使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/meitu/vmake-skills --skill vmake-image-remove-watermark开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Remove image text watermarks
Capability and preparation
Use image-remove-text to identify and remove text and text watermarks and repair the covered background. Proceed when the input and text-removal goal are clear; the service performs recognition and processing.
Use parameters supported by the current CLI; consult help or contracts for details. Requires vmake-labs-cli, targeting version 0.1.11, compatible with >=0.1.11 <0.2.0.
Before the first invocation, read CLI setup, resolve the executable, and check its version and tool image-remove-text --help. If JSON fields or media limits are unclear, query the relevant vmake contracts entry. For output parsing, service questions, or interrupted execution, read runtime and recovery.
Collect input
- Obtain one actual image file or URL from the request or attachments. Do not ask again when it is clear. If several candidates are ambiguous, ask the user to choose; process multiple inputs individually only when authorized, and keep a record for each.
- Use a real readable path for a local input. Download remote originals according to CLI setup, then submit the local file. If only a media ID is available, query its session for the real URL first. If downloading fails, ask for the original local file.
- Add
-r '<session_id>'to continue a task; omit it for a new one. Follow CLI setup for defaults, authentication, and missing information. Consult current help for optional parameters and omit unspecified values. - Keep the original source, local path, and session for recovery. When the user supplies a downloaded file, continue the established goal. Do not treat failed output as a source.
Upload and submit
Use the resolved executable with the prepared local file as an individual argument:
vmake tool image-remove-text --image-file '<local-image-path>'The CLI validates the file, uploads it through the built-in upload pipeline to obtain its CDN URL, then submits and observes the task. Do not first call create-room or add-media, or construct placeholder media references. Parameter files and separate preflight are optional; follow CLI setup for argument handling and cleanup.
Follow and deliver
- Preserve original inputs, real
session_id/room_url, and every Task ID. Read the original command's stdout and stderr separately without starting another concurrent observer. - Deliver only after
TASK_STATE_COMPLETEDand an actual usable output URL from this task. HTTP 200, stream closure, exit code 0, orend.reason=resultalone do not prove completion. Never treat an unknown state as success. - Use native previews or attachments when available, otherwise complete clickable URLs with their query parameters. Deliver every usable Part and any successful partial result. Include an actual returned session URL; do not construct one from an ID.
- For disconnection, timeout, or uncertain acceptance, query the original session and tasks first. Retry processing only after fixing a confirmed failure and obtaining authorization. If output download fails, recover the download without repeating successful processing. See runtime and recovery for details.