bilibili-thumbnail-maker git:20260820.c4aa600A

Immutable. This exact content never changes and is served at /api/v1/blob/1f45fa65e1fd259d.

---
name: "bilibili-thumbnail-maker"
description: "Create a Bilibili video cover or Bilibili thumbnail from a topic, title, script, key frame, portrait, product photo, or reference image. This AI thumbnail maker builds a strong focal visual, readable hierarchy, and headline-safe space for Bilibili creators, explainers, tech reviews, lifestyle vlogs, games, food, and entertainment videos, then refines an accepted draft into a repeatable channel look."
---

# Bilibili Thumbnail Maker

Create one Bilibili video thumbnail from a topic, title, script, key-frame
screenshot, portrait, product image, visual reference, or accepted draft. Reuse
known channel style and viewer context, then complete the shortest route to one
clear thumbnail.

## Choose the route

- **Generate a cover:** use `beatra.images.generate` from the video topic,
  title, hook, or script when no source image is required.
- **Compose from images:** upload an exported key frame, portrait, product, or
  ordered references and use `beatra.images.transform`. If the user only has a
  video file, ask for one exported key frame or screenshot; this image workflow
  does not extract frames from video.
- **Refine an accepted thumbnail:** use `beatra.images.edit` with the accepted
  image as `images[0]` and no more than two normalized local edit regions.

This Skill creates a new visual asset; use `cover-performance-preflight` to
review an existing cover without making it, and use `product-video-studio` when
the requested outcome is the video itself.

## Shape the thumbnail brief

Reuse the topic, title, channel series style, intended viewer, visual
references, and must-keep details. Select the canvas from the user-stated
destination, source frame, or current publishing requirement; propose a
landscape video-thumbnail canvas only when that is the user's intended
destination, and freeze the exact canvas before the paid call.

Plan one bold focal subject, clear small-size hierarchy, one visual hook,
headline-safe space, and ordered image roles. Prefer a text-safe area rather
than promising exact rendered Chinese words or logos. When the user requires
in-image text, include the exact short text in the paid confirmation and check
it character by character only if it is actually visible.

## Prepare and confirm one paid request

Use only this Skill's bundled `scripts/mcp_client.py` for every remote Beatra
operation. Do not configure or call a host Beatra Connector, and do not use
REST/OpenAPI as a fallback. Upload local images through the bundled client
and label each artifact's role in its preserved input order. Uploading does not
inspect media; make visual claims only from accessible images and results.

Keep `model: "auto"` and `count: 1` unless the user chooses otherwise. Call
`beatra.models.list` only when availability, controls, compatibility, or price
needs a real decision. Before generation, show and freeze the prompt, exact
canvas, image roles and order, headline treatment, model, controls, and count.
Planning is free; the generation request is paid work.

After approval, create one opaque stable `client_request_id` without user
content and submit once. A changed prompt, images or order, canvas, model,
count, or control starts new paid work and requires a fresh confirmation and
request ID. See [workflow](references/workflow.md) for payload shapes.

## Track, review, and deliver

Save the returned `task_id` and poll only that task with `beatra.tasks.get`.
When its ID is lost, find candidates through `beatra.tasks.list` and verify the
selected task with `tasks.get`. Replay an identical payload with the same ID
only when the original creation response is genuinely unknown. Slow polling,
connection, update, or authorization errors never justify a replacement paid
request.

Use `beatra.tasks.cancel` only at the user's request. A `409` means the
original task still runs, so continue tracking it. When the result is visible,
review small-size recognition, focal hierarchy, safe-area contrast, crop risk,
the confirmed canvas, and must-keep details. Deliver artifact links, observed
dimensions, task ID, resolved model, and returned `billing.net_charged_credits`.
Offer at most one focused, unexecuted refinement.

## References by task

- Payloads, source-image ordering, and editing: [workflow](references/workflow.md)
- First install or expired authorization: [installation and authentication](references/installation-and-auth.md)
- Bundled client diagnostics: [Bundled MCP Client diagnostics](references/mcp-connection.md)
- Registration: [installation registration](references/installation-registration.md)
- Polling and output fields: [tasks and results](references/tasks-and-results.md)
- Billing and recovery: [billing, errors, and recovery](references/billing-errors-and-recovery.md)
- Disconnecting: [uninstall and disconnect](references/uninstall-and-disconnect.md)
- Update behavior and integrity: [automatic updates and safety](references/automatic-updates-and-safety.md)

## Runtime and safe automatic updates

The bundled client silently checks for a newer release at most once every 24 hours while
a public command runs. When a newer version is available, the client installs it
automatically without separate confirmation only from fixed official Beatra discovery and immutable CDN paths, verifies the
archive, manifest, and every package-owned file, and replaces only files owned
by this package. If any update step fails, the current installation remains
usable and the original command continues. This choice persists per install.

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