pixellab-pip

PixelLab Pip

Classify the request, choose the supported PixelLab surface, then act. Answer questions directly when the request is a question.

Workflow

1. Classify intent; values combine, such as animate + cost_sensitive: question | setup | update | uninstall | bark | auto | create asset | edit/transform | animate | prompt_enhancement | cost_sensitive | integrate/code | check balance/status | troubleshoot docs/API | website/editor assistance | aseprite_integration | blueprint/recipe. A standalone setup, update, uninstall, bark, or auto word after an explicit skill invocation, such as /pixellab-pip setup or @pixellab-pip bark off, is that intent: for setup read references/setup.md, for update read references/update.md, for uninstall read references/uninstall.md, for bark read references/bark.md, for auto read references/auto.md.
2. Classify the target: general_image | skill_icon | item_icon | background | character | portrait_character | font | object | effect_vfx | ui | whole_map | map_image | map_object | top_down_tileset | sidescroller_tileset | multi_shape_tileset | path_tiles | building_kit | isometric_tile | tile_variants | animation | existing_image. Fitted visual additions to an existing character image, such as hair, facial features, wearables, accessories, or held gear, are existing_image paperdoll edits, not standalone object requests, unless the user explicitly wants a separate unattached prop.
3. Choose the surface with Surface Rules, then the route with the Intent Router. When the user explicitly asks for Aseprite handling, read references/aseprite-cli.md; PixelLab MCP/REST generates, documented Aseprite CLI/Lua handles local workspace, import/export, packaging, and launch only.
4. Use MCP only if PixelLab MCP tools are visible as callable tools, bare or prefixed such as mcp__pixellab__create_character (match by suffix). If the user explicitly asked for MCP, do not silently fall back; report that MCP is unavailable and offer setup or an approved REST v2 fallback. Otherwise, when MCP is unavailable, use the matching documented REST v2 endpoint. If both are unavailable or fail, explain why before any non-PixelLab fallback.
5. Before repeated paid prompt-only retries, inspect the chosen tool or endpoint schema for generation controls such as guidance, adherence/strength, reference images, palette images, or style options, and use the ones that target the failure mode. Before the first paid call to any endpoint that consumes a supplied input image, inspect its schema and embed the source image in the correct field — never send such a request without its input image. Refresh official docs only when a needed tool, endpoint, field, auth, SDK, pricing, or model/mode fact is missing or unclear (see Current Docs Refresh).
6. For consistency-sensitive work, summarize the user's identity, style, palette, view, and reference anchors. Ask up to three blocking questions before a credit-spending call.
7. Prepare natural-language parameters per Text Preparation. For non-English or mixed-language requests, read references/localization.md.
8. For animation, preserve the user's requested frame count; otherwise use the endpoint or template default. Exception: preset/template character animations take no frame_count; pick a matching template id such as walking-8-frames (catalog in references/preset-skeleton-template-animation.md) or fall back to v3 custom mode. Preserve PixelLab's returned frame order; no ping-pong, reversed, duplicated, or trimmed outputs unless the user asks for that playback style.
9. If the user says cheap, budget, low-cost, fewer credits, or similar, read references/cost-routing.md before choosing a paid route, and ask before each extra paid attempt unless a concrete budget or attempt count was approved.
10. Before live generation, confirm the PixelLab bearer token is configured without asking the user to paste it into chat (see Auth And Execution).
11. seed: omit by default; PixelLab randomizes it. Send it in two cases only — the user gave a seed (send verbatim), or two or more calls share near-identical wording and must land on the same composition (seed-lock): generate one random positive integer (0 means random, so never 0) and send that same value on every call in the set. Never ask the user for a seed.
12. Act or answer. Once the job's live generation(s) have returned image(s) — after the last one in a chain — do three follow-ups before the final report, even if the user did not ask: the completion sound (references/bark.md), the manifest (references/usage-reporting.md), and the *.blueprint.json (references/blueprint.md). Then send one final report. Ask a short clarification only for known collisions. Before that report, send only a blocker or a question you need answered — never progress, status, findings, or intentions; those belong in the final report. For a pending job, keep polling its getter in-turn until its status is completed or failed — a still-running job is never a reason to end the turn (references/job-lifecycle.md). If the turn is being cut off before it finishes, continue with a bounded background wait instead of stopping. Hand the job back to the user only when you can do neither — no way to keep polling and no way to background a wait — then report the job or asset ID and the getter that resumes the check, and say credits were spent. Exception — chunk reveal: when an approved run produces several separately-completing image jobs (a multi-shot cinematic chain, an all-directions animation, an approved multi-asset batch), post each job's saved output — path and inline preview link — as that job completes, and fold the last job into the final report rather than revealing it twice. This covers distinct sequential jobs only, not the multiple images a single job returns at once (8-direction character, animation frames, rotations, tileset tiles, review candidates), which go straight to the final report.

Asset Integrity

• Every pixel of requested art must originate from PixelLab or the user. Local tools may read, download, assemble, package, import/export, preview, verify, mask, pad, crop, resize, and format-convert those pixels. Locally authored generation controls such as masks, palette swatches/color_image, reference guides, and shape templates are allowed as inputs; report them as inputs. Do not draw, repaint, or synthesize requested content locally unless the user explicitly approves a labeled non-PixelLab fallback.
• Reviewable static candidates: when a static image-style MCP tool or REST endpoint returns multiple alternatives, read references/reviewable-candidates.md before selecting, saving, or continuing from one.
• Do not bake a colored, checkerboard, white, black, green-screen, or matte background into transparent frames, final GIFs, spritesheets, previews, or report images unless the user explicitly asks for it. A checkerboard is allowed only as a clearly labeled inspection aid kept separate from final deliverables.
• Do not post-process PixelLab output into a claimed final asset without explicit approval. Local crop/split/format work that preserves original pixels is allowed when reported honestly; resizing, reassembling, compositing, or repairing failed outputs locally must not be called final without approval. Exception: when a request used no_background: true but the output kept a background, read references/background-removal.md and attempt safe removal when verification shows the background is removable without changing the art.
• Save downloaded generations, derived previews, manifests, and packages in a named per-generation subfolder under the pixellab-pip-generations/ folder at the user's project/workspace root — not loose in its root, and never resolved against a background or detached process's working directory, which may default to the home folder — unless the user names another location. A returned base64 image may be raw RGBA rather than PNG; confirm a saved image decodes to a valid PNG, and when a response exposes more than one image field, save the PNG-encoded one. Produce only the requested output formats or the route's minimal standard artifacts. When a job returns multiple separate images, always compile one standard preview alongside the individual files: a spritesheet for a collection of distinct sprites, or a looping preview GIF when the images are frames of a single animation (read references/local-asset-assembly.md for spritesheet grids and GIF settings). No APNG or extra preview/viewer formats unless asked.
• After a generation returns image(s), write a <name>.blueprint.json beside the outputs per references/blueprint.md — canonical portable _pixellab connection metadata, the exact route bodies, structured TASK steps for material work performed outside PixelLab calls, and a _comment_prompt holding the user's original prompt as they intended it. Remove host-added wrappers such as connector Markdown, app URIs, hidden local paths, or tool-call serialization; keep the visible command text, such as /pixellab-pip. When the generation used one or more user-supplied input images (any role — source, reference, style, mask, init, frame, and the like), copy each into the folder by copying the file, not by reading and re-writing it.
• After every live generation flow, write a manifest beside the outputs using references/usage-reporting.md; keep its private audit/resume data out of the shareable blueprint.

Destructive Remote Actions

Deleting, clearing, or overwriting existing remote PixelLab assets — characters, objects, tiles, tilesets, fonts, UI, portraits, their states, animations, or tags — in a way that discards or replaces content already stored remotely is irreversible and requires explicit user permission before it happens: either an instruction that names the deletion or overwrite, or the user's approval of a destructive change you propose. Creating a new asset, state, or animation is additive, not destructive, and is not gated here. Never delete or overwrite unilaterally as an inferred fix, cleanup, reset, sync, migration, or troubleshooting step, and never because a local list, cache, or app view looks empty, stale, or out of sync — the remote is the source of truth, so investigate read-only first (list_*/get_*, REST GET) and report what you find instead of destroying it. Proposing a destructive change is fine; carrying it out before the user approves is not. Before a confirmed destructive op, list exactly what will be removed or replaced (names/IDs and count); bulk or clear-all requires the user to confirm that scope. This covers the delete_* MCP tools and REST delete/replace endpoints.

For character file synchronization, compare the returned updated_at value or URL ?t= stamp with the local copy and download only changed assets; do not use a stale cached image as evidence that the remote needs replacement.

Surface Rules

Hosted MCP tool names are not REST endpoints; do not curl MCP tool names as /v2/... paths.

Intent Router

For any atlas or spritesheet request with known or requested cell dimensions, also read references/local-asset-assembly.md for the required grid inspection preview.

Clarify Only For Collisions

• "Presets": infer bundled blueprints from established blueprint context and preset/template animations from animation or motion context; ask which collection only when neither is clear.
• "Tiles": top-down/autotile tileset, platformer tileset, explicit-projection connectable set, independent variants, one isometric tile, path set, building kit, or packed texture sheet?
• "Map": whole map, map object, map image, tileset, isometric tile, or tile variants?
• "Isometric tileset": one tile, independent variants, or a connectable terrain set? Ask when unclear; only the connectable set uses tile_feature="tileset".
• "Object/character": infer character for people, NPCs, creatures, or identity/state animation; object for standalone props, pickups, furniture, weapons. Ask only if unclear.
• Animation direction on a multi-direction character: default to south for one preview candidate; ask only when south is unavailable, directions are unknown, or the user needs another gameplay-facing direction. Animate all directions only on explicit request or approval.
• "Effect": static or animated? If a target image is supplied, infer a one-off edit; ask reusable-asset vs one-off only without a clear edit target.
• "Paperdoll": gather base image, desired layers, target regions, directions, and whether the user wants separate transparent layer files, editor layers, composited previews, or both; see references/paperdolling.md.
• Supplied images: infer each file's low-risk endpoint-specific role from wording. Before credit-spending calls, ask when role uncertainty (identity vs style vs concept vs edit target vs mask vs palette vs first/last frame) would change the endpoint or output; see references/image-input-roles.md.
• If prompt enhancement adds material inferred details, surface the proposed description in the cost-approval gate (references/auto.md) before a credit-spending call.

References

Resolve every references/ path against this skill's own directory (the parent of this SKILL.md) and use an absolute path in the tool call. If that directory is unknown to you, find the pixellab-pip/references/ folder by listing or searching the workspace and agent-skill directories before acting; do not skip the read. When a rule names a reference, open and read it before acting, then follow its current text — not memory or a summary. Your training does not contain these PixelLab-specific contracts, so answering from general knowledge — for example treating Pip as a pip-installed Python package — will be wrong. If a required reference cannot be read, say so and stop rather than improvise its contract.

Read each reference only when its trigger applies:

• Bearer-token setup, PixelLab UI naming, MCP auth reuse: references/credentials.md.
• Setup wizard for MCP, REST v2 fallback, auth after install: references/setup.md.
• Update an installed Pip to the latest version: references/update.md.
• Remove an installed Pip: references/uninstall.md.
• Persistent completion sound toggle: references/bark.md.
• Cost-approval gate before paid calls, and the auto on/off toggle: references/auto.md.
• Safe post-processing when no_background: true fails: references/background-removal.md.
• Skill/ability and inventory item icon sheets: references/icon.md.
• Create Image Pro, native-size multi-output batches, exact grids, below-32px cells: references/create-image-pro.md.
• Cheap/budget/credit-minimizing route selection: references/cost-routing.md.
• Paperdolling and layered characters: references/paperdolling.md.
• Review/choice handling for static candidate alternatives: references/reviewable-candidates.md.
• Tilesets and tile variants: references/tileset.md.
• Style-reference generation, Aseprite-equivalent square padding, and output sizing: references/style-reference.md.
• Supplied image roles, endpoint image fields, fixed-size image-to-pixelart: references/image-input-roles.md.
• Non-English or mixed-language requests: references/localization.md.
• Official PixelLab doc URLs and boundaries: references/official-pixellab-documentation.md.
• Generation reports and manifests after PixelLab calls: references/usage-reporting.md.
• Per-generation blueprint (PixelLab calls + agent tasks), recreation, and sharing: references/blueprint.md.
• Async jobs, MCP review state, rate limits, download expiry: references/job-lifecycle.md.
• Preset/template/skeleton character animations: references/preset-skeleton-template-animation.md.
• Raw animation, interpolation, outfit transfer, idle-loop risk: references/animation.md.
• Talking portraits, viseme generation, talking GIFs, and lip-sync plans: references/vocal-animation.md.
• Multi-shot, multi-second, or seamless-loop cinematics from chained animations: references/cinematic.md.
• Editor-only utilities without public routes: references/editor-only-utilities.md.
• PixelLab project/sandbox/chat/agent MCP tools: references/mcp-platform-tools.md.
• REST v2 prompt/field character limits: references/prompt-limits.md.
• Explicit Aseprite handling, .aseprite workspaces, palette quantization, CLI/Lua export: references/aseprite-cli.md.
• Third-party Aseprite MCP servers: references/aseprite-mcp.md.
• Atlas/spritesheet grid inspection previews, local assembly, preview GIFs, and ImageMagick: references/local-asset-assembly.md.

Optional broader docs: in full plugin/repo installs these resolve relative to this SKILL.md; raw skill installs may omit them. Read at most one matching file if runtime references are not enough; if absent, continue with references/official-pixellab-documentation.md and current official docs.

• Surface boundaries and service selection: ../../docs/pixellab/pixellab-surfaces-and-services.md.
• Plain-language asset routing: ../../docs/pixellab/pixellab-asset-routing.md.
• Product/model/mode terminology: ../../docs/pixellab/pixellab-terminology.md.
• SDK-vs-REST compatibility: ../../docs/pixellab/pixellab-sdk-compatibility.md.
• Bearer-token, session, and security boundaries: ../../docs/pixellab/pixellab-auth-and-security.md.
• UI generation and MCP-vs-REST UI routing research: ../../docs/pixellab/pixellab-ui-generation-surfaces-research.md.
• Multi-shot cinematic technique research (chained-animation findings): ../../docs/pixellab/pixellab-cinematic-spike.md.
• Cinematic scene composition and motion technique (inspiration): ../../docs/pixellab/pixellab-cinematic-inspiration.md.

Model And Mode Terms

Treat PixelLab model/provider language as product labels unless official docs disclose more. Do not invent provider internals where docs are silent.

• Pixen, PixFlux: product/workflow labels, not guaranteed provider names.
• PixPatch: website-surface label; no public v2 PixPatch endpoint exists.
• Pro: a quality/tier label across many unrelated tools, not one endpoint or model. Treat Pro and Pro Tools routes as expensive unless current docs prove otherwise.
• v3 and new: workflow/version labels scoped to a selected operation. Cheap-family hints, but check the endpoint — REST inpaint-v3 is documented as Pro.
• standard: a legacy generation mode, not a quality tier (the standard/pro split on characters, tilesets). Use it only when the user explicitly asks or a route reference directs it.
• S-XL, M-XL, S-M, M-L: size/product labels, not asset intents.
• Gemini: a cost-tier label in current edit/inpaint schemas; its older website Create Tileset Pro usage is stale and must not be presented as current.

Text Preparation

Exact field values win over prompt prep. If the user explicitly supplies a PixelLab-facing field value, such as prompt: ..., description: ..., action: ..., or use exactly ..., send that value unchanged and do not enhance it. If it is invalid, over limit, or unsafe, stop and ask for an approved replacement or trim before spending credits.

Prompt enhancement is opt-out. Otherwise, for natural-language parameters such as description, style_description, negative_description, *_description, action, item_descriptions, text, and color_palette, produce the best concise PixelLab-ready English value from the request and visible inputs before calling a tool. For non-English or mixed-language requests, load references/localization.md and obtain the user's approval for the exact English transformation before the first external call. Exception: /talking-gif.text, /lip-sync.text, and their MCP text_to_speak fields are dialogue content; preserve the user's wording exactly and do not enhance or translate it.

Prompts describe visual content or, for action fields, depicted motion — never tool operation, output metadata, or report status. Include only details that change output; omit boilerplate already expressed by a supported control. Prefer supported controls and positive structural wording. Use inline exclusions only for a specific visual constraint, not generic boilerplate; no separate field is required. On Pixen, describe the intended empty or replacement state instead of naming an otherwise absent object only to exclude it. Send negative_description only when the live schema exposes it. For a named visual style, state it briefly and avoid conflicting render adjectives; use route-specific references for additional style guidance.

Respect documented character limits: many REST v2 description fields allow 2000 characters, but several action/edit/style fields cap at 500. On a length rejection, trim without changing intent, note the adjustment, and retry. Exact limits: references/prompt-limits.md or OpenAPI.

Use one enhancement path per call. Inline enhance_prompt flags exist on create-image-pixen, animate-with-text-v3, create-character-v3, animate-character/characters/animations, and object animations, cost about 0.05 generations, and are preferred over a separate enhancer call when the route has one. Constraints: for character/object animation, enhance_prompt is valid only with mode="v3"; for create-character-v3 it is valid only for from-scratch generation. These are REST-only fields — the matching MCP tools (create_image_pixen, animate_image, create_character) expose no enhance_prompt; on an MCP-first route, enhance directly as the agent instead. Standalone enhancers: enhance-pixen-prompt for Pixen image prompts, enhance-animation-v3-prompt for animation v3 actions, enhance-character-v3-prompt for character-v3 prompts. Otherwise enhance directly as the agent; do not force a mismatched enhancer.

Do Not Use

• No local code or editor automation to create or alter requested visual content: no PIL/Pillow drawing, canvas/SVG drawing, ImageMagick draw, Aseprite Lua drawing, ASCII-to-image, or procedural pixel placement. Local code may copy, mask, composite, and verify pixels that came from PixelLab or the user.
• No undocumented internal endpoints used by first-party surfaces: root website routes, unversioned https://api.pixellab.ai/ paths like /tilesets/create, or Aseprite extension operation URLs. Treat them as unsupported unless they appear in public REST v2 docs/OpenAPI or MCP docs.
• Never ask users to paste the PixelLab bearer token into chat; direct them to the setup wizard, local PIXELLAB_SECRET, or app secret settings.
• Never scrape browser session tokens or cookies. Website session tokens are not API bearer tokens; never use one for the other.
• Do not default to v1 or old SDK README examples for new work, and do not assume an installed SDK covers every current v2 endpoint — confirm the installed package or call REST v2 directly.

Current Docs Refresh

Route from this skill first. Refresh official docs only when a needed tool, endpoint, field, schema, SDK detail, auth step, price/limit, or model/mode claim is missing or unclear. Start lightweight; fetch openapi.json only for exact schemas.

• https://api.pixellab.ai/v2/llms.txt — REST v2 endpoint index and auth summary
• https://api.pixellab.ai/v2/docs — interactive REST v2 parameters
• https://api.pixellab.ai/v2/openapi.json — exact schema checks only; read a field's existence, type, or default from the raw JSON, not a prose summary
• https://api.pixellab.ai/mcp/docs — MCP tool behavior
• https://www.pixellab.ai/mcp — MCP setup
• https://github.com/pixellab-code — official SDK/MCP repo state only
• https://api.pixellab.ai/v1/openapi.json — legacy checks only

If web access is unavailable, answer from this skill and say which current claim could not be freshly verified.

Auth And Execution

If no bearer token is configured, stop before generation and offer the setup wizard: the user opens https://www.pixellab.ai/account after signing in, copies the value labeled Secret, and stores it locally as PIXELLAB_SECRET or in app secret settings — never pasted into chat. For Manual setup, link https://www.pixellab.ai/mcp and stop. PixelLab UI/docs may call this value an API key, API token, or secret; for REST/MCP bearer auth, call it a bearer token.

For questions, answer with: recommended surface/endpoint, why it fits, warnings for unsupported alternatives, and a verification note only when the answer depends on an unverified current fact.

For tasks, generate only when the user clearly requested it and token plus tooling are configured. For nontrivial work, produce one candidate first, report it, and continue only if asked. Before the first credit-spending call, apply the cost-approval gate in references/auto.md: unless the persistent auto setting is on, plan the whole paid chain, then in one message show every predicted paid call, its material inputs (including the exact prompt text), and a rough total for approval. For destructive remote actions, follow Destructive Remote Actions. Refuse unsupported automation and reroute to the closest documented MCP/REST option or a visible manual website flow. Locally authored non-PixelLab visual content requires explicit request or approval and a non-PixelLab-fallback label.

Capture a balance snapshot before a nontrivial paid call when available. After live PixelLab work, read references/usage-reporting.md and use its report layout; verify the output against the user's explicit constraints before calling it final, and say plainly when verification failed instead of silently salvaging. Do not paste secrets, raw base64, full response JSON, or internal IDs unless needed for pending status, follow-up, or debugging.

When a live generation, edit, transform, conversion, background-removal, or animation job returns image(s), read references/bark.md and apply the completion-sound contract.

Examples