Initializing... drag & drop files here
Supports: RMVB
This page is for anyone holding an old .rmvb video — a 2000s anime episode, a Chinese TV rip, a recorded lecture — who needs a still image out of it. You will learn how to land on an exact timestamp, when to pull a whole sequence of stills instead of one, and the one thing most people get wrong about PNG output here: at every quality preset except the top one, the PNG is palette-quantized rather than true-colour.
.rmvb onto the page or click "+ Add Files". A full-length episode is fine — the practical ceiling is upload size and time over your connection, not the length of the clip.2.100 means 2 seconds and 100 milliseconds.Frame Selection is the control that decides whether this is a screenshot tool or a contact-sheet tool.
| Mode | What you set | What you get back |
|---|---|---|
| Specific Frame (preselected) | Time (seconds), opening at 0, fractions accepted | One PNG from the single instant you named |
| Multiple Screenshots | Capture Rate, opening at "1 second per frame" | One PNG per captured frame across the whole clip |
The Capture Rate list runs from "0.1s seconds (single frame at 10fps)" at the dense end to "10 seconds per frame" at the sparse end. It is expressed as seconds per frame, so larger numbers mean fewer files — a 20-minute episode at "10 seconds per frame" gives you about 120 stills, while the same episode at the 10 fps setting would give you around twelve thousand. Start sparse.
PNG is a lossless container, but that does not mean every PNG this tool writes is a pixel-exact copy of the decoded frame. The frame is first decoded to a lossless intermediate, and then the preset decides how it is written out:
| Quality Preset | What the encoder does | Result |
|---|---|---|
| Highest | True-colour PNG at maximum deflate compression, no quantization | Pixel-exact copy of the decoded frame, largest file |
| Very High (preselected), High, Medium, Low, Very Low, Lowest | Palette quantization is applied before writing | Smaller file, colours reduced to a palette — visible banding in gradients at the lower rungs |
So: choose Highest when the still has to be exact — reference frames, colour work, anything you will edit further. Leave it on Very High when you want a small, good-looking screenshot and a slight loss of colour precision does not matter. The Colors control pushes this further on purpose: switching it from Original to "By Color Reduction + Dither" caps the palette (the dropdown opens at 128 and goes down to 2), with a Dither checkbox that is off unless you turn it on. Compression speed (1-10, opening at 4) trades encoding time against how hard the quantizer works, and Compression level (1-10, opening at 6) tunes the deflate pass.
Image resolution opens on "Keep original", which matches the video's native frame size. The alternatives are Resolution Percentage, a Preset Resolutions list (these are height targets — width follows the source aspect ratio, so a preset never distorts the frame), and explicit Width, Height or Width x Height boxes that accept pixels or percent.
None of that adds detail. RMVB files in circulation are overwhelmingly standard-definition and were compressed hard so an episode would fit a small download, so a still is limited by what the original encode recorded. Upscaling with a preset enlarges the pixels and nothing more. One more thing worth knowing: nothing in this pipeline deinterlaces, so if your source was an interlaced broadcast capture, the comb lines on fast motion are baked into the still and will not be removed for you.
Frame extraction can only read what a player could read. If the .rmvb is truncated from an interrupted download, or carries DRM from an old subscription service, the decode fails or produces garbage, and no online tool can lawfully unwrap the protection. Transcoding the whole file first with RMVB to MP4 sometimes rescues a partially-damaged file, after which frame grabs behave normally. And if what you actually want is a short moving loop rather than a still, use RMVB to GIF, which produces a genuinely animated output.
Only at the Highest preset. That one writes a true-colour PNG with no quantization step. Every other preset — including the preselected "Very High (Recommended)" — runs a palette-quantization pass first, which is still stored losslessly inside the PNG but has already reduced the number of distinct colours. If you need a pixel-exact frame, set Quality Preset to Highest.
Yes. The Time (seconds) box accepts decimals, and the tool's own note spells it out: 2.100 means 2 seconds and 100 milliseconds into the video. That matters when you are trying to dodge motion blur or a transition, because a tenth of a second is often the difference between a smeared frame and a clean one.
No, and nothing you can set here will give it one. We decoded a frame out of a RealMedia file and it arrives as plain three-channel RGB with no alpha channel, because the RealVideo codecs in these files do not carry transparency. PNG can store an alpha channel, but there is none in the source to preserve, so the still is fully opaque. If you need a cut-out, you have to mask it yourself in an editor afterwards.
The video is RealVideo. The decoder family covers RV10, RV20, RV30, RV40 and RV60 — labelled RealVideo 1.0 through 6.0 — with the later RV30 and RV40 streams making up most of what circulated as .rmvb. The codec is fully decoded before any image is written, so it does not affect the PNG format itself. What it does affect is how much real detail exists to capture: a low-bitrate RealVideo encode limits every still taken from it.
One image per extracted frame, at whatever Capture Rate you chose. The setting is seconds per frame, so "5 seconds per frame" on a 10-minute clip yields roughly 120 files while "0.5 seconds" yields roughly 1,200. Pick a sparse rate first, look at what comes back, and tighten it only if you missed the moment you wanted.
PNG at the Highest preset if the still is going into an editor, a comparison, or anything where colour precision matters. JPG if it is going into a document, a forum post, or an email, because a photographic SD frame compresses far better as JPG than as PNG and nobody will see the difference at viewing size. RMVB to JPG does that directly, and RMVB to WebP sits between the two.
Because the source is interlaced and nothing in this pipeline deinterlaces. Broadcast captures store two fields taken a fraction of a second apart, and on anything moving those fields disagree, which reads as fine horizontal stripes on a still. There is no setting here to fix that — you would need to deinterlace the video in a desktop editor before extracting the frame.
Usually yes. The decode runs on our servers rather than depending on whatever codecs your machine has installed, so files that Windows Media Player or a browser rejects for lack of a RealMedia decoder still convert here. The exception is DRM-protected RealMedia, which cannot be decoded anywhere.