A SKAND canvas can hold the character assets, music, spoken lines, and shot videos for one short-drama episode. Recent updates prioritize lightweight previews, defer video originals and waveform extraction until playback, and reduce repeated work during dragging, saving, and job refreshes. Separate concurrency limits also keep individual batches of expensive work from occupying every processing slot.
Keep the episode's cast, audio, and shots together
Arrange character assets beside the scene references, shot videos, and audio tracks they belong to. The storyboard workbench reuses canvas material, and spoken lines bind to characters, so you spend less time finding the same assets across folders and tools.
The performance changes address that mixed-media workflow. Opening an episode, reviewing its overview, moving a group of shots, and waiting for several jobs to finish each require different work. They no longer need to reload every media item on the board.
Load visible previews before full media
Image cards prioritize thumbnails. Video cards show static posters and fetch originals when you press play. Audio can reuse a small waveform and duration cache; older files without that cache download and extract it on first playback, then reuse the result.
Nodes outside the viewport are culled. Dense zoomed-out views keep previews, names, and status while mounting fewer players and editors. Selection, references, and keyboard focus restore full cards. Preview authorization requests also batch together when many cards appear at once.
Do less work during dragging and saving
Dragging updates temporary positions; releasing the pointer creates one undoable move. Media results that finish during the gesture remain on the final board. Status refreshes cover all pending jobs, and unchanged responses no longer trigger repeated node updates and saves.
Autosave can send only changed nodes and connections while checking the board revision. The server still stores a complete snapshot. Current limits are 2,000 nodes, 8,000 connections, and 1,500,000 bytes of saved board data, checked independently. Long prompts or many versions can reach the byte limit first; an overflow error preserves your draft.
Write delivery packages one entry at a time
Episode packages read and write media one entry at a time. Browsers that support writing directly to a file can save the ZIP to your chosen location and abort unfinished output on cancellation or failure. Other browsers retain ordinary downloads, which still hold the final ZIP in memory.
Confirm the adopted shot versions and dialogue before exporting. Packaging does not review the episode or treat every candidate result as a final take.
Share processing resources across users
Idle and background tabs synchronize less often while active node selections keep renewing their locks. Generation submission, speech, result archiving, uploads, and document conversion have separate concurrency limits. Background conversations are claimed fairly across users so one person's queue cannot continually take every slot.
A paid request that cannot start because capacity is full receives a retry message before a charge is reserved. Uploads proceed one file at a time and preserve their upload identifier across retries. Document previews and exports share bounded conversion resources, so a long export can still cause waiting.
What we have checked
Mixed-media browser checks passed at 500 and 1,000 nodes. Isolated protocol tests also passed with 25, 50, and 100 distinct sessions, using mock storage and provider transport without paid generation. These checks verify loading and scheduling behavior; production capacity and responsiveness across devices and networks still require separate measurement.
One hundred people online across the service is a different load from one hundred editors in one project. The direct-collaborator limit remains 25 per project.
Last updated: