小豆

小豆 is a cloud-hosted cleanup service for people who need to strip burned-in captions and logo watermarks out of vIDEo footage without wrecking the rest of the frame. Rather than covering a caption with a mosAIc blur or slicing away the edge of the picture, the service asks you to draw a box around the offending area and then reconstructs what sits underneath using surrounding texture and lighting. The work runs on rEMOte GPUs, so nothing beyond a browser or a phone is required on your side.

What it does

The main job is subtitle and watermark removal on video. You upload a CLIP or paste a share link from a short-video platform, mark the region you want gone, and subMIT the task. The ouTPUt is rendered at up to 1080p. Two operating modes are offered: a Standard mode aimed at short CLIps, and a duration-based mode for longer recordings such as lecture captures or stream highlights. Alongside that, the Same workspace bundles a handful of asset-prep utilities that content repurposers tend to need in the same session.

Key capabilities

• Region-based video inpainting — select a fixed caption zone or APPly removal across the full frame, with the model filling the gap from nearby visual context.
• Watermark cleanup — erase on-screen logos either by selecting them manually or by paRSIng a link from a supported short-video platfORM.
• photo retouching — brush over text, stamps, or stray objects in a still image and let the model rebuild the background.
• speech-to-text extraction — transcribe spoken audio from a clip into text for rewriting or translation.
• Companion media tools — dubbing, vocal and instrumental separation, upscaling, and image compression.
• REST API — submit jobs, poll progress, and retrieve results so that editing platforms or agency pipelines can automate bulk cleanup.
• Two interfaces, one balance — a desktop web workbench for precise selection and a WeChat Mini Program for quick mobile jobs, sharing the same account entitlements.

Who it is for

Short-video creators repurposing clips, livestream highlight editors, course instructors Cleaning up screen recordings, and MCN teams that process material at volume. developers building asset pipelines are the SECondary audience, since the API is the piece most comparable tools in this niche lack.

How to get started

1. Sign in with a WeChat scan on the web workbench or open the mini program.
2. Upload a video or image, or paste a platform link to pull the source file.
3. Draw a box around the caption or watermark, or switch to full-frame mode.
4. Submit, wait for cloud processing, preview the repair, then download.
5. For recurring bulk work, wire the REST endpoints into your own system.

Pricing and access caveat

Access is metered. The short-clip tier typically grants a daily allowance of processing runs, with coin top-ups once that runs out, and coins are described as non-expiring across web and mini program. The long-video tier and API usage are sold as duration packs billed by the second of footage processed. ExACT prices, allowances, and file limits may change, so confirm them on the product's own pages before committing. Two practical warnings: inpainting over faces or dense fine detail can produce distortion or artifacts, and the tool only addresses hard-coded captions, not subtitle tracks. Use it on footage you own — stripping watermarks from someone else's copyrighted work for redistribution cARRies legal risk.