com.spriteship/spriteship

FREE. Abort a running job. Failed/incomplete provider work is refunded automatically. If the job is waiting for directional-pose approval, its successful reusable 360 and pose charges are retained; no dependent animation has started. Use this when a generation is clearly going wrong rather than letting it run to completion.

PAID (image-to-video; Wan 3.0 costs 150 credits/second, so each default 2-second pair is 300 credits; total cost scales with duration × pair count). Animate ANY asset type — the animation lands on the source asset and its still image is preserved. For grid assets (tileset/staged) pass pairs of tile indices; a self-loop { from: n, to: n } makes one tile loop seamlessly. ASK THE USER WHICH STAGES FIRST. Call get_asset, describe the tiles you can see, and get an explicit answer on which is the START and which is the END before you spend. "Animate the treasure box" does not tell you whether they mean closed→open, open→closed, or the lid alone, and guessing wrong bills them for a clip they cannot use. The server rejects a grid animate with no pairs (400) precisely so that this choice is always made deliberately — do not satisfy it by defaulting to 0→last. Confirmed background-removal provider failures already fall back automatically on the server, without a new user charge. There is no daily removal retry lockout; honor reported cooldowns and unknown-task safeguards if recovery still fails. For already-authorized generation, automatically execute a zero-credit recovery quote and poll it; no renewed approval or refund. Never duplicate a running recovery. Limit automatic retries to two per failed job per session; stop on recurring failure or any paid quote. DEFAULTS TO A COST PREVIEW — see the dryRun argument. Returns { jobId } — poll get_job.

Successful automatic poses now continue without review, including imports. Four-direction setup stays automatic and eight-direction keeps its existing selection. Imported generation preserves source rendering rather than project style. Use this tool only when get_job returns a pending automaticPoseFallback. If poseReview is present, this is a historical imported-pose checkpoint from an earlier flow: show originalUrl and poseUrl side by side and ask the user to check style, complete head/body and unwanted ground/scenery. Submit choice="generated" only after they accept that exact pose, or choice="original" to use their original character; include poseVersion from poseReview.version. Use cancel_job to cancel. Without poseReview, pose creation failed: ask whether to continue with the default pose or cancel, and use choice="original" to continue. Never approve automatically. Both choices resume the original approved animation request at normal video cost. Accepted generated poses are reused without another review until regenerated.

FREE approval checkpoint. Use only when get_job returns status:"waiting_approval", wizardPhase:"waiting-directional-pose-approval", and directionalPoseReview. Present every returned previewUrls to the user and let them adjust or review the proposed frameIndices and poseScales. If any remain unreviewed, get explicit confirmation to use those candidates as shown before submitting the complete returned index and scale maps. When directions is present, review only those four poses, choose unique indices 0–3, keep scale 100 and preserve unused map entries. Absent directions means eight poses. Closing or declining means do not call this tool: the job remains paused and no dependent animation credits are spent. The server rechecks the live credit balance immediately before resuming paid dependent animations.

FREE. Commit the exact draftHash as a recoverable revision. The existing map must still match its expectedRevision; conflicts require re-reading and merging the user’s changes. Reuse requestId for retry. The map immediately opens in the normal editor.

FREE and SYNCHRONOUS, no AI. Copies a legacy repeat-xy texture into a new immutable Brushable Material revision; the original revision is preserved. Topdown/isometric conversions can be ready immediately when the legacy seam check passed. Platformer conversions are marked needs-review because the old texture has no paired surface ribbon. Legacy strip-only textures cannot be converted because they have no interior fill. Returns the new iteration and quality status; call get_asset afterwards.

FREE, no AI. Promote a pose into a real 1-frame animation (a still/idle) that is scaled to match the character's other animations and exports to every engine. Pass poseId "__idle__" to use the character's default image when it has no poses. Returns a jobId — poll get_job.

PAID (one image call). Creates a NEW asset from a prompt in any of the six types. The default dry run now returns a strictly text-free artwork brief: no captions, labels, lettering, digits or pseudo-writing unless the user explicitly requests specific intrinsic writing on an object. Environment Kit overlays must not have die-cut sticker borders or backing. It returns both the exact semantic plan (`plan.preview`, item count/list, camera, project-derived look) and `planToken`. SHOW that preview to the user. A real call requires the same request plus that planToken; changing the prompt, project settings, type, model, or output settings invalidates it and requires a new dry run. Standard models cost 60 credits at 1K, 100 at 2K, or 150 at 4K per image/sheet, including automatic background removal and fallback; use the dry-run quote. Object (`assetType: "static"`, formerly Static Asset) defaults to four separately editable design alternatives of the requested object, not a state sequence. All four share project style and one of five world cameras; explicitly requested flat inventory/UI icons use front/content_type. Explicit one remains supported. Clean cutouts omit incidental scenery and sticker backing, not real outlines or integral bases. Never automatically reroute asset types. Object States (`staged`) is one object in exactly four ordered states on a 2×2 sheet; no filler or automatic animation. Preserve identity, shared physical scale and real size changes. Object Collection (`tileset`) and Environment Kit (`terrain`) propose 1–25 useful pieces when no count is specified; explicit lists/counts are preserved without grid-filler extras. Object Collections contain independently editable mixed items, variants, matching families, icons/symbols or cards/tokens—not automatic animation stages. World props follow project camera; flat icons/cards use front-facing presentation even in isometric projects, unless the user explicitly requests another view. Preserve requested intrinsic symbols/text, never external captions. Material uses one transparent 1K donor sheet through the configured Material provider route and yields two platformer fill+surface families or four topdown/isometric fills, and background to one image when the user did not specify a count. Explicit counts/semantic sets override defaults, and non-square counts for other grid types are filled to the smallest supported 2×2–5×5 square with disclosed supplemental items. Environment Kit camera follows the project; request a different camera explicitly in the prompt. Kit pieces are placed over Materials, not used as deterministic terrain donors or guaranteed seamless brushes. projectId is REQUIRED — the project supplies gameType, which decides how the art is drawn, and keeps the asset findable in the web app. A fan-out create returns one jobId — the asset ids appear in get_job as gridAssetIds once the job finishes. Environment Kit has NO per-tile edit path: revising a terrain asset regenerates the whole sheet. DEFAULTS TO A PLAN + COST PREVIEW — see dryRun. Returns { jobId, assetId } — poll get_job, then call get_asset. Generated atlases are cut by logical foreground objects, not trusted row geometry; get_asset exposes extraction status/warnings.

PAID (standard 1K: 60 credits; 2K: 100; 4K: 150; premium models may differ). Generate a NEW character from a text prompt. This is the entry point for building art from nothing — call create_project first if the user has no project. Artwork is strictly text-free by default: no labels, captions, lettering or pseudo-text; only specific intrinsic writing explicitly requested by the user is permitted. IMAGE ONLY: it produces the character still, not animations. Animation is a separate explicit step (generate_character_animation for a walk cycle, generate_character_turn for the 360 rotation), because animating costs several times more and the user should choose it. BY DEFAULT it requests FOUR DESIGN VARIATIONS FOR THE SAME PRICE, matching the dashboard. Set generateFourVariations:false only when the user explicitly wants one character. gridRows:2 + gridCols:2 remain the legacy trigger for the same four-character mode; one image call is split by complete character silhouettes, never fixed cells. Four is the expected count, but any result with at least two separate silhouettes succeeds; every clearly separated full character is preserved whether two, three, or more than four are found. Zero or one means the sheet was not extractable and refunds. Show the user every result and let them pick. Returns { jobId, expectedCharacterCount } — poll get_job; on status:"done" its characterIds and characterPreviews tell you which characters exist and what they look like. gameType defaults from the project but can be overridden for this character. Use topdown for the Top-Down Angled character pipeline, including characters intended for isometric projects. Passing isometric is accepted as an alias and creates a topdown character because that is the measured working directional pipeline. The STORED CHARACTER gameType — not the project gameType — controls every later animation and whether a user-facing 360 turn is available. A topdown character inside a platformer project therefore keeps the top-down directional pipeline and supports generate_character_turn. DEFAULTS TO A COST PREVIEW — see the dryRun argument.

FREE and SYNCHRONOUS (no jobId — do not poll). Creates the container everything else needs: characters and assets are created INSIDE a project, and the project's gameType is what decides how they are drawn and animated. Also seeds a starter level map, so the game preview has something real to render immediately. Call list_projects first — reuse an existing project rather than making a near-duplicate.

Create a free static composition of existing icon/frame revisions; indicators and functional widgets remain independent. No new image generation.

Create an empty project-owned UI Pack with its guide snapshot. Free; generation is a separate quoted action.

FREE, synchronous, no AI. Make an exact copy of the CURRENT saved animation under a new slug, preserving its selected frame order, loop/hold settings, compositing, mounts, and rendered pixels. The original is untouched. Returns { jobId: null, animation }; do not poll a job.

Engine-ready map JSON built from the SAVED level data, preserving placement name/gameplayId and baking custom-field values. Every engine retains materialRendering and materialRenderingResolved when present. Bundles include material-rendering.json and bake styled terrain/areas into PNGs: never apply shading twice or replace baked art with raw donor tiles. Pixel materials require nearest filtering without mipmaps; keep semantic terrain for gameplay. GameMaker uses the supplied tizo_draw_tile Draw helper; Tiled consumers configure filtering from metadata. All engine JSON exports retain materialAreas; bundles also retain semantic outlines in material-areas.json, alongside clipped artwork. Never render both as duplicate visuals. Phaser JSON materialAreaSource needs mask/blend rendering in URL mode, not the raw donor rectangle. For the zip bundle (per-tile PNGs, loader scripts, tileset files) use get_export_command instead.

LEGACY ONLY; new Brushable Materials reject this operation because they use world-space sampling. PAID (60 credits per standard-model edit attempt; use the dry-run quote). Make an old texture tile seamless. Legacy textures were created as raw grid cells, so CHECK get_asset first: metadata.seamless.seamMetric.pass is a FREE verdict on whether this cell already tiles cleanly, and metadata.seamless.fixed tells you if it was already fixed. Do not spend on a cell that passes. The fix lands as a new current revision. Confirmed background-removal provider failures already fall back automatically on the server, without a new user charge. There is no daily removal retry lockout; honor reported cooldowns and unknown-task safeguards if recovery still fails. For already-authorized generation, automatically execute a zero-credit recovery quote and poll it; no renewed approval or refund. Never duplicate a running recovery. Limit automatic retries to two per failed job per session; stop on recurring failure or any paid quote. DEFAULTS TO A COST PREVIEW — see the dryRun argument. Returns { jobId } — poll get_job.

FREE, and NO AI AT ALL — no video model, no credits, nothing to approve. The tiles of a tileset/staged asset ARE the frames: give the tile indices in playback order (repeats allowed) and they are composited into a real spritesheet animation on that asset, exportable to every engine. This is how you animate a coin flip, a flickering torch or a chest opening WITHOUT paying for image-to-video. Try this before animate_asset whenever the motion already exists across the tiles. Returns a jobId — poll get_job.

PAID, and the cost MULTIPLIES BY THE NUMBER OF ANIMATIONS (each is its own image-to-video call; Wan 3.0 costs 150 credits/second, so its default 2-second clip is 300 credits). All frames are text-free by default: no subtitles, captions or invented writing. This is real motion — a walk cycle, an attack, an idle bob — not the 1-frame still that create_animation_from_pose produces. Read the character's gameType and use the standard animation names for it (get_character shows what already exists); the project gameType is irrelevant after character creation. A topdown character inside a platformer project still uses top-down prompts and directional behavior. A name that already exists is NEVER replaced: the server allocates the next numeric slug (walk → walk_2 → walk_3), and the returned animations list contains the resolved names. Ask the user which animations they want before spending: 4 animations cost 4x. For a topdown character, resolve the animation direction mode BEFORE choosing those animation names. Read animationDirection from get_character and preselect its Luna-classified mode; let the user override it. Pass animationDirectionMode on both the dry run and confirmed request: overhead uses one engine-rotated sprite with base action names and no 360; four_direction uses S/N/E/W rows with FREE four-pose image setup; eight_direction also uses diagonals and keeps the regular 360 video setup (+200 credits when newly needed). Overhead setup is free; animation videos are priced separately. Setup and Luna run only after animation generation is confirmed; the server creates the poses and automatically assigns matching facings without pose approval for four_direction, then continues animation generation. Poll all operation jobs while queued or running; an approved directionalPoseReview is a completion record, not a user checkpoint. The chosen mode is saved on the character. A missing classification is not permission to guess silently. Every animation defaults to 2 seconds, including walks, runs, attacks, and loops. Omit animDurations and customAnims[*].duration unless the user explicitly asks for a different clip length; never infer a longer duration from the motion or from the fact that it loops. If the character came from a successful 2x2 create, there may be two, three, four, or more siblings and animating is per-character: show the user the characterPreviews from get_job and have them pick one before you spend. For multi-direction sets, mirrorAnims gives you FREE flipped copies of a direction generated in the same call (walk_left from walk_right) — prefer a real generation when the character is asymmetric, because a held item switches sides under a mirror. The server uses the standard pose-aware animation pipeline for new requests; there is no flag argument to set. When the user wants stored poses as the actual video endpoints, call list_character_poses and pass their ids through poseFirstFrameIds / poseLastFrameIds. Omit those maps to let the server plan anchors automatically; never pass a pose URL. Before each new clip the server classifies the effective anchor as empty-handed, held, or unclear and adds the matching inventory-preservation lock automatically; there is no inventory argument to send. For platformer walk/run, the server-created cached mid-walk anchor pose is FREE; explicit pose generation remains paid. For eight-direction and legacy video-based topdown/isometric clips with targetFacings, automatic planning uses one cached standing internal 360 rotation for every action, including custom actions; its one-time surcharge is included in the dry-run quote. Do NOT call generate_character_turn merely to prepare an animation — that explicit visible turn is a separate action. When that internal 360 is needed, the job pauses at waiting_approval before any animation video starts. Present directionalPoseReview from get_job and call approve_directional_poses only after the user accepts the full set, including explicit confirmation to use any candidates they did not review individually. The dry-run response includes animationPipelineAssignmentToken; copy it verbatim into the confirmed call so the quoted pipeline decision cannot change while the user is deciding. If it expires, run the same dry-run again. The same response includes preflight with per-animation resolved names, durations and credits; setup and total credits; model resolution; runtime frame size; and explicit unknown-until-generated frame count/FPS/ETA. Show that structured preview instead of inventing precision. API-key calls can use the account's full wallet balance; the dashboard-only 600-credit limit and its first-animation exception do not apply. The independent key spend cap, paid-attempt ceiling, and ordinary wallet balance still apply. Returns { jobId, animations, operations, rejectedOperations? }; jobId is the first operation only for backward compatibility. Poll, retry, or cancel every operations[].jobId independently. When an operation reaches done, it returns an animationQualityGate requiring get_character; finish that review/repair loop before exporting or saying the animation is ready. A prompt that composes too long for the model is rejected with PROMPT_TOO_LONG before anything is charged, and the response names the animation. Check isTemplate on that entry: false means YOUR text is what is long, so shorten it; true means the shipped template is over the ceiling and no prompt you send can fix it — report it rather than retrying. DEFAULTS TO A COST PREVIEW — see the dryRun argument.

PAID (200 credits — ONE 2-second rotation video), except retrying the exact failed job is FREE while its provider task/result or validated stored source survives. The default dry run reports 0 credits and recovery:"stored_artifact" in that case; confirmation revives the same job and never starts another paid provider task. Spins the character through a full 360 so the sprite exists from every side; its frames are also what the directional-pose extractor reads, which is what makes top-down characters usable in more than one facing (isometric projects use top-down characters). ONLY topdown and isometric characters are supported — any other game type errors 400 TURN_NOT_SUPPORTED (topdown_overhead: the engine rotates the sprite image; platformer/point_and_click: side/front views, mirror instead). Eligibility reads the STORED CHARACTER gameType, not its project: a topdown character inside a platformer project is supported. This explicit visible turn is separate from the internal mid-walk rotation that generate_character_animation may plan for eligible directional walk/run clips; do not call it as mandatory setup for every animation. Lands as the animation "turn", or "turn_<seedPoseId>" when seeded from a pose. An existing turn is NEVER replaced: an occupied name becomes turn_2, turn_3, and so on. It 409s (ENTITY_BUSY) while ANY other job is running on the character, because it rebuilds the spritesheet: poll that job first. Returns { jobId, animation, resumed } — poll get_job. A successful standalone 360 finishes at done after its animation is rebuilt; it does not open the directional-pose approval checkpoint. Only an internal 360 planned by generate_character_animation pauses for directionalPoseReview before dependent animation videos start. If a retained provider URL is repeatedly proven expired, support may set discardPendingResult:true and preview again; that explicit escape abandons the stale result and is a new paid generation. DEFAULTS TO A COST PREVIEW — see the dryRun argument.

PAID (standard 1K: 60 credits; 2K: 100; 4K: 150; premium models may differ). Generate a new pose for a character from a text description. DEFAULTS TO A COST PREVIEW — see the dryRun argument. Returns { poseId } once executed; poses have no jobId, so poll list_character_poses until status is done or error.

PAID but very cheap and SYNCHRONOUS. Use this for a broad game vision, story, or mood before create_project. Returns a proposed project name, game type, art style, world description, creative guidance, and a validated styleGuide object ready to pass unchanged to create_project. It stores nothing and does not create assets. Afterward, inspect the game repository and propose an MVP art plan before any image or animation spend. Skip this tool when the user already supplied a complete project/style guide or asked for one specific asset. DEFAULTS TO A COST PREVIEW — see the dryRun argument.

PAID but very cheap (one configured text-model call) and SYNCHRONOUS — the result comes back in this response, there is no jobId to poll. Turns a rough game idea into { aiVision, aiPromptGuidance, aiNegativeGuidance }: a written creative direction plus the positive/negative guidance to reuse in later prompts, which is what keeps a whole project visually consistent instead of every asset drifting. Stores NOTHING — feed the wording into create_project (description/style) and into your own prompts. DEFAULTS TO A COST PREVIEW — see the dryRun argument.

Get a cost preview for an exact component list and batch breakdown, then generate using the returned signed quote. Every paid raster/SVG batch is 100 credits; one pack action can contain several batches. Fonts are free. Defaults to dry-run; never alter the quote.

Free imports: import ONE still PNG or WebP as an Object or Character. Unlimited for every account, including free users, with zero credits and no subscription or allowance check. Before uploading, tell the user: transparency required, one clear complete subject, clean edges, consistent project style and camera, no text/watermark, up to 10 MB and 4096px per side; recommend 256px or more except intentional pixel art. No videos, GIFs or animated files; upload does not remove backgrounds or fix quality. Set confirmed only after this notice and approval. For local files, prefer POST /api/v1/asset-imports multipart file/projectId/name/kind/requestId rather than passing large base64 through model context. Persist requestId for safe network retries.

FREE, no AI. Flip every cached frame of ONE animation horizontally and rebuild it — the cheap way to get walk_left from walk_right. NOT idempotent: calling it twice flips the animation back to its original orientation, so call it once and check the result before calling again. Frame count and order are preserved, so any frame selection stays valid. To keep BOTH directions, call duplicate_animation first and mirror the copy. Returns { jobId, mirrored } (frames flipped) — poll get_job.

FREE, no AI. Carve ONE multi-state clip into separate game-phase animations by frame windows — the classic case is a jump becoming crouch/rise/fall/land, with fall looping while airborne and the rest playing once. Each phase becomes a real exportable animation; the source animation is untouched. Jump-family animations often carry a ready-made suggestion: get_character → animations[<anim>].phaseProposal — pass its phases through verbatim. Frame windows index the FULL capture. Returns { jobId, created } — poll get_job.

Compare the versions in assets/spriteship/spriteship.lock.json with current SpriteShip state in one call. Returns unchanged/changed/unavailable per character, asset, map, or UI pack; exact animation frame-count/FPS/collision, revision, runtime-intent, map-shape and UI component summaries when the previous lock entry includes the snapshot returned by an earlier call; and explicit fetch/download/review-removal actions. Existing v1 lock files without snapshots still detect version changes, then upgrade by saving each returned snapshot. Requests may contain 100 entries; honor Retry-After if the weighted sync budget is reached. Never delete local files for an unavailable entity without user confirmation.

FREE. Resolve owned artwork and validate a complete editable map without changing the live map. Preserve stable placement IDs on edits and pass expectedRevision from context. Returns immutable draft ID/hash and graph-only topology report. Inspect canonical overview and camera-sized region images before commit; no browser is required. Walkable surfaces/transitions are ground-view only, visualParts/reveal apply across game types. Graph reachability is not body-clearance or gameplay proof. Missing assets use existing quoted generation with user approval.

FREE. Apply up to 200 stable-ID operations against expectedRevision. Returns a validated draft and semantic diff without changing the live map. Preview/test then use commit_map_draft. Settings use null to remove optional values. Deleting referenced objects must repair their dependencies in the same batch.

FREE. Render the actual map composition using canonical terrain/export rendering. Omit viewport for an overview, or request world-pixel viewport and scale for camera-sized inspection (2048px output limit). Inspect this image and fix problems before saving; image output is sufficient for visual review without browser automation. Split-part clips and walking-surface draw order use canonical semantics. This static preview does not prove playability; test mechanics in the target engine.

FREE, no AI. Re-render one animation from its durable cached frames by passing animation. This is the repair/refresh tool: use it when a sheet looks stale or wrong after a series of edits, or when a previous rebuild failed. It never re-generates art, so it cannot change what the character looks like — only how the frames are packed and normalized. Omit animation only for an explicit full-character maintenance rebuild. Exact pose-owned scale references make the targeted path sibling-independent; legacy captures retain their compatibility sizing fallback. Image-only characters return NO_ANIMATIONS_TO_REBUILD without creating a job. A targeted rebuild returns an operationId/jobId — poll get_job.

Crop missing slots from a retained original PNG after visual review. No provider call or credit charge. Supply original-pixel rectangles and the current source hash; already delivered siblings remain untouched. Inspect validation after recovery.

PAID (Wan 3.0 costs 150 credits/second: 300 credits at the default 2-second settings). Re-run video generation for a SINGLE motion pair of an animated asset, leaving its other pairs alone. The new take lands as an additional iteration on that pair — earlier takes are preserved. DEFAULTS TO A COST PREVIEW — see the dryRun argument. Returns { jobId } — poll get_job.

FREE, no AI. Try a cleaner background removal on ONE animation's source video — the repair for a quality read showing deliveredIssues containing "matte" (green fringe, flickering holes, ragged edges). The cleaner version is kept ONLY if it measures better, so this can never make the animation worse; the motion itself is unchanged. One attempt per capture: if the same frames were already tried (matteRepair stamp), a repeat call is a cheap re-check + free re-render — regenerate the animation for a fresh take instead. Fails with SOURCE_VIDEO_EXPIRED when the source video aged past its 30-day retention. Returns { jobId } — poll get_job, then re-read the character's quality block: "matte" gone = the cleaner version shipped.

Rename an owned component for free. Supply its current name as expectedName to reject stale edits. Preserves IDs, artwork and historical revisions; re-sync updates its exported display name.

Save a locally edited PNG, revalidate behavior metadata and preserve the prior revision. Free. Maximum PNG 4 MB and 2048 pixels per axis; no SVG artwork editing.

FREE and SYNCHRONOUS (no jobId — do not poll), no AI. Rebuild a staged, tileset, or terrain collection's presentation layout: anchor chooses the shared fixed point and cellOffsets apply per-item nudges. Modern manifest revisions are recomposed from their independent item PNGs and are never re-sliced; legacy staged and tileset revisions retain the source-sheet re-cut path, where cellInset can remove separator bleed. For a manifest revision, fitItemIndices makes those items fill their shared cell as much as possible while preserving aspect ratio. Legacy terrain must be upgraded to independent items first. Omitted fields keep the revision's current values; send cellOffsets:[] or cellInset:0 to clear. Returns a grid summary; call get_asset afterwards for fresh presigned tile URLs.

FREE. Restore a saved revision as a new revision, preserving history. Pass the latest expectedRevision and reuse requestId on retry.

FREE. Resume a failed pose when list_character_poses reports removalRecoveryAvailable. Reuses the saved image; never generates or charges for another image. Respect the retry-after time on cooldown errors, then poll list_character_poses.

FREE. Resume background removal when a failed/uncertain illustrated batch reports removalRecoveryAvailable. Reuses its saved artwork and original billing identity. Respect cooldown errors; never create a replacement paid batch.

FREE when the original KIE task or a validated stored artifact can be recovered; otherwise PAID only for the sub-steps that still need to run. Re-runs the FAILED step of a job whose status is "error" — the prompt, model and every setting are already on the job, so this is the cheap way out of a failure instead of creating the whole thing again. Work that already succeeded is NOT redone or re-charged: a generation that failed after its image landed reuses the paid result. get_job reports retryMode, retryRecovery and retryCost, and the dryRun quote here is that same number. If get_job reports failureAction:"edit_input", do not use this tool; edit the prompt or references and create a new generation. A non-retryable phase answers 400 JOB_NOT_RETRYABLE — some late phases hold partially-committed state and deliberately have no blanket retry. Eligible overhead Material jobs use stored_artifact recovery to repair the original without new generation or background removal. A confirmed failed repair is not offered again by that processor; refresh get_job and honor server eligibility, even below the general two-attempt limit. No automatic refund is implied. Confirmed background-removal provider failures already fall back automatically on the server, without a new user charge. There is no daily removal retry lockout; honor reported cooldowns and unknown-task safeguards if recovery still fails. For already-authorized generation, automatically execute a zero-credit recovery quote and poll it; no renewed approval or refund. Never duplicate a running recovery. Limit automatic retries to two per failed job per session; stop on recurring failure or any paid quote. DEFAULTS TO A COST PREVIEW — see the dryRun argument. Returns { jobId } — poll get_job.

Quality Retry is the subscription compensation for AI art that invents details, ignores the request, or arrives unusable, letting the user try again without asking support for a credit refund. Costs ONE quality retry and ZERO credits; it is not a refund and preserves the original. Use get_subscription_allowances first and proceed only when features.retry is true; explain the purpose/cost and obtain approval before confirmed=true. This is distinct from retry_job (technical recovery) and ordinary paid revision. Do not use for arbitrary new ideas or imported still images. Persist requestId and reuse it after network errors; a returned allowance requires a new requestId.

PAID. Generate a new revision of an asset from feedback text. The previous revision is kept — revisions are additive, never destructive. Environment Kit (`terrain`) assets also accept the legacy templateId to pull a whole category pack (edges / grounds / props / slopes / textures / structures). Passing sourceItemIndex uses that extracted item as the visual source for a complete new related collection: the source stays untouched and every output is a separate new asset listed in get_job.assetIds. The default dry run returns the exact semantic revision plan and planToken; show the preview to the user, then resend the unchanged request with that token. DEFAULTS TO A PLAN + COST PREVIEW — see dryRun. Returns { jobId } — poll get_job.

Get a cost preview for a 100-credit selected PNG revision, then pass its signed quote to execute. SVG artwork editing is Phase 2; it is not enabled by this tool. Siblings and prior revisions remain available.

FREE bounded browser-free route/clearance or deterministic controller replay. Choose exactly one saved mapId or prepared draftId. Returns map hash, declared body/profile, trace, events, diagnostics and limitations. A passed movement test is not proof of combat, scripts or destination-engine gameplay. Save reusable scenarios in document.playtests; inspect canonical images separately.

Adopt a new immutable guide revision with optimistic concurrency. Colors, typography and camera apply to future generation, never silently rewrite prior assets.

Save supported fill, text and layout values for an exact current revision. Free and deterministic; cannot manufacture unsupported capabilities or edit arbitrary SVG geometry.

Save a pack-specific style for future quotes using the current pack updatedAt. Free; preserves project defaults and existing component revisions. A stale timestamp is rejected.

FREE, no AI. Curate which frames an animation keeps, by index into the full capture, in order. Pass an empty array to reset to all frames. Non-destructive: the full capture is retained, so you can re-select at any time. Use get_frames with format:"sheet" first to choose indices. Returns a jobId — poll get_job.

FREE and SYNCHRONOUS (no jobId — do not poll). Author the PHYSICS shape a game should build its collider from for an asset revision: rect, circle, polygon, compound of 1..16 simple disconnected bodies, or none, in NORMALIZED coordinates (0..1 fractions of the sprite/tile cell, origin top-left; a circle radius is a fraction of WIDTH). Set tileIndex to give one tile of a tileset/staged asset its own shape — it overrides the asset-level default for that tile. Pass collisionBody: null to clear/inherit, or {kind:"none"} to explicitly disable collision including bbox fallback. Compound children use the same cell space; preserve open archways between them. This is what stops a game from falling back to the sprite bounding box. The shape flows into every map and asset export made AFTER this edit — Tiled per-tile objectgroups, the Tiled map Collision layer, and the Phaser/Godot/Unity/GameMaker colliders — and it is resolved from the asset at export time, so it applies to tiles that were already placed on a map before you set it. Distinct from the FOOTPRINT, which is only the level editor's snap/tessellation shape and has no physics meaning.

FREE, synchronous metadata edit for an owned revision or collection tileIndex. Each independent role accepts a normalized simple body, disconnected compound, or none. Coordinates refer to the full sprite or padded collection cell; radius is a fraction of width. Omitted roles stay unchanged, null clears an override, none disables. Tile roles override asset roles independently. hitboxFrames is not supported on assets. Legacy collisionBody remains the movement/hurtbox fallback; hitbox requires explicit configuration. No paid job.

FREE and SYNCHRONOUS (no jobId — do not poll). Declare how an asset revision should be rendered at runtime: plain_image, particle_emitter, deformable_strip, repeat_fill, nine_slice or shader_effect. path_follow is accepted only for backward compatibility; do not set it on new assets. Draw and configure motion per placement in the level editor instead. The intent and its config flow into every engine export made AFTER this edit. Set tileIndex to target one tile of a tileset/staged asset.

FREE, synchronous metadata edit. Independent normalized full-frame movement, hurtbox and hitbox bodies. Character movement permits a simple body or none; hurtbox/hitbox also permit compound. Omitted keys stay unchanged; null clears that scoped override; none explicitly disables a role. Scope defaults to character base; select one poseId or animation. Only animation scope accepts hitboxFrames (inclusive zero-based delivered indices, maximum 64 windows). Omitted windows mean the explicitly configured hitbox is active for the whole animation; [] disables it for all frames. A hurtbox never inherits the new movement shape. Legacy collisionBody fallback remains supported. No paid job or implicit combat/faction rules.

FREE. Queue a durable complete Phaser, Godot or Unity project export. First review with get_project_export_review and pass its version and selected starting map/player. Returns a job ID immediately; poll get_project_export_status until ready. No generation credits are spent.

PAID but very cheap (one configured text-model call) and SYNCHRONOUS — returns { hint, cost } inline, no jobId. Reads a STAGED asset's prompt plus the stage pairs you intend to animate and writes the one-sentence motion description to pass as animate_asset's motionHint. Use it when you are about to spend 300 credits on a default video and are unsure what to write: a few credits here protects that spend. Object States assets only — any other type answers 400.

Suggest palette, treatment and existing font IDs from the available catalog. Free to customers; suggestions require explicit save to adopt.

Revalidate an exact current revision and apply only proven deterministic runtime layout repairs, including bar fill openings, panel text/slice regions, and explicitly named SVG button-state separation. Free; preserves original sources. This is not arbitrary SVG artwork editing and cannot promise a successful capability result.

One asset with its revision iterations and typed animation projection. Accepts the asset id OR its slug. New Material assets use the wire-compatible `texture` type and expose a typed `material` block containing the versioned donor recipe and its platformer, topdown, or isometric projection. Legacy textures may instead carry metadata.seamless.seamMetric and support fix_asset_seams. Generation Contract assets expose `generationPlan` plus `extraction`. `extraction.status:"exact"` means the planned number of logical foreground objects was recovered; `"recovered"` means small detached decorations were safely grouped; `"needs_review"` preserves every detected object and carries a warning instead of silently deleting or inventing one. `grid.tiles[].url` returns each extracted object as its own transparent PNG; the grid is a rebuilt presentation layout, not the geometry used to cut the model output. Legacy assets without `extraction` retain their historical fixed-grid metadata and should still be reviewed with the old grid fields. Backgrounds carry a `background` block echoing the sub-type, view and the aspect ratio actually generated — read it back to confirm the shape when you left aspectRatio unset.

The full PhaserCharacter JSON: animations, frameSize, gameType, animationDirection, per-group atlas + spritesheet URLs, poses and mount points. For topdown characters, animationDirection is Luna's saved mode suggestion or the user's last override; preselect it before asking for overhead, four-direction, or eight-direction animation generation. The top-level animationQualityGate is the delivery verdict per animation: ready:true means its deliveredIssues list is empty; needs_review means follow its actions before claiming the animation is ready; unscored means inspect the frames because no machine score can certify it. Check collision automatically: while an animation plays use animations[activeName].collisionBody before the character-level collisionBody; while displaying a pose use pose.collisionBody first. Fall back to a bbox only when the resolved body is absent. Bodies are normalized to the FULL untrimmed frameSize — never subtract atlas spriteSourceSize/alpha-trim offsets. Every binary URL is presigned and expires in ~1h — download immediately, never store or commit one. The in-app preview exercises shared export data, but validate the downloaded bundle and its supplied loader in the target engine before shipping.

The self-describing export contract (tizo-export.json manifest + its JSON Schema) without downloading the zip. Use this to learn the Phaser integration shape and field meanings before wiring an export into a game project. Its files[] is a logical canonical inventory, not an exact archive listing; use get_export_command and read the downloaded zip's tizo-export.json when exact shipped contents matter.

Current credit balance, and optionally the recent ledger. Every ledger entry carries apiKeyId/apiKeyName so you can see exactly what THIS key spent. Check the balance before proposing any paid generation. When the balance will not cover what the user wants, say what IS affordable rather than only that they are short — "1200 for all three animations, you have 900, so two now or top up for all three" is the useful answer. API-key calls can use the account's full wallet balance; the dashboard-only 600-credit experiment and its activation exception do not apply. freeCreditLimit therefore reports applies:false with exemptionReason:"api_key". Two remaining walls need OPPOSITE advice: running out of BALANCE is fixed by buying credits (the 402 carries a topUpUrl — send that link, never quote prices, they change), while hitting the key's DAILY SPEND CAP is not a money problem at all — the credits are already in the account, and the fixes are raising the cap or waiting for resetsAt. Never suggest a purchase for a cap trip. If the user asks HOW to raise the cap: the 403 carries capSettingsUrl — give them that link and tell them to edit the key there. YOU cannot change it, by design: the cap edit needs a logged-in session and is unreachable with an API key, because a key that could raise its own cap would not be a cap. Do not retry hoping it passes, and do not ask them to paste credentials.

Typed custom-field definitions for a character or asset, plus the derived builtins ($animation for characters, $autoplay/$loop for animated assets). Per-placement VALUES ride on get_map; map exports bake resolved values and ship fields-schema.json. Read-only here — definitions are edited in the dashboard.

Build the authenticated download URL + a ready-to-run curl line for a zip export (character engine bundle, asset bundle, or map bundle). Run the curl in your own shell — zips are multi-MB binaries and must not be passed through the model. Phaser character exports may select size 512, 256, 128, or 64; the selected sheet/atlas pairs keep canonical size-agnostic filenames. Sparrow XML is character-only: preserve the bundle's Sparrow prefixes and multisparrow assetPath routing. This tool exports whole characters, not a dashboard single-animation selection. Requires SPRITESHIP_API_KEY to be set in that shell.

Presigned per-frame thumbnails for a character animation or an asset motion pair — never inline base64. Use format:"sheet" to get ONE index-labeled contact sheet, which lets you judge a whole capture in a single look; that is the intended way to pick which frames to keep before calling select_frames.

Free advisory version check. Call once when starting SpriteShip work in a session, passing the version actually read from the installed skill; omit it when unknown. Report an available update with its release summary and manual update guide once, then continue compatible work. Unknown is not outdated; never downgrade a newer local copy. Does not install files, grant permissions, spend credits or require an update.

Poll a background job. Every write tool that returns a jobId is polled here. On status:"error" the response is enriched with failureKind, failureAction, attempt-scoped creditsRefunded, retryCost, retryMode, and retryRecovery. If failureAction is edit_input, change the prompt or reference inputs and create a new generation instead of calling retry. Poll every few seconds — generations routinely take minutes. THIS IS HOW YOU LEARN WHAT A CREATE CALL MADE: characterIds / assetIds list the entities this job produced — ONE normally, FOUR when create_character ran with the 2x2 variation grid — and on status:"done" characterPreviews carries a presigned image per character so you can show the user the options and let them pick. IMPORTANT FOR CHARACTER ANIMATION JOBS: status:"done" means processing finished, not that the animation is ready. The response carries animationQualityGate.status:"check_required" with the affected animationNames; you MUST call get_character and resolve each affected animation until its gate is ready (or report that it needs review). Do not export, sync, or claim completion before this check. queued:true means it has not started yet (waiting for a slot) and has spent nothing. On any TERMINAL status (done or error) the response carries a `credits` block — { charged, refunded, net, balance, spendCapDaily, spentLast24h, capRemaining }. TELL THE USER `credits.net` AND `credits.balance` when the work finishes, in one short line. Report `net`, never your own pre-flight estimate: the two differ routinely — a 2x2 grid is ONE charge for four entities and failed steps are refunded — so quoting the estimate can report a cost that never happened. Do NOT attach a purchase suggestion to a successful generation; only raise buying credits when a call is actually blocked or the balance will not cover what the user asked for next.

A map row plus its saved level data, revision and dependency-sensitive contentVersion for plan_sync. Preserve the returned contentVersion in the map lock entry after a successful import. Every placement has an immutable internal id and may have a user-authored mapData.materialRendering v1 snapshots the art style with optional profile, edgeMode and pixelSize overrides. Preserve it; it affects materials only, never geometry/physics. Hard edges suppress feathering without deleting saved blend widths. Absent settings retain original appearance. materialArea v1 visual fill: preserve its closed shape, material revision, repeatSize and blendWidthPx (inherits layer materialAreaBlendPx, default 16). Areas imply no swimming, damage or collision. Terrain layer can contain both materialGrid cells and materialArea objects: draw cells first, then shapes in array order. Honor materialArea.hidden. Terrain stays above Background and below object layers. Each placement can have a name plus map-unique gameplayId for game-code references; layers, Navigation paths, and per-placement custom-field values are included. Terrain materialGrid cells include elevation and continuous ramps; appearance.elevationShading (0–1, default 0.35) controls slope lighting, opaque cliff shading, and height-edge cues together, not collision. Zero disables all elevation lighting. Cell dimensions are authoritative; editor grid guides are not saved. Returns mapData:null when the map has never been saved.

Read a prepared map draft, resolved document, hash and expiry.

One project with its description, registered mount slots, and saved style guide. The legacy assetSuggestions field is currently always null because automatic suggestions are disabled; do not poll for it.

Return an authenticated shell download command for a ready project export job. This tool checks status first and never starts a build. Run the command outside the model; do not transmit ZIP bytes through MCP. Supports Phaser, Godot and Unity.

FREE. Check all saved maps, character and asset dependencies, choose the starting map/player, and inspect target-runtime compatibility before downloading an editable complete Phaser, Godot or Unity project. Errors block export; custom game code remains your responsibility.

FREE. Read a queued export job. Poll queued/running jobs about every two seconds. Ready jobs include the artifact size and expiry; failed jobs explain what to fix before retrying. Use get_project_export_command only once ready.

Read the current guide; old projects without an adopted guide receive an initial suggestion. Existing assets retain their saved guide versions.

Read the currently enabled subscriber benefits and their separate monthly limits, caps and balances. Disabled benefits return zeroed values and cannot be used. Unused units accumulate up to three monthly grants while the feature is enabled. These are not generation credits.

Read components, immutable revisions, capabilities, fonts and batch progress. Artwork URLs expire; native engine capability gaps are not successful tests.

Read one paid batch and retained source URL; provider completion does not prove usable component delivery.

Read original PNG dimensions, expiring image URL, source hash, and exact delivered/missing slots for an illustrated batch needing review. Inspect artwork before choosing crops.

Return an authenticated curl command for a whole/selected UI Pack ZIP or component source. Run locally; never put multi-MB ZIP data into model context. Phaser has browser behavior coverage; Godot/Unity native acceptance is pending; GameMaker/Unreal deliver artwork, Tiled transports metadata. Recovery artworkOnly must be explicit.

Assets (Object / Object Collection / staged / background / Environment Kit / Terrain Materials) with slug, hasAnimation, updatedAt, tags and favourite. Wire types remain tileset for collections, terrain for kits and texture for brushable materials. Object was formerly called Static Asset; its API type remains `static`. hasAnimation is data-derived: ANY asset type can carry an animation, so never infer it from the type.

All poses on a character: { id, name, prompt, status pending|done|error, url, direction, pendingSince, errorCode, creditsRefunded }. This is the poll target after generate_pose — pose generation has no job id. It is also the source of ready pose ids for generate_character_animation poseFirstFrameIds / poseLastFrameIds. A pose still pending long after pendingSince is stalled, not working.

Characters with the re-sync signals: contentVersion (opaque change token — compare for equality, never parse or order), hasActiveJob, apiReady, pose/animation counts, tags, favourite. Persist contentVersion per character and re-fetch only when it changes.

List recoverable map revisions from editor and agent saves.

Level-editor maps with updatedAt and hasMapData. hasMapData:false means the map was never saved and cannot be exported.

All of the user's projects: { id, name, gameType, style, itemCount, createdAt, updatedAt }. gameType (platformer / topdown / topdown_overhead / isometric / point_and_click) is the default perspective for new content — read it before generating anything. A character may override that default at creation; afterward its stored character gameType controls its own animation and turn behavior.

Read the retained licensed Google Fonts catalog. Existing font selection, preview and export cost zero customer credits.

List owned UI Packs, optionally within a project.

Read project art, material recipes, limits, shape classes, current map JSON and revision. Inspect the returned artwork visually before selecting assets. Read your game code to establish movement/mechanics; SpriteShip fields are flexible metadata.

Read the authoritative JSON Schema before composing a map. Use assetRef instead of inventing image URLs. Shapes, custom properties and paths retain their existing editor meanings.

Read owned project maps, reusable detached assemblies, destination spawn IDs, room links and broken-link diagnostics. No browser required.

Read saved camera, reference brief, gameplay, assemblies, links, tests and per-engine support. Runtime availability is distinct from playability evidence.

Verify the API key and report which powers it has. Returns { userId, username, keyId, keyName, scopes, mapLimits }. mapLimits contains maxMapsPerProject (null means unlimited), maxAreaPx (width × height allowance per map), maxWidthPx and maxHeightPx (independent technical side limits), and canResizeMaps. Paid plans allow any aspect ratio within both area and side limits. Admin accounts receive Studio map capacity even without a subscription. Read effective limits and the saved map before recommending gameplay-sized dimensions; starter dimensions are not an account limit. Free is fixed at 720 × 480; saved larger maps remain editable/exportable, but resizing requires a subscription. Map creation and editing use the web editor. Call this first when a tool fails with a scope error — the scopes array tells you what this key can and cannot do.