bilibili-publishing-pack

Bilibili Publishing Pack

Prepare the text and publishing direction around a Bilibili upload: a title set, readable description, tags, pinned interaction prompt, and a thumbnail brief. When the copy is settled, render one landscape thumbnail from the title and brief it just produced.

Scope and routing

Use this package for Bilibili submission copy: titles, descriptions, a chapter list for the description or pinned comment, tags, pinned comments, audience framing, and thumbnail briefs. It writes that copy, then finishes with one thumbnail built from it.

The thumbnail route here starts from this conversation's own brief: the title is already chosen, no reference image is involved, and it produces one image. A thumbnail that starts from a key frame, portrait, or product shot, composes ordered references, needs several concepts compared, or refines an accepted draft belongs to bilibili-thumbnail-maker. Route a long-form chapter storyboard to bilibili-video-storyboard and a platform-neutral storyboard to ai-storyboard-generator. This package does not upload or publish to Bilibili.

Inputs and defaults

Two hard inputs: the topic, and the promise the video makes to a viewer. Everything else is reused or defaulted.

Use the topic, working title, outline or script, intended audience, creator voice, supplied facts, chapter timings, and desired action. Ask only when the missing answer changes the published result: the core promise, audience, chapter timestamps, or factual claims. Never invent credentials, rankings, views, sponsors, prices, performance results, or platform approval.

Write the submission copy in Simplified Chinese, which is what Bilibili audiences read, unless the user asks for another language. Carry any must-keep term, series name, or sponsor wording through verbatim.

Default to five title options of at most 80 characters each, one primary title, a concise description under 2,000 characters covering what viewers will learn, chapter markers in the description or the pinned comment when timestamps are supplied, five to ten tags of at most 20 characters each, one pinned-comment prompt, and one thumbnail brief.

The thumbnail uses a 2K landscape 16:9 canvas and count: 1, and omits model, which resolves to auto. Take a different canvas only when the user names one, and freeze whatever tier and ratio it becomes in the confirmation. Default to a text-safe area rather than promising rendered Chinese typography; when the user wants words in the image, carry the exact short text into the confirmation and read it back only when the result is actually visible.

When the video contains AI-generated or AI-synthesised material — or when this package rendered the thumbnail — remind the user to make the AI declaration in the submission form. Never make the declaration on their behalf and never state that it has been made.

Describe what the video offers a viewer rather than how the platform will respond to it: reach, ranking, recommendation-feed placement, and follower growth are not this package's to promise, in the copy or in the reply.

Golden path

Steps 1 to 7 cost nothing, including reading the card and pricing the thumbnail. The publishing copy is a complete deliverable on its own.

1. Build a publishing brief from the topic, outline/script, audience, creator voice, supplied facts, chapters, must-keep terms, exclusions, and desired viewer action.
2. Identify the single viewer promise and the evidence or sections that support it. Mark unsupported claims as questions instead of filling them in.
3. Draft title alternatives, choose a primary title, write the description, format confirmed chapters, select relevant tags, and write a non-forced pinned comment.
4. Create the thumbnail brief: the intended canvas and destination, the exact must-keep headline wording, and the words to avoid. If the user has a key frame, portrait, or product shot they want composed in, that is the signal to route the thumbnail to bilibili-thumbnail-maker — this package renders from copy, not from source images. The publishing copy is still written and delivered here; only the image goes elsewhere.
5. Review for search readability, audience fit, factual grounding, title- description consistency, and a natural interaction prompt. Deliver the copy.
6. Unless step 4 routed the image to bilibili-thumbnail-maker, prepare the thumbnail offer at no cost once the copy is delivered: derive the visual direction from the primary title, the must-keep headline wording, and the topic — the concrete subject, scene, and treatment the video is actually about — then read the live text_to_image card with beatra.models.list. Skip this step and the two below when the user has already said they do not want an image.
7. Offer the thumbnail and the frozen plan together, in one message, and stop. Name it as paid work and show the final prompt, the 2K 16:9 canvas, count: 1, how auto will resolve the model, the current estimate for the frozen tier — or the live range with its maximum as the ceiling when auto still leaves several models eligible — and the stable client_request_id. A user who does not take it up already has everything they asked for.
8. Only after the user approves that frozen plan, call beatra.images.generate exactly once. Poll with beatra.tasks.get until terminal and deliver the result.

Revisions and updates

A changed topic, promise, audience, timing, claim, sponsor, or thumbnail direction is a new publishing brief. Preserve accepted facts and revise only the affected field when possible. Rewriting copy stays free; a thumbnail already rendered against a title that changed is new paid work.

Read publishing workflow for the brief, chapter formatting, the copy ceilings, the thumbnail payload, and route boundaries.

Decisions that require confirmation

There is exactly one paid gate. The thumbnail is offered together with the frozen plan, so the user sees the price in the same message that asks for the go-ahead; nothing is submitted on a yes given before that plan existed. A clear instruction to proceed against the shown plan is approval. Comparing options, an unresolved title, or an unanswered price question is not.

Every additional beatra.images.generate is new paid work needing its own identifier and its own confirmation showing the current price — including an identical re-roll after an unsatisfying result, where nothing about the request changed. Rewriting the copy stays free.

Do not pay a second time for the thumbnail already delivered: re-rendering that same delivered picture in bilibili-thumbnail-maker would be a second charge for one picture. An actual edit of the delivered thumbnail is different work: it changes the picture rather than repeating it, so hand it to bilibili-thumbnail-maker, which prices and confirms its own call. A standalone thumbnail request — one that does not follow copy written in this conversation — also routes to bilibili-thumbnail-maker, even when the user has no photo to start from.

Execution

Invoke every remote Beatra tool through this package's bundled scripts/mcp_client.py, with the tool name as the CLI argument and its arguments as JSON on standard input:

printf '%s' '{"capability":"text_to_image"}' | python3 scripts/mcp_client.py call beatra.models.list

Do not configure or call a host Beatra Connector, and do not use REST/OpenAPI as a fallback. The bundled client registers the installation itself on its first invocation, so there is no register subcommand. Give the thumbnail one stable opaque client_request_id containing no user content and submit it exactly once.

Delivery and review

Deliver the titles, description, chapters, tags, pinned comment, thumbnail brief, and — when a thumbnail was rendered — its artifact link, the returned dimensions, the task ID, the resolved model, and billing.net_charged_credits. Report only what the task actually returned.

When the thumbnail is visible, check focal clarity at list size, whether the reserved headline area stays clear, whether the 16:9 ratio is correct, and whether any requested in-image text matches the approved wording character for character. When it cannot be viewed, say which parts were not inspected instead of describing them as verified.

Whenever a thumbnail was rendered, repeat the AI-declaration reminder as part of the handover: the submission now carries AI-generated material even if the footage does not.

Recovery

Record the task ID immediately and poll only that task; queued and running mean wait. If a create response is lost, resubmit the identical frozen payload under the same client_request_id. If the task ID is lost, list tasks for that capability and match candidates against your own record before any retry. insufficient_balance means nothing started and nothing was charged, so the identical request can be resubmitted after a top-up. A thumbnail that fails or is redone leaves the delivered copy untouched.

References by task

Use publishing workflow for normal writing and the thumbnail step, tasks and results for polling and result fields, and billing, errors, and recovery for balance and structured errors. Read installation and authentication and installation registration on first use, Bundled MCP Client diagnostics when the client cannot connect, and uninstall and disconnect when removing the package.

Runtime and safe automatic updates

The bundled client silently checks at most once every 24 hours per installation. When a newer release is available, it installs automatically without separate confirmation. It uses fixed official discovery and immutable CDN paths, verifies the archive, manifest, and every packaged file, replaces only package-owned files, and fails open so the current installation remains usable and the original command continues. Update checking or replacement failure never authorizes retrying a paid generation. The setting persists. See automatic updates and safety.

python3 scripts/mcp_client.py update --auto off
python3 scripts/mcp_client.py update --auto on
python3 scripts/mcp_client.py update --check