If you’ve spent any time generating AI videos, you’ve probably hit the same wall I have. You create something incredible—maybe a talking avatar or a cinematic b-roll shot—and then the platform slaps a jittery, semi-transparent logo right in the corner. It ruins the immersion, and if you’re trying to repurpose that clip for a client project or a cross-platform upload, it’s a dealbreaker.
The two biggest names in this space right now are ByteDance’s JiMeng (即梦) and its international counterpart, Dreamina. They are essentially the same engine, but they serve different audiences. The question I get asked most often is: which one is actually harder to clean up? Let’s break down the technical reality of removing watermarks from both, because the answer isn’t as straightforward as you might think.
Before we compare, we need to understand what we’re dealing with. JiMeng (即梦) is the Chinese-language AI video generator, while Dreamina is the international version rolled out for global users. Under the hood, they share the same generation models and, importantly, the same watermarking strategy.
Unlike traditional video editors that burn a static logo into the pixels, ByteDance’s AI platforms use a dynamic, non-static watermark. It often shifts position slightly or changes opacity during playback. This isn’t an accident—it’s specifically designed to break simple crop tools and blur filters.
Here is the critical part: the watermark is not part of the original render. When you generate a video in JiMeng (即梦) Camera Mode or Dreamina, the platform encodes the clean file, then overlays the watermark as a separate layer during the export to your device. This means the pristine version still exists on their CDN servers.
This is why traditional editing fails. You can’t "un-blur" a watermark baked into the file. But you *can* extract the original source file if you know how to grab it. This is where the difficulty curve splits.
Let’s look at the practical differences in cleaning these up. I’ve tested both extensively, and here is the reality of the situation.
| Feature | JiMeng (即梦) - Chinese Version | Dreamina - International Version |
| :--- | :--- | :--- |
| Watermark Type | Dynamic overlay, moves slightly | Dynamic overlay, moves slightly |
| Watermark Position | Bottom right, large logo | Bottom right, smaller logo |
| Camera Mode (出镜模式) | Heavy watermark on AI avatar | Same heavy watermark logic |
| Extraction Difficulty | Moderate | Harder |
| Link Sharing | Share links are short and direct | Links are longer, often require login to view |
| CDN Protection | Standard, extraction works | Aggressive token validation |
| Success Rate (Manual) | Low (requires technical scripts) | Very Low (requires session tokens) |
Here is the nuance that most people miss. While the *app* is the same, the web infrastructure is different. The international version (Dreamina) has stricter anti-leeching protocols on its CDN. The links expire faster, and they often require a valid session token from a logged-in browser to fetch the source file.
On the other hand, the Chinese version (JiMeng) uses a slightly more relaxed API endpoint for share links. This makes it *marginally* easier to locate the clean file, but it still requires you to intercept network traffic or use a developer console to find the `playUrl` without the watermark parameter.
The Verdict on Difficulty: If you are trying to do this manually with browser tricks, Dreamina is harder. The token expiration is a nightmare. JiMeng (即梦) is still hard, but the share-link structure is a bit more forgiving.
If you’re reading this, you’ve probably tried the "cover it with a sticker" or "zoom in 110%" trick. Let’s talk about why that’s a losing battle, especially with these AI platforms.
When you generate a video in JiMeng (即梦) Camera Mode, the AI creates a synthetic version of you. The watermark is placed directly over your face or body in some cases. Cropping it out means losing the subject. Blurring it looks unprofessional.
The only way to get a truly clean file is to extract the original source. This bypasses the re-encoding process entirely. You don't get a "cleaned" video; you get the exact file that the AI generated before the watermark was applied. This is 100% quality, zero loss.
This is where OffWatermark comes in. I built my workflow around this exact problem. OffWatermark is a web-based tool that handles the API extraction for you. You don’t need to open a developer console or write Python scripts.
Here’s how the process works, and why it solves the "Dreamina vs. JiMeng" problem:
This works for both versions because the tool handles the token negotiation and CDN routing on the backend. It doesn't matter if the link is from `jimeng.jianying.com` or the international Dreamina domain—the extraction logic adapts.
There is another major difference between these platforms that affects your workflow: account safety.
JiMeng (即梦): Because it’s tied to the Chinese ecosystem (Douyin/ByteDance), aggressive scraping—like downloading 50 videos in a minute—can trigger risk control on your account. You need to be patient.
Dreamina: The international version is a bit more lenient with rate limits, but it is stricter on *where* you download from. If you try to pull the video in a browser that isn't logged in, you’ll get a 403 error.
The advantage of using a dedicated extraction tool is that it doesn't use your account credentials. It uses the public share link. This means your generation account is never at risk of being flagged for "suspicious download behavior." You are simply viewing a public link.
To put it simply:
But honestly, neither should be done manually anymore. It’s a waste of time. The watermark is a layer, and the clean file is sitting on a server. You just need the right key to unlock it.
If you are generating content with AI avatars for client work, or you simply want your personal videos to look clean without a "Powered by" logo, don't fight the pixels. Use extraction.
I’ve found that using OffWatermark cuts the process down to about 10 seconds per video. It handles the heavy lifting for both the Chinese and international versions of the tool, plus all the other major short-video platforms.
If I had to pick one that is *harder* to clean, I’d say Dreamina due to its aggressive CDN token validation. However, the difficulty gap is irrelevant when you use a server-side extraction tool.
Stop trying to blur or crop. It looks bad, and it degrades the AI video quality you worked hard to generate. Instead, extract the original source file.
---
> Disclaimer: OffWatermark is an independent tool and is not affiliated with, endorsed by, or connected to ByteDance, JiMeng (即梦), Dreamina, Kuaishou (快手), or Xiaohongshu (小红书) in any way. All trademarks belong to their respective owners. Users are solely responsible for ensuring their use complies with applicable laws and terms of service. Only remove watermarks from videos you personally created.
← Back to OffWatermark