A contestant pack replaces one character. A judge pack replaces five at once, since the panel you face every round comes bundled together as a single piece of content rather than five separate downloads. That makes it a bigger project than building your own contestant, though not a more complicated one — it’s just five of everything instead of one.
Every judge in the lineup needs its own dedicated character art, numbered from one straight through to five so the game knows who’s who. Nothing about how these images behave is unique to judges — it’s the same handling every character portrait in the game gets. A transparent PNG holds up better than a JPG once the character is sitting against the 3D backdrop of the studio, and whatever the lowest point of your image file is becomes the point that touches the floor, which means lifting a character visually is a matter of leaving blank space beneath them in the file rather than nudging a slider somewhere in the game. None of it scales automatically either, so building each portrait at a consistent size — somewhere around 500 pixels across, roughly double that in height — keeps the whole row of five looking like they belong on the same stage.
Land a favorable vote and two separate things fire off — a visual cue on that judge’s podium, and a sound. They’re worth understanding as two unrelated systems, because building both means thinking about them differently.
On the visual side, there’s a single shared image the whole panel falls back on by default whenever any of them votes for you — think of it as the panel’s default celebration graphic, sized to a wide rectangular shape rather than a square one. Nothing stops you from breaking a specific judge out of that shared default and giving them their own unique version instead; doing that just means naming a file after that particular seat rather than using the generic shared name, and any seat you don’t bother customizing simply keeps using the shared graphic.
The audio half splits into two entirely separate ideas that are easy to mix up. One set of sounds is tied to the position a vote falls in during a round — first vote gets one sound, second gets another, and so on, completely independent of which of the five judges actually cast that vote. The other set works the opposite way: tied permanently to one specific judge, playing whenever that individual votes for you regardless of where in the sequence it happens. Most packs use both together, though there’s a config-level switch buried in the pack’s settings file that lets you silence the vote-order sounds entirely if you’d rather rely purely on the judges’ individual voices — flipping that particular setting to off in the JSON file does the trick without touching the individual voice clips at all.
Names for the five judges themselves aren’t a filename thing at all — they get typed directly into the game’s own character-editing menu, the exact same way a contestant’s name gets set.
Realistically, building a full panel from scratch means working through five character portraits before anything else, since that’s the part with zero flexibility — a judge without art simply isn’t a judge. From there, a shared podium reaction image gets you a functional pack, and per-judge overrides can wait until you’re polishing rather than just getting something playable. Sound is the most optional layer of all three; a silent panel that still lights up correctly plays perfectly fine, it just won’t have much personality until you circle back and record something for each seat.
Everything described here isn’t limited to the regular studio experience. Streamers running the mode that lets Twitch usernames sit in judge seats instead of the computer are drawing on this exact same pack — the portraits, the podium graphics, the sounds all still apply, they’re just triggered by a real viewer’s vote instead of an algorithm’s. Any seat a streamer doesn’t hand off to chat just falls back to being computer-controlled, using the same custom pack either way, so there’s no separate content to build for that mode specifically.