Find a partner
You are queued as the seat you picked. Someone on the other side gets matched with you.
LARP 2.0 is natively multimodal. One prompt, and it answers in text, voice, images or video — with a warmth, hesitation and hand-made quality that other models still can't reach. Ask it for anything. Watch how close it gets.
LARP 2.0 was tuned on one objective: sound like someone who is actually there. Here is what that changes.
Holds a persona across a whole session without drifting, flattening, or reminding you what it is.
Outputs carry the small imperfections other models sand off. Line weight wobbles. Breath is audible.
Answers take up to 120 seconds. LARP 2.0 does not rush, and you are not billed as if it did.
Pay per call. No subscription, no seat licence, no waitlist. 120 credits land with your profile.
Every call is priced before you send it. Attach a reference image or video and LARP 2.0 will work from it directly.
Conversational generation with persistent persona. Warm, specific, and willing to sit in a scene with you.
Speech synthesis with natural breath and timing. Whisper, shout, or read a line back to you.
Image generation with a deliberately hand-drawn grain. Reference-guided when you attach one.
Short-form video generation. The heaviest call in the model, and the one worth waiting for.
One balance, both directions, held on the server so nobody can edit theirs. Every profile opens with 120 credits and no card on file.
Enough for roughly a dozen text calls, or two videos and change.
Switch sides and serve calls instead of making them. Credits land the moment you deliver.
Out of credits just means no new calls. Your conversation stays exactly where it was.
Both sides move together. A slow response is discounted for you by exactly what it takes off the inference payout.
LARP 2.0 holds full rate for 120 seconds. Past that the price slides down every second until it settles at a 40% floor — it never reaches zero, and a call is never cancelled out from under you. The timer runs on the server, so the number you see is the number you are billed against.
Attach an image or video and it arrives pinned to your prompt, driving the output directly.
A canvas is built into the session. Rough something out and the model works straight from it.
Nothing is locked in. Switch between prompting the model and serving its calls whenever you like — your credits follow you.
Open a session and use it like any other assistant. Ask for the impossible and see what comes back inside two minutes.
Take the other side and be LARP 2.0 for someone. Calls come in, the clock starts, you generate the output yourself and never break character.
No email address ever leaves your hands, because we never ask for one. The random name is what the session calls you — set a password only to pick your credits and recent session back up on another device.
You are queued as the seat you picked. Someone on the other side gets matched with you.