Streaming

Live streaming means broadcasting video and audio over the internet in real time. It differs from a recording in three ways: it happens now, viewers can step into the action (chat, questions, polls), and you can’t fast-forward it. A recording is convenient; a live broadcast is an event. People set aside time for it, talk to you and to each other, and leave feeling that they were there. That’s why it only makes sense to go live with things that gain something from being live: results, an interview with questions, a training session with discussion, a match. A lecture without interaction is better off as a recording.

The scale is huge. According to Stream Hatchet, people watched 36.4 billion hours of live streams in 2025, only 2% less than at the pandemic peak in 2021 (37.1 billion). For scale: 36.4 billion hours is roughly 4.2 million years of viewing in a single year, and that’s only platforms like Twitch, YouTube Gaming and Kick, not counting Teams, Zoom or TV websites. Twitch holds more than half but is losing ground: in 2025 it lost 8.3 percentage points of share (from about 61% to 52.8%) after a crackdown on fake viewers, known as viewbotting, and because of Kick, which grew 131%.

The most interesting shift is in content: people watch people. On Twitch, the Just Chatting category (a talking head with no game) grew 25% in 2025 and IRL streams (in real life: the camera outdoors or in everyday life) grew 19%; non-gaming content now makes up 22% of Twitch viewing. And in the Czech context: during the Paris 2024 Olympics, Czech Television, the public broadcaster, recorded 10.2 million online stream starts alongside 4.6 million TV viewers. Live sports can no longer do without streaming; more on that in the Media and Sports chapters.

Live video on the internet is older than YouTube, but it went mainstream in 2007, when Justin Kan launched Justin.tv and broadcast his own life 24 hours a day with a camera on his cap. The gamers who streamed games there soon outgrew the founders, and so on June 6, 2011, the spin-off Twitch was born. That same year, on April 8, 2011, YouTube announced live streaming for its partners, YouTube Live. In 2012, Hugh Bailey released the first version of OBS (Open Broadcaster Software), a free, open-source tool that is still the default choice for streamers around the world. And on August 25, 2014, Amazon bought Twitch for $970 million in cash.

The second wave came with the phone in your pocket. Meerkat came out at the end of February 2015 and became the hit of the SXSW conference in March; on March 26 it was steamrolled by Periscope, which Twitter had bought even before its launch. Facebook opened live video first to celebrities (August 2015) and to everyone in January 2016; Meerkat was gone within a year. LinkedIn Live arrived on February 11, 2019, as an invitation-only beta, and from the start only through partner tools. Periscope disappeared from the app stores on March 31, 2021. The lesson: platforms come and go; your gear, your script and your audience stay.

The 2020 pandemic turned a niche skill into a necessity. Zoom went from 10 million daily meeting participants in December 2019 to more than 300 million in April 2020. Twitch grew from 11 billion hours watched (2019) to roughly 18.6 billion (2020), an increase of about 70%, and peaked in 2021 at 24.3 billion hours.

On November 15, 2024, Netflix streamed the Jake Paul vs. Mike Tyson fight: 60 million households, a peak of 65 million concurrent streams, and, in Netflix’s own words, it “had our buffering systems on the ropes.” Even the world’s biggest streaming company found out that live broadcasting is a different discipline from video on demand: you can’t pre-encode it, cache it or replay it. Remember that the next time you promise your boss a stream for 10,000 people.

A studio isn’t a room; it’s a decision about what the viewer should see and hear. We’ll distinguish three levels, and each has a different purpose. The most common mistake is buying level-three gear and using it for level-one jobs: a panel with an audience needs three cameras, a two-person interview one to three, and a call with a client none at all; a webcam and a good microphone are enough. The level prices are estimates in Czech crowns, excluding the space, furniture and people; the prices of individual items in dollars and euros are manufacturers’ indicative prices as of October 2026.

Notice what changes most between the levels: not the resolution, but the number of people you look after and the number of things that can go wrong. At level one, one thing can go wrong (the internet). At level three there are twenty: the battery in a mic, the NDI license, an overheated camera, the HDMI cable to the TV, a guest who talks away from the mic. That’s why level three needs a person who does nothing during the broadcast except keep watch.

The video signal path always has the same links: the source (camera, phone, screen), transport (cable or network), capture (a capture card or network input on the computer), switching (OBS, vMix, ATEM), the encoder (which packs the picture into H.264, HEVC or AV1) and the platform. Every link can fail, and every link adds delay. The camera itself matters less than people think; what matters is the lens, the light, and whether the camera has a clean HDMI output without icons and menus and doesn’t switch itself off after half an hour.

Cables have physical limits. HDMI reliably handles about 15 m at 1080p and less at 4K, has no lock, and drops out when someone bumps into it. SDI (coaxial cable with a BNC connector) goes 100 m over a single cable, locks in place, and is used by every outside broadcast truck: 3G-SDI is enough for 1080p60, 12G-SDI for 4K60. NDI (Network Device Interface) sends video over a regular Ethernet network: the camera sits on the network like a printer, OBS sees it as a source, and a single cable powers it at the same time (PoE). The variants: NDI High Bandwidth takes about 130 Mbps at 1080p60 and 250 Mbps at 4K60, NDI HX3 50–62 Mbps with latency under 100 ms. So a gigabit network can carry a few full streams or around ten HX3 streams, but definitely not over office Wi‑Fi.

A capture card turns an HDMI output into a USB webcam that Zoom, Teams and OBS all recognize. The Elgato Cam Link 4K, at $99.99, handles 1080p60 and 4K30 and is the cheapest step from a webcam to real video. A hardware switcher such as the ATEM Mini Pro ($349, 4× HDMI) replaces the computer: buttons switch the cameras, it has its own encoder and it can start the stream by itself. It’s useful where there’s nobody around who understands OBS. A PTZ camera (pan, tilt, zoom) replaces a camera operator: you save positions as presets and switch between them from the Stream Deck.

Viewers will forgive a soft picture, background noise or a darker shot. They won’t forgive audio they can’t understand, and they’ll leave within a minute. That’s why audio comes first in both the budget and the checklist. The basic rule: the mic as close to the mouth as possible. A lavalier mic 15 cm from the mouth sounds better than a CZK 20,000 microphone on a table a meter away. Wireless kits (the current third-generation RØDE Wireless GO: range up to 260 m, 7 hours of battery) give guests freedom and give you the job of keeping them charged.

More mics means mixing. Software can do it, but in a live broadcast you want a physical controller: a mixing console like the RØDECaster Pro II has 4 combo inputs, 6 faders, two USB-C outputs to two computers, and programmable buttons for jingles. When a guest starts whispering, you reach for the fader and that’s it. In software, you’re hunting for the right window, and meanwhile ten seconds of silence have gone by. Headphones on the operator’s ears are a must: what you hear from the speakers in the room isn’t what the viewer hears.

Light is the cheapest way to improve the picture. The classic is three-point lighting: the key light from the side at an angle of about 45°, a weaker fill light from the other side, and a back light that separates the person from the background. In practice, two LED panels with softboxes are enough, plus switching off the overhead fluorescent light, which creates shadows under the eyes. Keep one color temperature across the whole room: 3,200 K (tungsten) or 5,600 K (daylight), not a mix, or one face will be orange and the other blue. A window is a great key light in the morning and an enemy in the afternoon; the curtain is part of the gear.

Production software combines sources (cameras, screen, graphics, guests) into scenes and sends the result to the encoder. There are three families: desktop (OBS, vMix, Ecamm Live) runs on your computer and needs processing power, cloud (StreamYard, Restream Studio) runs in the browser and invites guests via a link, and hardware (ATEM) runs in a box without a computer. The choice depends on who operates the stream and where the guests are.

OBS Studio is free, open source, around since 2012, and runs on Windows, Mac and Linux. It handles scenes, transitions, NDI via a plugin, a virtual camera (OBS shows up as a webcam in Teams or StreamYard) and full-quality recording alongside the stream. The downside: you have to bring in remote guests some other way (Zoom, vdo.ninja), and the settings have a hundred options. vMix (Windows only) is a one-time purchase: Basic HD for $60, HD for $350, 4K for $700, Pro for $1,200; it has its own video calls (vMix Call), titles and instant replay for sports. Ecamm Live (Mac only) costs $16 a month (Standard) or $32 (Pro) billed annually and is the fastest route to a good-looking result for one person.

StreamYard lives in the browser: a guest gets a link, clicks it and is in the studio, while you switch layouts and put chat comments on screen. It’s free for up to 6 people on screen including you, with the StreamYard logo; Core costs $44.99 a month ($35.99 billed annually), with no logo, 10 people on screen and 3 destinations; Advanced costs $88.99, with 8 destinations. Restream is primarily a multistreaming tool: it sends one stream to several platforms at once (free for 2 channels, Standard at $19 a month for 3, Professional at $49 for 5), and it has its own studio too. The combination in Studio HOPI Connect: OBS collects the cameras over NDI, a Stream Deck switches between them, and the virtual camera goes into StreamYard, which sends the stream to YouTube, LinkedIn or Teams.

The destination is where the viewer watches the stream, and it’s the audience that decides it, not the technology. The same logic applies as with the PESO model in the Media chapter: YouTube, LinkedIn and Twitch are shared platforms with someone else’s rules and algorithm; a player embedded on your own website is an owned channel. In practice, you combine the two: the stream runs on YouTube and is embedded on the club’s website, where the program, the partners and the registration are.

YouTube is the default choice for the general public: no viewer limit, the recording stays on the channel, and you can stream as unlisted for internal purposes. LinkedIn Live is for B2B. Access goes to a profile or page with more than 150 followers or connections, an account older than 30 days, and no violations of the community policies. You broadcast through a partner tool (Restream, Socialive, StreamYard, Switcher Studio) or from your own encoder such as OBS via RTMP (Real-Time Messaging Protocol, the standard protocol for sending a stream to a platform). On top of that, since June 22, 2026, LinkedIn has scrapped spontaneous broadcasting: every live must be set up in advance as an Event, even if only a few minutes before it starts. Twitch and Kick are about community and chat: games, Just Chatting, IRL; for a company, they’re only worth it where there really is someone to chat.

Microsoft Teams is for the internal world. A regular meeting holds 1,000 active participants and up to 10,000 more as view-only attendees; events range from interactive ones for up to 1,000 people to broadcasts for 100,000 attendees with registration, moderated Q&A and organizer roles. Microsoft is retiring the old Teams Live Events: after June 30, 2026, they can no longer be scheduled, and they have been replaced by the unified Events app, where town hall and webinar are just settings. For external guests without an account, though, Teams is unfriendly: it shows them sign-in windows, and some guests give up.

The destination also decides who makes money from the stream and how much the platform keeps. On Twitch, viewers pay for channel subscriptions in three tiers and send Bits, a virtual currency they use to cheer in the chat; on top of that, the streamer gets a share of ad revenue. From subscriptions, a streamer gets half of the net revenue by default. Since May 2024, the Plus Program raises the share to 60% when a channel holds at least 100 points for three months in a row, and to 70% from 300 points; one Tier 1 subscription is one point. Since May 2026, even streamers who aren’t Affiliates or Partners yet can turn on subscriptions and Bits, but Twitch only pays them out once they reach Affiliate status, which since June takes 25 followers instead of 50. YouTube leaves the creator 70% of Super Chat and Super Stickers, paid messages and stickers in the chat, and of channel memberships, plus 55% of ad revenue, live streams included. But it only unlocks viewer payments for a channel with at least 500 subscribers and enough watch time, for example 3,000 hours over the past year. Kick lures creators with 95% of subscription revenue and also pays partners in its program for the streams they broadcast. In September 2026, it changed how those rates are calculated so that payouts fluctuate less; the subscription split stayed the same.

The second big source is money from brands. A sponsored stream takes several forms: graphics over the picture with the partner’s logo, a segment in which the creator presents a product or plays a sponsored game, and a link with a discount code in the chat. YouTube wants creators to mark such content as paid promotion, and it then shows viewers a label. In August 2026, Twitch opened brand campaigns to smaller streamers with Affiliate status, too. What this looks like in the Czech Republic is shown by Agraelus, still the biggest Czech streamer according to Forbes. Forbes estimates his earnings at CZK 25 million before tax for the year to June 2026. Most of it comes from subscriptions and fan donations during live streams, but more and more from long-term partnerships with domestic brands, such as Komerční banka or the football club Slovan Liberec, and from merchandise sales.

For a club, a federation or a company, the math works out differently. The average Twitch channel has around 25 viewers, as the Retention section will show, so viewer payments won’t pay for the stream, and its value is mostly indirect: the partner in the picture, reach, and a recording that keeps working after the broadcast. Direct money goes to those with a strong brand and the rights in their own hands. According to Forbes, tickets and paid broadcasts, so-called pay-per-view, made up roughly two-thirds of Oktagon MMA’s revenue, which topped CZK 250 million in 2022; TV rights and sponsors carry the rest. Since June 2026, when qualification for the Olympic Games in Los Angeles began, the International Judo Federation has been streaming every World Judo Tour event live only to JudoTV subscribers, and it wants their money to pay for better production. Platforms guard rights, too. YouTube scans every live stream, and when it finds third-party content, such as music or someone else’s live broadcast, it can replace the stream with a placeholder image or interrupt it. That can happen even with a license if the rights holder doesn’t add your channel to their allowlist. In a professional competition, the league sells the rights, not the club: since the 2024/25 season, the Czech football league has been broadcast by O2 TV, now Oneplay, its broadcast contracts bring the league more than CZK 430 million a year, and clubs can now share match footage on their websites and social media.

The encoder takes the finished picture and sound and compresses them into a data stream at a given bitrate, which goes out via RTMP to the platform. A higher bitrate means a better picture and higher demands on the internet connection. YouTube recommends 5–14 Mbps for 1080p30 in H.264, 6–17 Mbps for 1080p60 and 3–8 Mbps for 720p; with the newer AV1 and HEVC codecs, about a third less is enough (1080p30: 4–10 Mbps). Audio at 128 kbps AAC stereo, a keyframe every 2 seconds, constant bitrate (CBR). Twitch is stricter: about 6,000 kbps is the ceiling for non-partners, and above it you have no guarantee of transcoding into lower qualities.

The rule for the internet: upload at least twice the bitrate, because the same line also carries the call with the guest, the chat and your colleagues. For 1080p30 at 6 Mbps, you want a stable 12 Mbps up, not the “up to 100 Mbps” from an ad that’s promising download speed. Measure the upload on site, over a cable, on the day of the broadcast, and let a test stream run for at least 10 minutes: the Wi‑Fi holds up for three minutes, and then someone in the room starts a phone backup.

Latency is the delay between something happening and the viewer seeing it. Normal latency on YouTube means tens of seconds and the least buffering; ultra-low latency under 5 seconds is a must for auctions, quizzes and real-time voting, but it’s more sensitive to dropouts. For a town hall with questions from the chat, low latency (under 10 s for most viewers) is enough: the host reads a question 8 seconds later and nobody notices. What you must never skip: a backup. A second encoder (a phone with LTE and the same stream key), local recording, and a pre-made “technical break” slide that you put up when a guest drops out.

A format is a deal with the viewer: what they get, how long it will take and what’s asked of them. Live broadcasting has several established formats, and each has different technology, a different show flow and a different success metric.

For a company or a sports organization, the most valuable format is one you can repeat: the same day, the same length, the same jingle. Regularity builds an audience; a one-off show only gets one-off numbers. In judo, the rhythm is set by the tournament calendar, so the publishing plan comes from the calendar, not the other way around. The same goes for a quarterly company town hall or a club podcast every Monday.

And the live broadcast is only the beginning of the content’s life. An hour-long webinar becomes a recording, a transcript, an article, five clips for social media and a newsletter. If you count a stream’s costs only against the live viewers, you’re counting wrong: according to ON24 benchmarks based on 2024 data, 45% of webinar attendees watched the recording on demand, not the live broadcast. More on repurposing content in the Internet chapter.

The tech gets the stream on air; the show flow keeps people watching. Before every broadcast, go through the chain from the Internet chapter: Target → Audience → Message → Channel → Form → Result. What I want (sign-ups for a camp), who I’m talking to (parents of judokas aged 9–15), what I’m saying (dates, price, what the kids will experience), through which channel (YouTube embedded on the website, a link emailed to parents), in what form (30 minutes: a coach, last year’s participant, questions), and how I’ll know the result (sign-ups within a week). Without answers to these six questions, there’s no point in turning on the camera.

Then comes the script, in TV language the rundown or run sheet. It’s a table where every item has a time, a duration, what happens, what’s on screen, what’s in the audio, which device it runs on and who does it: Time · Dur · Item · Screen · Audio · Source · Who. The times are calculated from the durations, so when you move a block, everything shifts automatically.

Roles: the host keeps the pace and talks to the viewers (reads the chat, welcomes, wraps up), the director keeps the picture (cameras, graphics, the remote guest), and the producer keeps time and decides what gets cut when you’re running late. In a small team that’s two people, but never one: whoever is talking can’t be clicking at the same time. Graphics are part of the show flow: lower thirds with name and title (shown for 8 seconds at the guest’s first words), an opening loop with a countdown, a question slide, a break slide, and a closing slide with a call to action and a QR code.

Stream numbers are harsher than TV numbers, because you see every minute when people left. Three metrics are enough: peak concurrent viewers (how many people were there at once), average watch time (how many minutes they lasted) and, for webinars, the registration-to-attendance ratio. In its benchmarks based on 2024 data, ON24 reports that 57% of registrants watch the webinar (live or on demand) and stay for 51 minutes on average; for 2025 data, it reports 60% and 49 minutes. In other words: for a webinar with 200 registrations, expect 110 to 120 people, and they’ll watch an hour of content to the end if it has structure.

On open platforms, reality is more modest. In 2024, Twitch averaged 2.38 million concurrent viewers across 95,700 channels live at the same time, which works out to about 25 viewers per channel (calculated from TwitchTracker). Most streams in the world have fewer viewers than a school class, and that’s fine: 25 of the right people (customers, parents, journalists) are worth more than 2,500 random ones. For a company stream, “what percentage of the target group watched” is a better metric than an absolute number.

The retention curve has almost the same shape everywhere: a steep drop in the first 2–5 minutes (people find out what it is and leave), a plateau in the middle, and a second drop at the first “and finally.” So: start with content, not with waiting for latecomers; put the strongest thing in by minute 10; and announce the end only in the last sentence. After the broadcast, open the retention graph (YouTube Analytics, StreamYard, Teams) and find the minute where the most people left. That’s where the show flow went wrong, not the tech.

Studio HOPI Connect was built on the HOPI campus for internal streams, podcasts and video calls, with the goal of handling more than a laptop with a webcam while being run by one person. The full signal path is described at jecminek.co/studio (in Czech); here’s a short version for class.

Video: three Obsbot Tail Air PTZ cameras, powered over Ethernet (PoE) from a switch, send video over the network as NDI to OBS on a Mac; each camera is a separate source. A Stream Deck switches cameras with a single button, with no mouse and no digging through menus. OBS sends the selected shot through the virtual camera to StreamYard, and from there the stream goes to YouTube, LinkedIn or Teams.
Tail Air ×3 → PoE switch → NDI → OBS (Mac) → virtual camera → StreamYard → YouTube / LinkedIn / Teams

Audio: wireless RØDE Wireless GO mics on the guests pair with the RØDECaster Pro console, the mix happens on physical faders, USB-C goes to the Mac, and StreamYard picks the console as its microphone.
Wireless GO → RØDECaster Pro → USB-C → Mac → StreamYard
From the Mac, a mirrored picture goes over HDMI to a TV behind the main camera: the guests see themselves, the presentation and the stream layout, and because the TV sits behind the lens, they look into the camera.