Argil 500-character moment limit vs Sume's 4-60 second script window

Argil limits every moment in a video to 500 characters of transcript, so a long script has to be cut into many moments. Sume does not count characters for a whole script: it accepts a script or multi-scene plan whose estimated duration lands between 4 and 60 seconds, and rejects anything outside that window.

Both limits come from vendor pages read on 2026-10-02: Argil's Create a new Video and Create a video pages, and Sume's Generate avatar video. This post is about how to plan a script against each limit, not about which avatar looks better.

What exactly does Argil limit per moment?

On the create-video endpoint, moments is a required array and each moment needs an avatarId plus either a transcript or an audioUrl, never both. The transcript of a single moment is capped at 500 characters. Argil's guide adds that a video can hold 60 paragraphs of up to 500 characters each, which it says can reach 10-15 minutes depending on how fast the avatar speaks.

The practical effect is that you chunk by characters. A 2,400-character narration becomes at least five moments, and you decide where the breaks fall.

How does Sume size a script instead?

Sume takes either a script or ordered video_inputs on POST /v1/avatar-1.0/talking-video, and exactly one of them. It estimates the spoken duration and accepts the request only if that estimate is 4-60 seconds inclusive. For a longer piece, the docs say to shorten the script or split it into multiple jobs.

Multi-scene plans use the same window for the total planned duration. Each spoken scene sets voice.type: "text" with a script or input_text and a duration, and a voice.type: "silence" scene is a non-speaking beat that needs a duration and no text.

Side by side

The two limits measure different things, so the comparison below is about units, not about which is larger.

How should I split a script for each?

For Argil, split on sentence boundaries and count characters before you send, so no moment is rejected. Keep each moment a complete thought, because the moment is also the unit you can attach gestures, zoom or b-roll to.

For Sume, estimate seconds rather than characters. A short product pitch fits one job. A longer explainer becomes several jobs that you submit with distinct Idempotency-Key headers and join afterwards. Sume's Jobs and results page says a queued job is a normal accepted state, so submit the pieces together and poll or use a webhook rather than waiting on one at a time.

What does Sume not do here?

Sume does not give you a single avatar video longer than 60 seconds, and the docs state that one final video uses one resolved avatar. If you need a ten-minute talking presenter in one render, Argil's moment model is built for that length and Sume's is not. The Sume route fits clips: hooks, product pitches, short explainers, and per-recipient variants. See AI avatar video longer than 60 seconds for the join-it-yourself approach.