v1.4.0

kyutai-labs/hibikiv1.4.0Apr 1, 2026by shrimbly

AI Summary

Adds audio-to-video generation capabilities and a new array batch mode for sequential workflow execution.

Key Highlights

Video generation nodes now accept audio inputs for audio-driven workflows.
New array batch mode for sequential generation from array items.
Undo/redo memory usage optimized to prevent bloat.
Output gallery now correctly preserves all batch-generated images.

New Features

Audio-to-video generation
Array batch mode

Full Release Notes

## Added

- **Audio-to-video generation** — Video generation nodes now accept audio inputs, enabling audio-driven video workflows with handle rendering, connection validation, model discovery, and drop-menu wiring
- **Array batch mode** — New batch execution mode that sequentially generates from all items in an array, with shared helper logic across all execution entry points

## Fixed

- **Undo/redo memory bloat** — Eliminated excessive memory usage caused by deep-cloning base64 image blobs in history snapshots; clipboard and snapshot operations now use a string-preserving clone
- **Cancellable batch execution** — Wired AbortController into `regenerateNode` so batch runs can be properly cancelled
- **Output gallery correctness** — Output gallery now reads fresh node data to preserve all batch-generated images
- **Array batch behavior** — Batch mode is now derived dynamically from the source node rather than being statically configured
- **UI polish** — Normalized button sizes in array node headers and repositioned batch/auto-route controls inline with split rows