Install the Skill first, then let your Agent follow the official docs to integrate video translation, subtitle removal, dubbing, and image translation.
Send the prompt below to your Agent to add the Skills.
Visit https://github.com/zhaoli-ai/GhostCut-ai/blob/main/skills.md and follow the instructions to install GhostCut Skill for me.
Use the npx command to get started.
npx skills add zhaoli-ai/GhostCut-ai
Subtitles / text / logos
Multilingual translation / AI voices
OCR / ASR / SRT burn-in
Image translation / text removal
Why GhostCut
Cover the complete workflow from import, translation, proofreading, removal, dubbing and burn-in to final video download. It supports both fast API + website collaboration and full API automation.
Covers leading short drama globalization customers as well as AI short drama production and delivery teams.
More than 2,000 titles flow through daily production, suitable for batch task queues.
Turn translation, dubbing, subtitle removal, and final video retrieval into reusable production actions.
Customer Paths
Choose integration depth by business model: companies can call standalone capabilities, globalization teams can combine localization workflows, and Agents can orchestrate GhostCut as a tool.
Integrate image translation, image text removal, video text removal, subtitle generation, and other standalone capabilities into your existing backend.
For short drama and AI drama globalization teams, connect subtitle translation, removal, dubbing, and final video retrieval into a stable production workflow.
Let Codex, Cursor, Claude Code, or internal bots read the Skill first, then submit tasks, query status, and handle callbacks according to the docs.
Localization Workflow
This section explains how a short drama moves from the source video to a localized final video. You can connect only the front-end and back-end workflow to the API while keeping detailed refinement on the website, or integrate translation, proofreading, removal, dubbing, and final video download fully through the API.
These two modes are not sequential steps; they represent different integration depths.
Best for teams that want a fast integration. Use the API first for the most time-consuming and standardized steps: project creation, video upload, subtitle upload, subtitle translation, and final video download. Keep complex interactions such as subtitle proofreading, subtitle removal, and AI dubbing on the GhostCut website.
Best for teams with an internal system or automation platform. Source video upload, subtitle processing, proofreading, removal, dubbing, burn-in, task status, and result retrieval are all handled through the API.
Whichever integration mode you choose, a short drama usually goes through these stages.
Create a project through the API, upload the video and any existing subtitles. If there are no subtitles, continue with recognition or translation later.
Translate Chinese subtitles into the target language. When human review is needed, proofread on the website; when systematic processing is required, continue through the API.
Remove original subtitles based on the source video, burn in new subtitles, and choose whether to use AI dubbing, voice cloning, or keep the background audio.
After the task is complete, get status through callback or polling, download the final video, and send it back to your content library, ad system, or publishing workflow.
Minimum Call Shape
After installing the Skill, the minimum call is not an isolated curl command, but a repeatable loop: credentials -> payload -> signed POST -> work ID -> status query.
import hashlib
import json
import os
import requests
APP_KEY = os.environ["GHOSTCUT_APP_KEY"]
APP_SECRET = os.environ["GHOSTCUT_APP_SECRET"]
BASE_URL = "https://api.zhaoli.com"
VIDEO_URL = "https://example.com/input.mp4"
def post(path, payload):
body = json.dumps(payload, ensure_ascii=False, separators=(",", ":"))
body_md5 = hashlib.md5(body.encode("utf-8")).hexdigest()
app_sign = hashlib.md5((body_md5 + APP_SECRET).encode("utf-8")).hexdigest()
response = requests.post(
BASE_URL + path,
headers={
"Content-Type": "application/json",
"AppKey": APP_KEY,
"AppSign": app_sign,
},
data=body.encode("utf-8"),
timeout=30,
)
response.raise_for_status()
return response.json()
payload = {
"urls": [VIDEO_URL],
"needChineseOcclude": 2,
"videoInpaintLang": "all",
"extraOptions": json.dumps(
{"extra_inpaint_config": {"model": "advanced_lite"}},
separators=(",", ":"),
),
"videoInpaintMasks": json.dumps(
[{
"type": "remove_only_ocr",
"start": 0,
"end": 99999,
"region": [[0, 0], [1, 0], [1, 1], [0, 1]],
}],
separators=(",", ":"),
),
}
result = post("/v-w-c/gateway/ve/work/free", payload)
work_id = result["body"]["dataList"][0]["id"]
print("work_id:", work_id)
status = post("/v-w-c/gateway/ve/work/status", {"idWorks": [work_id]})
print(json.dumps(status, ensure_ascii=False, indent=2))Replace VIDEO_URL with your own public video URL. Local videos must be uploaded first to obtain an accessible URL. Set AppKey and AppSecret through environment variables before running.
The example serializes payload into compact JSON, generates the request header with AppSign = md5(md5(body_str) + AppSecret), and sends the exact same body_str.
After /work/free succeeds, save body.dataList[0].id. Query normal video results with /work/status; processStatus == 1 means processing succeeded.
For local media, request upload credentials first, upload the file to OSS, and use the returned URL in later API calls.
Demo Scenarios
Below are parameter examples for three high-frequency tasks. The sample code assumes the signed request helper is already wrapped: ghostcutPost(path, payload)。
Use full-screen relative coordinates to validate video inpainting, task submission, and result retrieval.
Translate a Chinese short drama into English, keep the original background audio, and burn in the translated subtitles.
Recognize Chinese image content, translate it into English, and synthesize a new image.
const payload = {
urls: ["https://gc100.cdn.izhaoli.cn/demo/1691660164246.mp4"],
needChineseOcclude: 2,
videoInpaintLang: "all",
extraOptions: JSON.stringify({
extra_inpaint_config: { model: "advanced_lite" }
}),
videoInpaintMasks: JSON.stringify([{
type: "remove_only_ocr",
start: 0,
end: 99999,
region: [[0, 0], [1, 0], [1, 1], [0, 1]]
}])
};
const task = await ghostcutPost("/v-w-c/gateway/ve/work/free", payload);
const workId = task.body.dataList[0].id;const payload = {
urls: ["http://gc9.cdn.izhaoli.cn/ghostcut_demo/HighEmotion_Dubbing_cn2en.mp4"],
needWanyin: 1,
sourceLang: "zh",
lang: "en",
wyTaskType: "VOICE_CLONE_PRO",
wyNeedText: 1,
removeBgAudio: 0,
wyVoiceParam: JSON.stringify({
font_param: {
style: "tpl-31-1-T",
font_size: 32,
position: 0.727,
subtitleLang: "en"
}
})
};
const task = await ghostcutPost("/v-w-c/gateway/ve/work/free", payload);
const workId = task.body.dataList[0].id;const payload = {
downloadInfo: JSON.stringify({
url: "https://gc100.cdn.izhaoli.cn/ve_material_image/A-00b4943646344cfc/e2d6dd82903f4059bd94c9b660f5f7b6/1782789120948.webp"
}),
translateOn: 1,
srcLang: "zh",
tgtLang: "en",
commodityFilterOn: 0,
synthesisOn: 1
};
const task = await ghostcutPost("/v-w-c/gateway/ve/image/translate", payload);
const result = await ghostcutPost("/v-w-c/gateway/ve/image/translate/query", {
id: task.body
});Integration Path
From credentials to processed results, GhostCut API integration is split into four steps: prepare access, install the Skill, submit the first task, and receive or query results.
After logging in to GhostCut, open User Center, create API access under "My Account", and get the AppKey and AppSecret for request signing.
Agents should read skills.md and install ghostcut-skill first. Developers can run npx skills add zhaoli-ai/GhostCut-ai.
Use the minimum call example to submit a video text removal task and verify the AppSign signature, task creation, and status query flow.
In production, pass callback when creating the task and keep polling as a fallback. Query video results with /work/status and image tasks with /image/translate/query.
Billing
API billing follows the same rules as the GhostCut website. You can first run tests with your website account, then choose the right package based on task volume, concurrency, and delivery schedule.
Production Rules
Current media duration is limited to 15 minutes, and each file must be under 1000MB. Split longer videos first and compress oversized files.
Video, image, and SRT URLs must not depend on login state, intranet access, or local paths, and URLs should not contain Chinese characters.
Images must be no larger than 2000 * 2000 px and 50MB. Supported formats: png, jpeg, jpg, bmp, webp. Upload local images first.
Verify Callback-Sign, deduplicate by work ID or task ID, and return HTTP 200 with a non-empty body after persisting or enqueueing the event.
Failed callbacks are retried roughly every 30 minutes within 24 hours. Even when using callback, keep compensation polling.
For AI Coding Tools
Give Codex, Cursor, Claude Code, internal bots, or automated Agents the installation entry, minimum loop, auth signing, async callback, and machine-readable index. Credentials should only be provided through environment variables, and real fields are governed by the Skill docs.
FAQ
Clarify the most common concepts, access, scenarios, async tasks, media, and billing questions before integration to reduce trial and error for developers and Agents.
GhostCut API provides video translation, subtitle removal, dubbing, image translation, and related capabilities. GhostCut Skill is the integration guide read by AI Agents and AI Coding tools so they can install, authenticate, sign requests, and call the API through the official workflow.
Log in to your GhostCut account, open User Center, create API access under "My Account", and get the AppKey and AppSecret for request signing.
It is suitable for short drama globalization, AI short drama localization, ecommerce asset localization, video subtitle removal, image translation, batch dubbing, internal content production systems, and automated Agent workflows.
Yes. You can integrate project creation, media upload, subtitle translation, proofreading, subtitle removal, AI dubbing, final video download, and related steps, or use selected capabilities together with the GhostCut website.
Video and image processing usually run as asynchronous tasks. After submitting a task, first get the task or work ID, then retrieve the final result through the status query API or callback. In production, prefer callback and keep polling as a fallback.
No. Media must be a publicly accessible URL. For local media, request upload credentials first, upload the file to OSS, and then use the returned URL in later API calls.
First check whether AppSign was generated from the exact same request body, whether the media URL is publicly accessible, whether file size and format meet the limits, whether the correct task ID was used, and whether the final status was confirmed through the correct query API.
GhostCut API uses the same billing rules as the GhostCut website. API tasks and website tasks share membership, credits, and quotas under the same account. Before going live, confirm account balance, expected task volume, concurrency needs, and delivery schedule. Packages, quotas, and prices are subject to the billing document.