Twitch Video Quality Explained: Source, Transcodes and What You Can Actually Download

Updated: 2026-07-29 · Written by the vodfetch founder

Every download starts by picking a line from a dropdown most people cannot read. This guide explains where each rendition in a Twitch quality list comes from, why the ladder differs from one broadcast to the next, and why the number after the "p" is the least useful part of the label.

Three different machines decide what you can download

The quality list in a downloader is not generated by the downloader. It is read out of the HLS master playlist Twitch publishes for that video, and by 2026 three separate systems can contribute entries to it. The broadcaster's own upload becomes the source rendition. Twitch's servers may add standard transcodes after the stream arrives. And the streamer's own machine may send several encodes at once through Enhanced Broadcasting, before Twitch sees anything at all. Which of those three ran decides what you get to pick.

"Source" — Twitch's playlists usually label it chunked — is the untouched arrival: the exact bitrate, resolution and framerate the broadcaster's encoder pushed, stored without a second pass. Everything below it in the list is a smaller copy that something else made. That is the whole reason source is worth choosing: it is the only rendition that was never re-encoded, so it carries whatever detail the stream actually had. It is also why a source rendition can legitimately read 1280x720. Source means original, not high.

Twitch defines the rest in its own words: when a stream starts, Twitch "may transform your broadcast into multiple quality formats so viewers can choose the option that works best for their device and network connection." The operative word is may. There is no rule that every Twitch video carries a 1080p, a 720p, a 480p and a 160p rung. A playlist with one entry and a playlist with seven are both entirely normal, and the difference usually has nothing to do with the video in front of you.

Who gets transcodes, and why channel size has nothing to do with it

Twitch publishes the allocation rule plainly. "Partners receive transcodes on every broadcast (first-priority access)." Everyone else falls into one bucket: "Non-partners (Affiliates and all other streamers) receive transcodes based on resource availability each time they go live." Note the shape of that. For transcoding, Affiliate status buys nothing at all — an Affiliate sits alongside a channel that went live for the first time yesterday. Partner is the only status that guarantees anything, and it guarantees it every single broadcast.

That shape is worth holding next to the other Twitch clock people confuse it with. Past broadcast retention is tiered three ways: 60 days for Partners and for channels with Prime or Turbo, 14 days for Affiliates, 7 days for everyone else. Transcoding is tiered two ways, and the line sits in a different place. Knowing which document you are reading matters, because "Affiliates get more" is true of retention and false of transcodes.

The most persistent myth is that transcodes are a reward for growing an audience. Twitch denies it outright: availability "is not influenced by follower count, viewer count, or past enforcement actions. It is determined by server resource availability at the time your stream starts." It is re-decided per broadcast rather than per channel, which is exactly why one archive from a channel can offer four rungs and the next offers one. Twitch's own answer for streamers who lack guaranteed transcodes is not an appeal or a request form; it is Enhanced Broadcasting.

Twitch adds a warning that belongs on a downloader site: "Do not trust third-party services or websites that promise to transcode video after the broadcast is sent to Twitch." Nothing outside Twitch can add a rendition to a Twitch playlist. So when a quality list comes back short, you are usually not looking at a broken tool or a blocked request — the playlist for that video lists what it lists, and a past broadcast from a stream that received no transcodes commonly shows only the source rendition. /blog/twitch-downloader-not-working separates that from an actual failure.

Enhanced Broadcasting: the ladder can come from the streamer's GPU

The modern answer to "why does this video only have source" is no longer just transcoding. Enhanced Broadcasting, which Twitch says is now available to everyone, moves the work to the broadcaster. It requires OBS Studio 30.2 or newer, or XSplit Broadcaster 4.5.2406.1801 or newer, and it sends several encodes from the streamer's own machine simultaneously, so viewers get quality options whether or not Twitch had spare transcoding capacity that night.

Twitch draws the line itself: "Standard transcodes are created by Twitch servers after your stream is received." Enhanced Broadcasting encodes are made earlier, by the broadcast software, before the stream ever reaches Twitch. For anyone downloading, that is not trivia. It means the ladder was configured automatically by software on someone else's machine, can differ from one broadcast to the next on the same channel, and owes nothing to server capacity — which is precisely the point of it.

Twitch publishes ranges rather than a fixed ladder for this. The number of renditions varies between five and two. The highest track's maximum resolution varies from 1080p60 down to 480p. Per-rendition video bitrate runs from 6 Mbps (for example 1080p AVC) down to 200 Kbps (for example 160p AVC), at either a high framerate (60, 50 or 48 fps) or a standard one (30, 25 or 24 fps), across resolutions spanning 1080p down to 160p. Those figures describe Enhanced Broadcasting specifically; Twitch publishes no equivalent table for standard server-side transcodes.

2K, Dual Format, and a second clock that costs you quality on day eight

On 17 June 2026 Twitch announced Dual Format and 2K streaming. Dual Format — a horizontal and a vertical layout delivered simultaneously — arrived "for all streamers on Twitch." 2K (1440p) is limited to Partners and Affiliates, uses HEVC rather than the AVC used at 1080p and below, and raises the ceiling to "up to 9 Mbps for 1440p and 7.5 Mbps for 1080p". Both features run on Enhanced Broadcasting, so both depend on the streamer's hardware, not on Twitch's spare capacity.

Twitch also added server-side transcode support for these streams, for "Partners and many Affiliates" — the help documentation describes the same thing as Partners and select Affiliates, and tells streamers to check Advanced Broadcast Settings in the Creator Dashboard to see whether they have it. Mechanically it is a hybrid: the documentation describes using your GPU "to encode the highest-resolution horizontal and vertical streams, while using server-side transcoding to fill in all the lower resolutions." A single playlist can therefore contain rungs produced in two entirely different places.

Then comes the part that changes when you should download rather than what you should pick. The 2K and vertical renditions carry their own, much shorter expiry. Twitch's documentation for these features states that the 2K version of a VOD expires seven days after the stream, after which the 1080p version remains for the rest of the channel's normal retention period; the vertical version of a Dual Format broadcast runs on the same seven-day clock, leaving the horizontal one behind. The video does not disappear — its top rung silently does.

Work that through on an Affiliate channel streaming in 2K: 1440p for seven days, then 1080p for the remaining seven of the 14-day retention window. On a Partner channel it is 1440p for a week and 1080p for the next 53 days. Nothing on the video page announces the change; the quality list is simply shorter on day eight than it was on day six. If a 2K or vertical broadcast matters to you, the real window is one week, not the retention figure. /blog/download-twitch-vod-before-deleted covers the deletion clock itself.

Why a video tops out at 720p when the stream looked better

Work through it in order, because only one of the causes is fixable. First, read the resolution field on the top entry. If it says 1280x720, the question is settled: that is what the broadcaster's encoder sent and Twitch never held anything higher. A missing transcode ladder cannot be the cause here — transcodes only ever add lower rungs beneath the source, never a higher one, so a short list is not by itself a reason for a low ceiling.

Second, check the date against the seven-day rule above. A 2K or Dual Format broadcast older than a week has lost its 1440p and vertical versions by design. Third, remember that Enhanced Broadcasting reconfigures itself every stream: Twitch's published range permits a highest track anywhere from 1080p60 down to 480p, so a streamer whose upstream was constrained that night genuinely broadcast lower than they usually do, and the archive faithfully reflects it.

One check settles most arguments. Compare the resolution and bitrate the playlist reports against Twitch's own recommended numbers — 6000 kbps at 1080p60, 4500 kbps at 720p60. A top entry reading 1920x1080 alongside 3.0 Mbit/s tells you the stream was nominally 1080p but bitrate-starved, which in motion looks worse than a clean 720p60. That is a broadcaster-side encoder setting, and no download choice recovers it.

Bitrate matters more than the number after the "p"

Twitch's Broadcasting Guidelines put the trade-off in one table. 1080p60 is 1920x1080 at 6000 kbps. 1080p30 is 4500 kbps. 720p60 is 1280x720 at 4500 kbps. 720p30 is 3000 kbps. Read the middle two rows again: 1080p30 and 720p60 are given the identical 4500 kbps budget. Same data per second, spent two different ways — one buys detail inside each frame, the other buys twice as many frames. Neither is the better setting in the abstract.

This is why resolution alone is a poor quality label. Motion-heavy content — a first-person shooter, a racing game, a handheld camera — spends its bits encoding change between frames. Give a 1080p60 stream 4500 kbps instead of 6000 and each frame receives roughly a quarter less data, so busy scenes soften and blocking appears in exactly the moments worth keeping. A chess stream or a talk show at the same numbers looks fine, because very little changes between one frame and the next.

Twitch recommends CBR rather than VBR, and the reason survives into the archive: VBR drops the bitrate during lulls and spikes it during action, and the spike is precisely when a marginal upload connection starts dropping frames. Keyframe interval is two seconds. Audio is AAC-LC with a documented maximum of 160 kbps — relevant if you are archiving a music or talk stream, because that ceiling holds no matter which video rung you choose. Twitch also caps a single broadcast at 48 hours.

The practical instruction is short: compare the Mbit/s figures in the list, not the resolutions. Where both are published, a rendition labelled 1080p at 3.0 Mbit/s will hold up worse on fast gameplay than one labelled 720p60 at 4.5 Mbit/s, even though its resolution reads higher. Where bitrates are not published — and that happens — you are back to resolution and framerate as the only signals available.

How the dropdown is actually built from the playlist

Every entry is read out of the master playlist with nothing added. For each variant the tool takes the NAME from the EXT-X-MEDIA line, and the RESOLUTION and BANDWIDTH from the EXT-X-STREAM-INF line, then joins whatever exists with middots: "1080p60 · 1920x1080 · 6.0 Mbit/s". The first field is the NAME; where no NAME is published it falls back to the resolution, and where neither exists it prints the bare word "source". The resolution segment is dropped when the playlist reports no RESOLUTION, and the Mbit/s figure when it reports no BANDWIDTH. Nothing is estimated or filled in.

The star is a text match rather than a measurement: a rendition is flagged as source when its NAME or its EXT-X-MEDIA line contains "chunked" or "source", which are Twitch's own words for the untranscoded copy. The sort is source first, then descending bandwidth — that is the entire comparator. Audio-only is detected separately, and the only thing that detection does is hide the MP4/TS switch, which is meaningless for audio; it gets no special treatment in the sort and settles at the bottom because it declares the least bandwidth. A variant publishing no bandwidth counts as zero and lands down there too, so read the label rather than trusting position.

The request itself shapes what appears. The playlist is fetched with source and audio-only both explicitly allowed, and with framerate inclusion switched on — that last flag is the reason a label can read "1080p60" rather than a bare "1080p". The request also advertises support for av1, h265 and h264, so an HEVC ladder from a 2K broadcast is not filtered out before it ever reaches the list.

Two visible behaviours follow from bandwidth being optional. The "≈" size estimate is the declared bandwidth divided by eight, multiplied by the selected duration, so a rendition with no bandwidth shows no estimate at all, and the estimate is computed for past broadcasts only. And the suggested filename uses only the first field of the label — "1080p60", never the resolution and never the Mbit/s. The download mechanics behind all this are covered in /blog/download-twitch-vod-1080p60.

Why a clip's quality list looks poorer than a VOD's

Clips do not come from a playlist at all. Twitch's clip API returns quality and framerate values with no bandwidth data whatsoever, so labels are assembled as resolution plus framerate — "1080p60" — and can never carry a resolution field or a Mbit/s figure. That is the honest explanation for the sparser dropdown: it is not the tool withholding detail, it is the API publishing none. For the same reason, a clip never shows a size estimate before you download it.

The star on a clip is weaker than it looks, too. There is no chunked rendition to detect, so the first entry Twitch returns is simply marked as the top one. It is Twitch's highest available version of that clip, which is what you want in practice, but it is not the "this is the untranscoded original" claim the same star makes on a past broadcast. Treat it as a ranking, not a provenance guarantee.

No downloader can exceed what Twitch published

The ceiling is structural rather than a policy choice. When a rendition is saved, the bytes are either written out exactly as fetched, or passed through an in-browser transmuxer configured to remux only and keep the original timestamps. There is no encoder anywhere in the tool. That is what "no generation loss" concretely means here: the output is a container rewrap of the same compressed video, not a fresh encode of it. It is also why the MP4 and TS options are identical in picture quality — /blog/convert-twitch-vod-to-mp4 covers that choice.

The corollary is the part people resist. Something that only copies cannot manufacture a rung that was never published. If the source rendition reads 1280x720, there is no 1080p copy sitting on Twitch's servers, and no setting, no paid tier and no competing site can produce one. Upscaling is a separate operation performed afterwards on your own machine, and it invents detail rather than recovering it. When a playlist yields nothing usable the tool stops with "No qualities available", which means the list came back empty, not that a better version is hidden.

One modern edge case is worth planning around. Because the playlist request advertises av1 and h265 alongside h264, a 2K broadcast's ladder can include renditions the in-browser MP4 step may not be able to parse. The likely symptom is the message "MP4 conversion failed — choose Format → TS and retry.", and the remedy is exactly what it says: the TS path writes the same bytes through without parsing them, and you can rewrap the result locally afterwards with a copy-only remux that re-encodes nothing.

Frequently asked questions

← Back to the Twitch Downloader