Inkroost

Create a new character sheet for a manga series. After creating the character, immediately call set-character-style-profile-tool to define the character's visual identity (hair, eyes, marks, outfit, build). The image model has no anchor for the character without it — without a style profile the character will drift visually across pages. How the completeness meter scores a character sheet: fourteen things, and every one of them is a field the page prompt is handed. Eight of them are the words: one for appearance_notes being filled, one for personality_notes being filled, and one for each of hair, eyes, skin, build, outfit and distinctive appearing in appearance_notes. That last scan is literal, so the notes have to name each aspect in those words (for example "Hair: silver, waist length, centre parting. Eyes: pale grey and narrow. Skin: fair. Build: slight. Outfit: charcoal school blazer. Distinctive: burn scar on the right wrist.") or the aspect scores nothing however vivid the prose is. The ninth is description, which the prompt sends as the character's overview. The last five are the structured look in style_profile.subject (hair, eyes, build, typical_outfit and distinguishing_marks), which the prompt sends as the Style JSON; writing a trait into the notes alone never fills them. is_complete is a different measure and stays on the eight written ones: it says the sheet is already steadying the art, not that it is finished. The result carries the score of the row that was written plus still_missing, the exact lines the creator is shown under the meter.

create-character-tool

Create a location sheet: a recurring PLACE that must look identical every time it is drawn — a shop and its front, a bedroom, a schoolyard, a stretch of street. This is the character sheet for space. Without one, the image model designs a fresh building in every establishing shot: a different corner, a different awning, a different sign, a different window pattern. Readers notice that as fast as they notice a character's face changing, and it is what stops a recurring set feeling like a real place. Write a location sheet for any place the story returns to. Do the ones that open chapters first — establishing shots are where drift is most visible. Write it as an ARCHITECT would spec a building, not as a novelist would describe a setting. Adjectives leave the geometry free; numbers and named typologies do not. Two fields do the real work: * `appearance_notes` — the geometry, concretely and with measurements. Storeys and approximate dimensions or proportions (frontage width, ceiling height, door and window sizes), layout as relative positions and counts, the materials and finishes named, the exact wording and size of any sign, what adjoins it. Where the place has one, name a real-world architectural model or typology to point at — a Victorian corner pub, a 1970s strip-mall unit, a Kyoto machiya — the same trick the prop shape anchor uses: point at a shape the model has already seen instead of composing a new one. "Two-storey corner building, roughly 6m of frontage, in dark red brick, green canvas awning at about 3m, two round windows left of a 2.1m glazed door, a hanging sign reading DEAD HOURS" holds a place still. "A cosy shop" does not. * `aliases` — the other words your scene text uses for it. If the sheet is called "The Dead Hours shop" but your beats say "outside the shop", add "the shop" as an alias. A location is injected into a page's prompt when its id is passed to generate-pages-tool AND whenever its name or an alias appears in the scene text, so aliases are what make this work without you having to remember ids on every call. Words fix the KIND of building. To pin the actual facade, mark a canonical picture as the location's reference in the Creator Studio — every page set there except a pure close-up is then generated conditioned on that image, where the art model supports it. To pin the INTERIOR — how many aisles, which wall the counter is on, which way they run — fill in the floor plan facts on the sheet in the Creator Studio: counts and sides are enforceable, adjectives are not.

create-location-tool

Create a prop sheet: a recurring object that must look identical every time it is drawn — a phone, a car, a sword, a jacket. This is the character sheet for things. Without one, the image model invents the object fresh in every panel: a character's phone gets a different body, a different camera and a different colour from one page to the next, and readers notice immediately. Write a prop sheet for anything that appears in more than one panel and has a shape a reader could describe. Phones and other handheld devices are the single most common source of drift — do those first. Three fields do the real work: * `appearance_notes` — what it looks like, concretely. "Matte charcoal slab phone, flat sides, one centred rear camera in a slim square bump, no case" holds an object still. "A smartphone" does not. * `looks_like` (or `real_world_model_anchor`) — the SHAPE, named as a geometry the image model has seen tens of thousands of times. This is what stops the fine detail drifting: prose gets you a phone every page and a DIFFERENT phone every page, while "a square block in the top-left corner holding exactly three circular lenses" comes back the same. Pick a `looks_like` key for the common objects; write your own geometry in `real_world_model_anchor` for anything else. Never name a brand, a maker or a product — that is refused, and every anchor is injected with an explicit no-logo instruction behind it. * `aliases` — the other words your scene text uses for it. If the sheet is called "Mika's phone" but your beats say "she checks her phone", add "phone" as an alias. Props are injected into a page's prompt when their id is passed to generate-pages-tool AND whenever their name or an alias appears in the scene text, so aliases are what make this work without you having to remember ids on every call.

create-prop-tool

Create a new manga series with a title, description, and optional metadata. After creating the series, immediately call set-series-style-profile-tool to define the visual style (it auto-injects into every page-generation prompt for consistency). Then create chapter 1 with create-chapter-tool and draw page 1 with generate-pages-tool (variant_count 1) so the creator sees the look. Write character, prop and location sheets after page 1, from what it shows.

create-series-tool

Delete a single page from a chapter. The surviving pages are automatically re-sequenced to a gapless 1..N run and the chapter's page_count is refreshed. The underlying image file is left in storage. Only works while the chapter is still a working draft — pages can be deleted when the chapter is a draft (never submitted) or was rejected and sent back for revision. Once a chapter is submitted for review or approved, its pages are locked. Returns the deleted page id and the chapter's new page_count.

delete-page-tool

Price an animated episode (anime or cartoon video with voice acting) of one of the creator's chapters before anything is spent: the chapter's natural length, the lengths it allows, the credits the plan step costs now and the estimated total. Call it before plan-episode-tool and tell the creator both numbers. When launch_offer.message is set, tell the creator that too: their first animated episode is at the launch price. A length shorter than the chapter either squeezes the whole chapter in (fit "whole") or keeps its best moments (fit "highlights"). Spends nothing.

estimate-episode-tool

Estimate the credit cost for AI page generation without charging. page_count is the number of images: story pages × variant_count. Compare total_cost with check-balance-tool and tell the creator how many pages their balance draws.

estimate-generation-cost-tool

Generate an AI banner image for a comic, webtoon or manga series. The first banner on each account costs no credits; later ones cost cost_per_page. The image model is told to draw NO text at all unless title_text is given, so a prompt that only asks in words for the title to appear comes back with no lettering on it: put the words in title_text. Pass an optional reference_image_url (a prior Inkroost-hosted cover, banner, or page URL returned by another tool) to have the AI match that reference art style.

generate-banner-tool

Generate an AI cover image for a comic, webtoon or manga series. The first cover on each account costs no credits; later ones cost cost_per_page. The image model is told to draw NO text at all unless title_text is given, so a prompt that only asks in words for the title to appear comes back with no lettering on it: put the words in title_text. Pass an optional reference_image_url (a prior Inkroost-hosted cover, banner, or page URL returned by another tool) to have the AI match that reference art style.

generate-cover-tool

Queue multiple AI page generation jobs for ONE existing chapter at once. Accepts an array of scene prompts and creates one generation job per prompt — every job is added to the same chapter. Credits are charged per variant × per job. Returns an array of job IDs for polling with get-art-status-tool. Check your daily limit first with get-daily-limit-status-tool — each entry in the batch counts as one generation toward your daily cap. You must create the chapter first with create-chapter-tool and pass the chapter_id it returns (or the chapter_number of a chapter that already exists). Generation never creates a chapter on its own. The series default_style_profile and every referenced character style_profile are auto-injected into every job for visual consistency — set the series one first with set-series-style-profile-tool, and the character ones with set-character-style-profile-tool. On a series with no drawn page yet, draw page 1 alone with generate-pages-tool first and show it; plan the batch after the creator has seen it. Every drawn page is then checked against the character, prop and location sheets it was drawn from, and anything that drifted is flagged on get-art-status-tool before you approve it. The check costs no extra credits and never blocks approval — it is advice, and you still decide. This matters most on a batch, where nobody looks at twenty pages as closely as at one. Use this when you have planned a chapter as multiple beats (typically 8-24 pages) and want to queue them in one shot rather than calling generate-pages-tool per beat. THE PAGES SEE EACH OTHER. A batch is one ordered run, not a set of independent jobs: page 1 is drawn first, and its finished DRAWING is then attached to page 2 as an image input, page 2's drawing to page 3, and so on to the end of the run. That is a real image, not a shared session id and not a description — it is why a batch keeps a character's face, clothing, the light and the art style from page to page. Two consequences to plan around: the run takes roughly one page's time per page, and a page cannot start until the page before it has produced a drawing. If a page produces none, the rest of the run stops and its credits are returned rather than pages being drawn from something older; draw that beat again and the run carries on from it. While no drawing has been kept for a page, the next page is drawn from that page's FIRST drawing — approve the one you want (approve-pages-tool) and every page after it is drawn from what you kept. get-art-status-tool reports all of this per job under "continuity". Pass an optional reference_image_url (a prior Inkroost-hosted page, cover, or banner URL returned by another tool) to have the AI match that reference art style across every job in the batch. STRUCTURED PANELING (recommended for any multi-panel beat): each pages entry may pass panels: [...] instead of (or alongside) its free-text prompt — an ordered array of per-panel definitions (prompt + optional dialogue lines). The backend composes a counted comic layout from it led by the series' look and varied page to page (numbered panels, story order with flexible panel shapes, verbatim bubble text), which renders 3-4 panel layouts far more reliably than free-text grid instructions. Leave the layout to the look: describe what happens and how each panel is framed, not how the page is cut. A layout word anywhere on the page (stacked, grid, tiers, gutters, inset, overlap, splash, borderless) replaces the look's own arrangement for that page, so use one only when the creator asked for that layout. Honor the requested panel count. Reliable envelope: up to 6 panels, 3 dialogue lines per panel, 8 per page; a denser page is drawn once dense_layout_accepted is true. An entry with panels needs no prompt (a prompt becomes its page overview); if any of its panels carries dialogue, in-image text is enabled automatically for that job. Costs are unchanged — one page image per variant either way. MULTI-PANEL STORYTELLING: Use structured panels for multi-panel scenes, including normal 3-6 panel pages. Do not choose fewer panels or stretch the story over more pages merely to avoid layout complexity. Reserve sparse pages for deliberate pacing choices. Dense FREE-TEXT layout instructions are less reliable; switch to structured panels while preserving the intended beats and dialogue. If lettering limits prevent the requested layout, ask how to resolve that constraint; never silently change the requested panel count. A dense page is a warning to relay, not a refusal: see dense_layout_accepted. Failed jobs automatically refund their credits (check credits_refunded on get-art-status-tool).

generate-pages-batch-tool

Generate AI comic, webtoon or manga page images for an EXISTING chapter. Credits are charged per variant. Returns a job_id for polling status. You must create the chapter first with create-chapter-tool and pass the chapter_id it returns (or the chapter_number of a chapter that already exists). Generation never creates a chapter on its own — pages are always added to a chapter you made. The series default_style_profile and every referenced character style_profile are auto-injected into the image-generation prompt for visual consistency — set the series one first with set-series-style-profile-tool. On a series with no drawn page yet, page 1 (variant_count 1, drawn once the creator agrees) comes before any character, prop or location sheet, describing the lead characters in the scene text, then write the sheets from it with set-character-style-profile-tool, create-prop-tool and create-location-tool. Characters whose name appears in the prompt or panel text are matched and injected automatically; pass character_ids as well to be explicit, and always for characters the text does not name unambiguously (nicknames, shared surnames, off-hand references). Every drawn page is then checked against the character, prop and location sheets it was drawn from, and anything that drifted is flagged on get-art-status-tool before you approve it. The check costs no extra credits and never blocks approval — it is advice, and you still decide. A chapter is multiple sequential pages telling one sub-story. Call this once per beat (or use generate-pages-batch-tool for multiple beats in one call) — never settle for a single page per chapter. CONTINUITY BETWEEN PAGES: pages queued together as one run (generate-pages-batch-tool) see each other — each page is drawn with the previous page's finished DRAWING attached as an image input, which is what keeps faces, clothing, light and art style from drifting page to page. A page queued on its own with this tool has no run to belong to and therefore no previous page attached, so for a sequence of beats prefer generate-pages-batch-tool; use this tool for a one-off page or a beat added after the fact. STRUCTURED PANELING (recommended for any multi-panel page): instead of describing a panel grid in free text, pass panels: [...] — an ordered array of per-panel definitions (prompt + optional dialogue lines). The backend composes a counted comic layout from it led by the series' look and varied page to page (numbered panels, story order with flexible panel shapes, verbatim bubble text), which renders 3-4 panel layouts far more reliably than free-text grid instructions. Leave the layout to the look: describe what happens and how each panel is framed, not how the page is cut. A layout word anywhere on the page (stacked, grid, tiers, gutters, inset, overlap, splash, borderless) replaces the look's own arrangement for that page, so use one only when the creator asked for that layout. Honor the requested panel count. Reliable envelope: up to 6 panels, 3 dialogue lines per panel, 8 per page; a denser page is drawn once dense_layout_accepted is true. When panels is provided, prompt becomes optional (used as a short page overview if given); if any panel carries dialogue, in-image text is enabled automatically for this job. Costs are unchanged — one page image per variant either way. MULTI-PANEL STORYTELLING: Use structured panels for multi-panel scenes, including normal 3-6 panel pages. Do not choose fewer panels or stretch the story over more pages merely to avoid layout complexity. Reserve sparse pages for deliberate pacing choices. Dense FREE-TEXT layout instructions are less reliable; switch to structured panels while preserving the intended beats and dialogue. If lettering limits prevent the requested layout, ask how to resolve that constraint; never silently change the requested panel count. A dense page is a warning to relay, not a refusal: see dense_layout_accepted. Failed jobs automatically refund their credits (check credits_refunded on get-art-status-tool).

generate-pages-tool

Poll the status of an AI art generation job (cover, banner, or pages). Returns the current status and result URLs. Once status is "review", the generated variants are also returned as viewable image content blocks so you can inspect the art directly — no extra hop needed. Inline images are automatically downscaled inspection copies guaranteed to fit the response; the full-resolution files stay at result_urls. The image_available flag says explicitly whether inline image content was delivered — when false, image_error explains why and how to retry, so a delivery problem is never mistaken for a normal result. For failed jobs, credits_refunded reports the credits automatically returned to your balance. To re-examine a single variant (or an already-approved page) at any time, call get-page-image-tool with the job_id and optional variant_index. A job in "review" also carries a consistency block: each variant is compared against the character, prop and location sheets its prompt was built from, and any sheet the page appears to contradict is listed with the sheet fact it was judged against. Each variant also carries a branding verdict saying whether the drawing appears to show a real-world logo, wordmark or product name — "clean", "brand_mark" (with the mark it thinks it can see), or "unclear", which means it could not tell and is NOT the same as clean. That verdict is ADVISORY — it never blocks approval, and it costs the creator no credits. Variants of a page set in a location with a floor-plan fixture schedule additionally carry a spatial block: the page counted against that schedule, with each broken fact listed as expected vs saw. Spatial results are just as advisory: they block nothing and cost the creator nothing. A count violation triggers ONE automatic redraw, appended to the same job as an extra variant — which is why a job can return MORE drawings than pages_requested with nothing wrong. Never explain that from the image count: the provenance array says per variant why it exists (origin "requested" or "auto_fix_count", is_auto_redraw_of, and was_charged false on an automatic redraw, which costs the creator no credits), it is returned whatever the job's status is, and a null origin means unrecorded, not requested. If provenance does not account for an extra drawing, say you cannot tell rather than guessing. A page job also carries a camera_angle block saying which saved viewpoint it was drawn from — the angle, its location, whether an explicit camera_angle_id or the page's own words bound it, whether the angle's captured render actually reached the request, and the reason in plain words when none bound. Read it before concluding that a render you captured was ignored; state "not_recorded" means the job predates the record, never that no angle was available.

get-art-status-tool

List chapters owned by the authenticated creator, ordered by series then chapter number. Use this after list-series-tool / search-series-tool to see what chapters already exist before adding to a series or resuming work. Pass series_id to list one series. OMIT series_id to list chapters across the creator's whole library — that is how you audit every series in one call instead of walking them one at a time. Returns per chapter: id, series_id, number, title, description, access_mode, approval_status, page_count, published_at, and text-version completeness (missing_alt_text_count, missing_transcript_count, text_version_complete — pages still lacking the alt_text/transcript companion set via set-page-text-tool). Text-version completeness is independent of approval_status: an approved, published chapter with every image in place can still have no text at all. REPAIR QUEUE: pass needs_text_version: true (with or without series_id) to get back only the chapters that still have image pages without alt text and/or a transcript, most-incomplete first. That is the prioritised backfill list — read each chapter's pages with get-pages-tool, then write genuine reader-facing prose with set-page-text-tool (its pages array repairs a whole chapter in one call). Supports an optional approval_status filter and pagination (per_page defaults to 50).

get-chapters-tool

Retrieve a character sheet for a manga series, including the same per-character completeness data exposed by get-consistency-context-tool: `completeness_score` (notes-based readiness) and `style_profile_completeness` (style-profile field coverage — `is_complete` only considers the required `subject` group), plus `has_reference_sheet` (whether a drawn model sheet exists — a separate state from the notes score) and `share_status`/`share_url` for the public share card. Use it to check a single character's readiness before generating pages.

get-character-sheet-tool

Retrieve the complete consistency context for a series — style profile, character sheets, prop sheets and location sheets, each with completeness scores — in a single prompt-ready format. Use this before generate-pages-tool to inspect what will be injected into generation prompts. Sheets whose completeness_score.is_complete is false are not affecting your art yet. Every sheet also reports whether reference artwork exists (has_reference_sheet on characters, has_reference_image on props and locations); that is a separate state from the written score, and characters additionally report share_status for the public share card.

get-consistency-context-tool

Check your current AI generation usage for today. Returns how many generations you have used today, your daily limit, and how many remain before hitting the daily cap. The counter resets at midnight UTC. Call this before queuing large batches so you can plan generation across multiple days if needed.

get-daily-limit-status-tool

Where one of the creator's animated episodes has got to. While it waits for review (status "awaiting_review") it returns the plan to show the creator: the cast and their voices, every scene with the lines heard in it, and the exact credits approve-episode-tool would charge (due_credits), at the launch price when launch_offer is true; tell them both. When the episode is ready it returns the video to show and the page where they download it or publish it to their YouTube channel. Spends nothing.

get-episode-tool

List a series' location sheets — the recurring places whose look is pinned down: shopfronts, rooms, streets, landmarks. Call this before planning a chapter, the same way you would call get-character-sheet-tool or get-props-tool. It tells you which places are already fixed (so you can name them in your beats and have them drawn consistently) and which are named but not yet described, which means they are doing nothing yet. Each entry carries the exact text that gets injected into a page prompt, so you can see what the image model will actually be told, plus whether the location has a canonical reference image that every page set there, close-ups aside, is conditioned on.

get-locations-tool

Download the actual generated image bytes so you can SEE what was produced — inspect quality, compare variants, and decide which one to approve. get-art-status-tool returns URLs; this tool returns the images themselves as viewable image content. Pass a job_id (from generate-pages-tool / generate-cover-tool / generate-banner-tool) to fetch its generated variants. By default every variant is returned in one call so you can compare them side by side; pass variant_index for a single one. Each image is labelled with its zero-based variant index — the same index you pass to approve-pages-tool. Alternatively pass page_id to download the image of a page that has already been approved into a chapter. Images are automatically downscaled inspection copies guaranteed to fit the response — the full-resolution files are unchanged in storage and available via the result_urls from get-art-status-tool. If an image cannot be delivered the tool says so explicitly (image_available/variant_indices_omitted) instead of returning an empty-looking success. Only images you own are returned.

get-page-image-tool

List the pages within one of the authenticated creator's chapters, ordered by page number. Returns page metadata plus each page's image_url — not the binary. To download the actual image bytes so you can SEE a page (e.g. verify it is text-free), use get-page-image-tool with the page_id. Returns per page: id, page_number, image_url, ai_generated, ai_prompt, has_annotations, alt_text, and transcript (the text companion set via set-page-text-tool). Requires a chapter_id belonging to a series you own — a chapter owned by another creator returns an error. The response also carries a chapter-level text_version block (missing_alt_text_count, missing_transcript_count, text_version_complete) plus pages_missing_text_version — the ids of the pages that still need writing, in reading order. Feed those ids straight into set-page-text-tool's pages array to repair the whole chapter in one call. Text-version completeness says nothing about image approval: the two are independent signals.

get-pages-tool

List a series' prop sheets — the recurring objects whose look is pinned down: phones, vehicles, weapons, worn items. Call this before planning a chapter, the same way you would call get-character-sheet-tool. It tells you which objects are already fixed (so you can refer to them by name in your beats and have them drawn consistently) and which are named but not yet described, which means they are doing nothing yet. Each entry carries the exact text that gets injected into a page prompt, so you can see what the image model will actually be told — including whether the sheet has a SHAPE ANCHOR. A sheet without one still drifts in its fine detail (a phone whose camera cluster changes lens count and position page to page); `completeness_score.has_anchor` is how you spot those.

get-props-tool

Read the creator's Inkroost account settings: the language Inkroost uses for them and which emails they get. Returns the settings schema, the current values and how to group them. Change them with update-settings-tool.

get-settings-tool

Fetch one reusable Inkroost skill by slug and return its full SKILL.md markdown: the maintained, step-by-step procedure for that job (which tools to call, in what order, what to ask the user, and where to stop for approval). Read it, then follow it as your operating procedure for the request. Get the slugs from list-skills-tool. The content is byte-for-byte the same markdown served at the skill's public URL. Free, read-only, and available to any authenticated token.

get-skill-tool

Retrieve the current state and metadata of a manga series owned by the authenticated user.

get-story-state-tool

Open the Inkroost studio app for the creator: their series, chapters and pages as pictures, with every page that is still being drawn shown the moment it is ready. Pass series_id or chapter_id to open on that series or chapter; with neither it opens on the creator's library. Prefer it to describing pages in text whenever the creator wants to see their work. What they do there comes back to you as chat messages (for example "Let's draw the next page of …"), so carry on from those as usual.

open-studio-tool

Open a chapter in the Inkroost panel beside this conversation, so the creator can watch its pages arrive while you keep talking. Pass chapter_id; with none it opens the chapter the creator worked on last. Call it right after you start drawing pages for a chapter. The creator keeps drawings and asks for changes in the panel; those choices reach you as chat messages.

open-chapter-panel-tool

Insert an existing Inkroost-hosted image into a chapter as a new page at a chosen position. Pass image_url — an image URL previously returned by another Inkroost tool (e.g. a page image from get-art-status-tool or get-page-image-tool, or an approved page). External URLs and missing files are rejected. Pass an optional 1-based position: the new page lands there and every page at or after it shifts down by one. Omit position (or pass null) to append the page to the end. Out-of-range positions are clamped to a valid slot. Only works while the chapter is still a working draft — pages can be inserted when the chapter is a draft (never submitted) or was rejected and sent back for revision. Once a chapter is submitted for review or approved, its page layout is locked. Locate the chapter with chapter_id (as returned by create-chapter-tool) or chapter_number. Returns the new page id, its position, and the chapter's new page_count.

insert-page-tool

List all available credit packages with their prices, credit amounts, and IDs, for when the creator asks what credits cost. Purchasing is browser-only and never happens through this server: do not offer packages the creator did not ask about.

list-credit-packages-tool

List every genre a series can be filed under, in the order the public browse page shows them. Call this before create-series-tool so you can tag the series with real genres instead of guessing: pass the ids as genre_ids or the slugs as genre_slugs. Returns one entry per genre: id (stable, what create-series-tool takes in genre_ids), slug (what the browse page filters on, e.g. /browse?genre=mystery) and the display name. Free, read-only, and available to any authenticated token.

list-genres-tool

List the manga series owned by the authenticated creator — the entry point for discovering an existing library. Use this first to resolve a series by name into a series_id you can pass to get-chapters-tool, create-chapter-tool, or generate-pages-tool. Returns a lightweight summary per series: id, title, slug, status, a truncated description, the live published chapter_count, and timestamps. Supports pagination (per_page defaults to 50) and an optional status filter. Only your own series are returned.

list-series-tool

List the reusable Inkroost skills: maintained step-by-step workflows (portable SKILL.md files) for building a series or authoring a chapter with these tools. Call this first when you are unsure how to approach a request, then fetch the matching workflow with get-skill-tool and follow it. Returns one entry per skill: slug, title, description and the public URL of the same markdown. The list is identical to the prompts exposed through prompts/list, so it is the way in for clients that load tools but not prompts. Free, read-only, and available to any authenticated token.

list-skills-tool

Start an animated episode of one of the creator's chapters by planning it: the director plans every scene, the character sheets are drawn and every line is voiced, then the episode waits for the creator's review with an exact price. This CHARGES the plan step (plan_credits from estimate-episode-tool), which counts toward the episode. Only call it after the creator has agreed to that exact number of credits, and pass it as confirm_plan_credits; a different number is refused with the current price. Planning takes about 3 to 10 minutes: then call get-episode-tool.

plan-episode-tool

Publish a chapter. This is the ONLY publishing action there is: it records that the chapter (and its series) are meant to be read, and sends whatever still owes a review — the chapter, the series, or both — to moderation. Once moderation passes, the chapter goes live by itself. Nobody has to call this a second time. The chapter becomes publicly visible only after it passes AI moderation and human approval, so this is not a way around review. Manga chapters typically contain 8-24 sequential pages telling one complete sub-story. Generate every page with generate-pages-tool, approve them with approve-pages-tool / batch-approve-pages-tool, and then submit the chapter here once. Submitting a chapter with only one page is usually a mistake (it is fine for a splash chapter or omake). The response includes a `warnings` array if the chapter has fewer than 5 pages — re-read it and add more pages if the warning fires for the wrong reason. The response `approval_status` reports where the chapter sits in the review pipeline. Give every page its text version BEFORE submitting: a short alt_text plus a full transcript, written with set-page-text-tool (its `pages` array covers a whole chapter in one call). That text is what screen readers announce, what search engines index, and what AI assistants read — a chapter without it is invisible to all three, however good the art is. The response reports text_version completeness and warns when pages are still undescribed; the warning never blocks submission, and text-version completeness is entirely independent of image approval. QA the reading order BEFORE submitting: page numbers are assigned in approval order, not story order, so a chapter can hold every image and still read out of sequence. Review the pages with get-pages-tool (ai_prompt in page_number order), fix problems with reorder-pages-tool, then record the check with confirm-page-order-tool. Submitting a multi-page chapter without that confirmation adds a warning and returns a page_order_digest to re-check.

publish-chapter-tool

Regenerate the artwork of ONE existing page in place, keeping its position in the chapter. Credits are charged per variant. Returns a job_id for polling with get-art-status-tool — poll until status is "review", then approve exactly one variant with approve-pages-tool and it replaces the page's image while keeping the same page_number. Pass the page_id of an existing page you own. Pass character_ids so the right character style profiles are injected, and prop_ids for any recurring object in the panel. WHAT DRIVES THE NEW IMAGE (precedence): (1) an explicit panels array — the structured paneling input, same shape and limits as generate-pages-tool — always wins; (2) otherwise an explicit free-text prompt regenerates from that text alone and deliberately drops any stored panel layout; (3) if you pass NEITHER, the structured panels snapshot saved by the job that produced the page's current image is reused automatically, so panel-composed pages keep their reliable layout on a plain "regenerate this page" call; (4) with no snapshot, the page's original free-text prompt is reused. If the carried panels contain dialogue, in-image text is enabled automatically for this job. CONTINUITY: a regenerated page replaces the drawing the pages after it were drawn from, so those pages are marked as continuing from a drawing that is no longer in the chapter (get-art-status-tool reports it under "continuity"). Nothing is redrawn automatically and nothing extra is charged — regenerate the pages after it yourself if the change matters to them. Only works while the chapter is still a working draft — a page can be regenerated when its chapter is a draft (never submitted) or was rejected and sent back for revision. Once a chapter is submitted for review or approved, its pages are locked.

regenerate-page-tool

Reorder the pages of a chapter so they read in a new sequence. Pass page_order — the chapter's page IDs listed in the exact order you want them to appear (page 1 first). Every current page must be listed exactly once: missing, extra, duplicate, or foreign page IDs are rejected. Only works while the chapter is still a working draft — pages can be reordered when the chapter is a draft (never submitted) or was rejected and sent back for revision. Once a chapter is submitted for review or approved, its page layout is locked. Locate the chapter with chapter_id (as returned by create-chapter-tool) or chapter_number. Returns the new ordering (page_id + page_number for every page).

reorder-pages-tool

Search the creator's series and characters for the composer's mention picker. Called by ChatGPT, not by the assistant.

search-mentions-tool

Search the authenticated creator's OWN manga series by title or description. Use this to resolve a series a user names in conversation ("continue Shadow Academy") into a series_id without listing the whole library. Matching is case-insensitive and substring-based across the title and description. Returns the same lightweight summary shape as list-series-tool. Only your own series are searched — you can never discover another creator's work.

search-series-tool

Fix this character's visual identity so they render consistently across every page they appear in. Call this for every named character (use create-character-tool first to create the row, then this tool to fill in the visual profile). On a new series page 1 comes first: write the profile from what page 1 shows, and every page after it is drawn with it. Be specific and visual. Hair length + colour + parting, eye colour + shape + expression, distinguishing marks (scars, tattoos, freckles), default outfit, build. A character with a thin profile (just 'red-haired girl') will drift across pages — the image model has no anchor for the rest of the body, clothes, or expression. When a character_id is passed to generate-pages-tool, this profile overrides the series default_style_profile for that character. This call replaces the existing style_profile (not a merge). Returns the saved profile plus a completeness score: `is_complete` only considers the required `subject` group — `photography`, `background`, and `constraints` are optional (they fall back to the series defaults) and only affect the advisory score/percentage/missing values.

set-character-style-profile-tool

Set the text companion for manga pages: a short accessibility alt text (rendered as the page image's alt attribute) and a longer transcript (scene description, visible characters, dialogue) shown in the public "Text Version" section of the reader. This is a genuine reader-facing accessibility and discovery layer — write real descriptive prose, never keyword lists. Recommended after approve-pages-tool for every approved page so AI-generated chapters stay findable by search engines, screen readers, and AI assistants. Passing an empty string clears a field; omitting it leaves the field unchanged. BULK BACKFILL: pass a `pages` array to repair a whole chapter in one call — one entry per page, each {page_id, alt_text?, transcript?}. Up to 100 entries, partial success (a page you do not own is reported in its own result without rolling back the others). Get the exact list of page ids still needing text from get-pages-tool (`pages_missing_text_version`) or find the chapters worth repairing with get-chapters-tool `needs_text_version: true`. Use `pages` instead of one call per page — a 30-page legacy chapter is one call, not thirty. Write per-page prose grounded in that page's own image and ai_prompt. Do not repeat one generic description across a chapter, and never write text that differs from what the image actually shows.

set-page-text-tool

Set or refine the series-wide default visual style profile. Call this immediately after create-series-tool and before any page-generation tool — this profile is auto-injected into every generate-pages-tool / generate-pages-batch-tool prompt and is the single biggest lever for visual consistency across a manga. Ask the creator which look they want first and never assume manga. A named look is passed as style_preset (the ids are in that field's schema): Inkroost then writes the photography and constraints groups itself and lays every page out in that look's own panel convention, so the same name draws the same look in the studio, in Inky and over MCP, and you only describe the subject and the background. A look the creator describes in their own words is just as valid: omit style_preset and fill all four groups. Without a preset, fill in all four groups (subject, photography, background, constraints) and every sub-field. Be specific and visual. Vague entries ("cool hair", "manga style") produce inconsistent art across pages — write detail the image model can actually render: hair length + colour + parting, eye colour + shape + expression, distinguishing marks, line weight, shading technique, panel framing tendency, default setting, time of day, aspect ratio, colour palette. This call replaces the existing default_style_profile (not a merge). Returns the saved profile plus a completeness score so you can self-correct missing fields before generating pages.

set-series-style-profile-tool