–10min, as in the web app), "image" for reference images. The user adds up to max files; when they press Continue, Lupa uploads them and returns their public cdn.lupa.art URLs, to use as image_input / image_url / video_url in the original request. Not needed when the user already has a public URL.
lupa_record_3d_export
ChatGPTAdds one export made in the Lupa 3D object card (format, texture size, outcome, file size) to the workspace's download history, as Export does in the web app. Called by the card only and hidden from the model. It does not produce a file; lupa_get_download_link returns a 3D file.
lupa_remove_background
ChatGPTRemoves the background of an image and saves the result, a transparent PNG, to the connected Lupa workspace — the image editor's Remove BG (Bria RMBG 2.0). Spends a flat 5 credits from the workspace. Usually takes a few seconds. The result joins the source image's versions in the Lupa image editor. The call returns at once with the pending job; the Lupa card shows its progress and the image when it is ready, so no follow-up call is needed. Failed jobs are refunded automatically. idempotency_key is optional: the server makes one when it is omitted; sending the same key again returns the same job without a second charge, so a retry with it is safe. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_space_motion_add_scenes
ChatGPTAdds photos as new Scenes at the end of an existing Space Motion project, free of charge. Close-ups start only with closeups=true, for when the user asked for them (5 credits per close-up). Same photo rules and project limit (20 images) as lupa_space_motion_create. Display: in clients that render MCP Apps, the result opens in a Space Motion card with the storyboard, every close-up, camera-move pickers, more close-ups, the Generate button with its price, live progress and the finished cuts with download and share. The card keeps itself up to date, so no polling is needed.
lupa_space_motion_card_add_scenes
ChatGPTAdds photos the user uploaded in the card as new Scenes at the end of a Space Motion project, as the card's "Add Next Scene" does. Free; close-ups are a separate step. Photos that cannot be animated are left out and listed. Called by the Space Motion card only and hidden from the model.
lupa_space_motion_card_create_project
ChatGPTCreates an empty Space Motion project with the name and aspect ratio entered in the Space Motion card's new-project form; the user then adds photos in the card. Free. Called by the card only and hidden from the model; lupa_space_motion_create creates a project from photos.
lupa_space_motion_card_generate_closeups
ChatGPTWrites and renders close-ups for one Scene, as the card's "Generate Close-ups" button does: fills the scene up to 4 close-ups within the project limit, 5 credits each (refunded if one fails). Called by the Space Motion card only and hidden from the model.
lupa_space_motion_card_render_video
ChatGPTRenders a Space Motion project into its finished video cuts when the user presses the card's Generate button, which shows the price: 80 credits per image in the storyboard. Takes about 10–20 minutes; a failed render is fully refunded. Called by the Space Motion card only and hidden from the model.
lupa_space_motion_create
ChatGPTCreates a Space Motion project from photos of a space. Space Motion turns photos of a space (interiors, architecture, real estate) into a short music-cut video reel: each photo is a Scene; Lupa writes camera close-ups of each Scene (up to 4, 5 credits each), every image becomes a 3-second camera move, and Generate renders them into 3 finished cuts, each edited to a different song. A project holds at most 20 images (scenes + close-ups) and needs at least 5 to generate. Generate costs 80 credits per image (lupa_space_motion_get quotes a project's exact price) and takes about 10–20 minutes. To have the finished video made from the photos rather than a project to edit, use lupa_space_motion_quote_video instead. When the user wants a new project but has not provided photos, lupa_space_motion_open_form opens the new-project form in the card, where they name it, pick the format and add their photos. Pass the photos in playback order (each becomes a Scene). Photos must be Lupa media URLs; for local files call lupa_open_upload_panel first. Photos that cannot be animated (under 300×300, over 20 MB, or a height-to-width ratio under 0.40 or over 2.50) are left out and listed. Creating the project and its scenes is free. Close-ups cost 5 credits each (up to 4 per scene) and are not started by default; set closeups=true only when the user asked for close-ups. Rendering the video is a separate step (lupa_space_motion_render_video). Display: in clients that render MCP Apps, the result opens in a Space Motion card with the storyboard, every close-up, camera-move pickers, more close-ups, the Generate button with its price, live progress and the finished cuts with download and share. The card keeps itself up to date, so no polling is needed.
lupa_space_motion_generate_closeups
ChatGPTWrites and renders more camera close-ups for Scenes of a Space Motion project, as "Generate Close-ups" does on the web. Fills each scene up to 4 close-ups within the project limit, at 5 credits each (refunded if one fails). Omit scene_ids to fill every scene with free slots. Display: in clients that render MCP Apps, the result opens in a Space Motion card with the storyboard, every close-up, camera-move pickers, more close-ups, the Generate button with its price, live progress and the finished cuts with download and share. The card keeps itself up to date, so no polling is needed.
lupa_space_motion_get
ChatGPTReturns a Space Motion project: its scenes, close-ups and their camera moves, the exact render price, render progress and, once finished, the video cut URLs. With wait_seconds it waits server-side while close-ups or the video are still rendering. Use it to open a project the user names (lupa_space_motion_list finds it). Read-only. Display: in clients that render MCP Apps, the result opens in a Space Motion card with the storyboard, every close-up, camera-move pickers, more close-ups, the Generate button with its price, live progress and the finished cuts with download and share. The card keeps itself up to date, so no polling is needed.
lupa_space_motion_list
ChatGPTLists the workspace's Space Motion projects, most recent first (id, name, status, aspect ratio). Read-only. In clients that render MCP Apps they appear as a project grid in the card, where the user can open one; otherwise lupa_space_motion_get opens a project.
lupa_space_motion_make_video
ChatGPTStarts a Space Motion video that lupa_space_motion_quote_video priced, after the user agreed to the quoted total. Creates the project, starts the close-ups (5 credits each) and renders the video when they finish, without exceeding confirmed_generate_cost (if the price would be higher, it stops and says why). Returns the project; lupa_space_motion_get shows its progress.
lupa_space_motion_open_form
ChatGPTOpens the new Space Motion project form in the Lupa card, as "Create new video" does on the web: project name, portrait or landscape, then the storyboard where the user adds their photos. Creates nothing and spends nothing. Use it when the user wants a new Space Motion video and has not provided photos (with photos, use lupa_space_motion_create). Pass name / aspect_ratio only if the user gave them. Display: in clients that render MCP Apps, the result opens in a Space Motion card with the storyboard, every close-up, camera-move pickers, more close-ups, the Generate button with its price, live progress and the finished cuts with download and share. The card keeps itself up to date, so no polling is needed.
lupa_space_motion_quote_video
ChatGPTPrepares a Space Motion video made automatically from a set of images, for when the user asks for the finished video (e.g. "make a Space Motion video from these photos / from this link") rather than a project to edit. Space Motion turns photos of a space (interiors, architecture, real estate) into a short music-cut video reel: each photo is a Scene; Lupa writes camera close-ups of each Scene (up to 4, 5 credits each), every image becomes a 3-second camera move, and Generate renders them into 3 finished cuts, each edited to a different song. A project holds at most 20 images (scenes + close-ups) and needs at least 5 to generate. Generate costs 80 credits per image (lupa_space_motion_get quotes a project's exact price) and takes about 10–20 minutes. Pass the images in playback order: public image URLs (e.g. from lupa_list_page_images) or Lupa media URLs. This step spends nothing: it checks the images (unusable ones are listed with the reason) and returns the plan (one scene per image, close-ups on every scene, then the video) with its total price. The card shows the plan and the total, and the user confirms there; after that the close-ups and the video run without further calls.
lupa_space_motion_refresh_project
ChatGPTReturns the current storyboard of a Space Motion project for the card, optionally waiting up to wait_seconds while close-ups or the video are still rendering. Called by the Space Motion card only and hidden from the model.
lupa_space_motion_remove_closeup
ChatGPTRemoves one close-up from a Space Motion storyboard, as the card's remove button does. The close-up can be brought back with lupa_space_motion_restore_closeup. No credits are refunded for a removed close-up. Called by the Space Motion card only and hidden from the model.
lupa_space_motion_render_video
ChatGPTRenders a Space Motion project into its finished video cuts. Spends 80 credits per image in the storyboard (scenes plus finished close-ups); lupa_space_motion_get quotes the exact price. Requires at least 5 images and no close-ups still rendering. Takes about 10–20 minutes; a failed render is fully refunded. Use it when the user asks to render or generate the video. Display: in clients that render MCP Apps, the result opens in a Space Motion card with the storyboard, every close-up, camera-move pickers, more close-ups, the Generate button with its price, live progress and the finished cuts with download and share. The card keeps itself up to date, so no polling is needed.
lupa_space_motion_restore_closeup
ChatGPTPuts a removed close-up back into its Space Motion storyboard (the card's Undo). Free. Called by the Space Motion card only and hidden from the model.
lupa_space_motion_set_closeup_camera_move
ChatGPTSets the camera move of one close-up in a Space Motion storyboard (null restores the default move). Free. Called by the Space Motion card only and hidden from the model.
lupa_space_motion_share_project
ChatGPTCreates the public share link of a finished Space Motion project, as the card's Share button does; anyone with the link can watch the video cuts. Called by the Space Motion card only and hidden from the model.
lupa_upscale_image
ChatGPTUpscales or enhances an image and saves the result to the connected Lupa workspace. Spends credits from the workspace. Models: "precision" (keeps the source detail; 4k/6k/8k), "texture-enhancer" (Texture Enhancer: boosts the realism and material texture of a render, to about 8K, for a flat 95 credits; creativity -10 to 10 and an optional prompt describing the materials; Pro and above, and every subscriber gets one Free Generation at 0 credits, once per user and shared with lupa.art, which below Pro is the only run allowed), "creative" (adds new detail; 4k/6k), "flux" and "classic" (scale_factor x2–x10), "realistic" (turns a 3D render into a photographic image), "real-skin" (skin retouch), "old-photos" (restores and colorizes old photos). Most upscales take 1–20 minutes. The call returns at once with the pending job; the Lupa card shows its progress and the image when it is ready, so no follow-up call is needed. Failed jobs are refunded automatically. idempotency_key is optional: the server makes one when it is omitted; sending the same key again returns the same job without a second charge, so a retry with it is safe. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_upscale_video
ChatGPTUpscales a short video to a higher resolution and/or raises its frame rate by interpolation, and saves the result to the connected Lupa workspace. Two processing types: "standard" (default; keeps the look of the source) and "pro" (reconstructs skin and fine texture, aimed at footage of real people; costs 1.2–4x more and takes longer). Spends credits from the workspace; the price depends on duration, fps, output resolution and processing type and is computed server-side. Requires a Pro, Premium or Business plan. The source must be an MP4 reachable by URL, 1s–10min long; for a local file, call lupa_open_upload_panel with kind "video" first. Returns a jobId immediately; video upscaling takes minutes. When no card is following it, check it with lupa_get_job (outputVideoUrl holds the result). A failed job is refunded automatically. idempotency_key is optional: the server makes one when it is omitted; sending the same key again returns the same job without a second charge, so a retry with it is safe. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_video_preset_run
ChatGPTVideo Presets are ready-made real-estate and interior videos, the same as the Presets tab on Lupa: day-to-night (1 photo → a 7s locked-off day-to-night timelapse with activity and sound), before-after (exactly 5 photos of the finished project → a 10s video that opens on the first photo under construction and cuts the rest to music) and ugc-tour (a presenter shows 1–4 photos of a property, speaking the chosen language, 4–30s). Each has a fixed price per quality (480p/720p/1080p; ugc-tour is priced per second). The aspect ratio follows the photos: portrait photos make a portrait video, landscape ones a landscape video. Runs one preset and saves the video to the connected workspace, spending its fixed price (lupa_estimate_cost kind "video_preset" quotes it when the user has not seen the price). Photos must be Lupa media URLs (for local files, call lupa_open_upload_panel first), in the order the video should follow (before-after: the first one becomes the construction intro). ugc-tour needs a host: host_slug for a stock host from lupa_video_presets_list, or host_image_url for the user's own presenter photo, which requires likeness_consent: true: ask the user to confirm they have the right to use that person's likeness before setting it. Lupa checks an uploaded presenter before charging and refuses minors, public figures and explicit images (nothing is charged then). Returns a jobId immediately; a preset takes about 5–8 minutes (several stages). Failed runs are refunded in full. Display: in clients that render MCP Apps, the result opens in a Lupa card that shows the image or video with Download, Upscale, Open in Lupa and Copy link, and that keeps following a job that is still running until it finishes; the result text says when the card is showing it. In other clients the result text carries the output URL on cdn.lupa.art (the user's own media), to show as an image or a link. To use this result in a follow-up step (upscale it, change its camera angle, animate it, upscale the video), pass its job id as source_job_id or first_frame_job_id; the server waits for that job to finish. lupa_get_download_link returns the file of a finished job.
lupa_video_presets_list
ChatGPTVideo Presets are ready-made real-estate and interior videos, the same as the Presets tab on Lupa: day-to-night (1 photo → a 7s locked-off day-to-night timelapse with activity and sound), before-after (exactly 5 photos of the finished project → a 10s video that opens on the first photo under construction and cuts the rest to music) and ugc-tour (a presenter shows 1–4 photos of a property, speaking the chosen language, 4–30s). Each has a fixed price per quality (480p/720p/1080p; ugc-tour is priced per second). The aspect ratio follows the photos: portrait photos make a portrait video, landscape ones a landscape video. Lists the presets available now with what each needs, its price per quality, and the stock UGC hosts (slug, name, photo) for lupa_video_preset_run. In clients that render MCP Apps, a preset picker card shows preview videos, prices and the stock hosts; a choice made there arrives as a chat message. Spends nothing.