SeedAgent.ai - AI Video Production

v2.33

Sep 3, 2026

Latest

Cut your takes into one film without leaving the canvas

Turning several finished takes into one video meant downloading each of them and opening an editor. The new Video Edit node does the assembly on the canvas: trim every clip, put them in order, add crossfades, subtitles, a colour grade and background music, and the finished cut renders in the background.

NewNew Video Edit node: wire finished videos into it or tick them from anywhere on the canvas, set in and out points on each clip, reorder them, choose a cut or crossfade between clips, and render one film — with optional burned-in subtitles, a colour grade, loudness normalisation, 720p or 1080p output and a music track mixed underneath. · Canvas

NewA render keeps going if you pan away or reload — the node picks the job back up and shows which step it is on — and the finished cut can be played, downloaded or re-edited right on the node. · Canvas

NewAI cuts on any clip: the take is transcribed, filler words, false starts, repeats and long pauses are marked for you, and you confirm or change the call word by word before they are cut out of the render. · Canvas

NewDescribe the cut in plain language and the node lays out the timeline for you (the model reads the transcripts, never the pixels); plus the canvas Assistant can now build and run edits. · Canvas

v2.32

Sep 2, 2026

Take history is back on portrait video nodes

A 9:16 video node is narrower than a widescreen one, and the row of buttons under the clip had quietly outgrown it — the version switcher, sitting at the far left, was pushed past the card edge and cut off. Portrait nodes looked like they had no earlier takes at all. The row now wraps instead of clipping.

FixedPortrait and square video nodes show the version switcher again, so you can flip back through earlier takes instead of only seeing the newest one. · Canvas

v2.31

Aug 31, 2026

Polishing from a text card knows what it is feeding

A text card wired into a video node shows a polish button, and so does the video node — same text, two buttons, and the one on the card was quietly the worse of the two: it did not know your model or which images were connected. Now it uses the wired node's settings and says so.

FixedPolishing from a text card now uses the model, scenario, and connected images of the video node it feeds, instead of a generic pass that could rewrite your @references away. · Canvas

ImprovedBoth polish buttons say which node's settings they are using, so a text card wired to a video node no longer leaves you guessing which one to press. · Canvas

ImprovedA text card feeding several video nodes asks which one to polish for instead of quietly picking one. · Canvas

FixedVoice transcription now detects the spoken language instead of assuming English, so Chinese and mixed-language clips stop coming back as garbled English guesses. · Review

ImprovedPer-line pronunciation scores now use precise audio alignment, so Chinese dialogue gets scored line by line instead of coming back empty. · Review

NewGenerate subtitles for any finished video right on its node — auto-transcribed from the audio or matched word-for-word to your script, previewed in place, downloadable as SRT/VTT for CapCut or Premiere. · Canvas

NewNew Sound FX node: describe an effect — a door slam, rain on glass, a crowd gasp — and get a clip up to 30 seconds you can preview, download, or wire straight into a video node as reference audio. · Canvas

NewMusic finally makes it into the film: wire a music node into a video node's BGM port and merging mixes it under the original audio — pick soft, normal, or loud, loop short tracks, and it works on a single video too. · Canvas

ImprovedThe merge button on video chains is back, now with the BGM mix controls built in. · Canvas

NewNew Dialogue node: write or paste a multi-character scene, give each character a voice, and render the whole conversation as one expressive track — with emotion tags like [whispers] and [laughs] inline. · Canvas

NewScript breakdowns get a one-tap Dialogue VO button per scene — it lifts the scene's lines, keeps them grouped by character, and drops a ready-to-voice dialogue node on the canvas. · Canvas

Changelog

Cut your takes into one film without leaving the canvas

Take history is back on portrait video nodes

Polishing from a text card knows what it is feeding

Built with our own crew.