v0.2.5

This release focuses on Scope LTX-2 improvements and a sweep of bug fixes and stability improvements.

Highlights

• Scope LTX-2 Improvements — Skip CPU offload on high-VRAM GPUs (via the upstream scope-ltx-2 bump), A/V buffers flushed on pipeline discontinuity to fix desync after hold/resume, and the first pipeline call now waits for stream-wired prompts so the desired prompt lands on generation #1 instead of #2.
• Bug Fixes & Stability — Workflow media bundling for portable exports, pipeline failures surfaced in the UI as toasts and inline overlays, fail-fast LoRA compatibility validation, local Syphon/NDI/Spout sinks restored when connected to cloud, connected param-edge values honored at session start, and trickle JSONL reader sizing for workflow imports with embedded media.

What's Changed

• LTX-2 — scope-ltx-2 bump (skip CPU offload on high-VRAM GPUs), A/V flush on pipeline discontinuity, and gating the first pipeline call on stream-wired prompts by @leszko in #1034, #1031, and #1032
• Workflow portability — bundle referenced images/audio/video into exported workflow JSON via base64 with idempotent extract, and raise/tune the trickle JSONL reader cap so imports with embedded assets succeed by @leszko in #1027, #1035, and #1037
• Cloud sinks — restore local Syphon/NDI/Spout output when connected to cloud (strip hardware sinks from the runner graph, tee cloud-rendered frames into local senders) by @leszko in #1029
• Pipeline error surfacing — structured pipeline_error notifications with toast + inline error overlay on sinks and Video Output, plus fatal-error escalation by @leszko in #1026
• LoRA validation — fail fast on incompatible LoRAs with a typed error naming the mismatched layer, and warn in the LoRA picker when a configured path is missing by @leszko in #1028
• Graph param edges — use connected param edge values in the initial session payload so producer-node values aren't lost to stale defaults on first generation by @leszko in #1030
• Tests — unblock AudioProcessingTrack recv() silence tests by @leszko in #1022

Full Changelog: v0.2.4...v0.2.5

Preview Build

Preview build from main branch (2026-05-15 11:48 UTC).

Commit: 2aced4d

This is an automated preview — not a versioned release. Do not use for production.

v0.2.4

This release introduces the Node Abstraction, LTX-2 improvements, and a sweep of UX polish and bug fixes.

Highlights

• Node Abstraction — You can now ship any custom node for Scope as a plugin. Sample plugins included for a YouTube input source and a local LLM Prompt Enhancer.
• LTX-2 improvements — Real-time pacing controls, ~18× faster model loading, faster prompt changes, and Ampere GPU compatibility — see scope-ltx-2 for full details.
• Bug Fixes & UX Improvements — Graph editor edge unattach / quick-connect, in-node audio playback, a "?" cheat sheet, install hints for unavailable integrations (NDI/Spout/Syphon), recording reliability, and many smaller fixes.

What's Changed

• Node abstraction by @leszko in #903, #904, #906, #980, #987, #988, and #1002
• Graph editor UX — edge unattach, quick-connect drag menu, in-node audio playback, "?" cheat sheet, and unavailable-integration install hints by @BuffMcBigHuge and @hthillman in #1012, #1013, and #1014
• LTX-2 plugin bumps, LTX-specific cloud asset upload, and audio-flush fix on prompt change by @leszko, @j0sh, and @BuffMcBigHuge in #981, #983, #1000, #1010, and #1015
• Recording and WebRTC reliability — ScopeMediaRecorder replacement, incoming-timestamp preservation, dedicated record-node queues, A/V start-sync delay, and local media upload to cloud by @j0sh in #985, #992, #993, #994, and #1001
• Cloud and Livepeer — gateway update, readiness check before websocket accept, source routing, manifest-ID logging, Cloud Relay removal, and resolve-proxy fix by @j0sh and @leszko in #972, #978, #979, #989, #991, and #1020
• OSC parameter coverage — expose any node param via OSC with per-param opt-in and forwarded callback updates by @hthillman and @j0sh in #996 and #1008
• Bug fixes and polish (VACE ref propagation, peft pin for torchao, Spout pixel-format flip, friendlier HF /blob/ LoRA error, reference-image log spam, video source toggle and credits display, source-less workflow start, packaged Electron commit hash, post-session NPS survey) by @leszko, @j0sh, @hthillman, and @gioelecerati in #877, #975, #977, #986, #995, #997, #1003, #1005, and #1007

Full Changelog: v0.2.3...v0.2.4

v0.2.3

This release introduces Synced Audio + Video, the move to Livepeer Network as the default cloud backend, and Cloud Billing.

Highlights

• Synced Audio + Video — PTS now rides with every video and audio packet end-to-end, with cumulative-drift pacing that keeps output steady over long runs.
• Cloud switched to Livepeer Network — Cloud mode now runs on the Livepeer Network by default, with audio and multi-track support.
• Cloud Billing — Credit-based billing with subscription management.

What's Changed

• End-to-end timestamp propagation, audio sync, and Livepeer output pacing by @j0sh in #864, #892, #901, #902, #919, and #941
• Livepeer cloud default and connection reliability by @j0sh in #893, #894, #907, and #951
• Cloud credits and billing by @gioelecerati in #954
• Graph mode visual redesign and brighter background grid by @gioelecerati in #886 and #961
• Bug fixes and polish (headless recording, Safari sink stats, Syphon flip/reload, macOS Sink fullscreen, Livepeer runner Kafka parity, session-start params, LTX-2 plugin pin) by @j0sh, @hthillman, @emranemran, and @leszko in #910, #913, #927, #928, #930, #934, #949, #956, #958, #959, #960, #964, #967, and #969

Full Changelog: v0.2.2...v0.2.3

v0.2.2

This release introduces Multi Source / Multi Sink graphs, an Audio In node, and a number of bug fixes across graph mode, cloud, and recording.

Highlights

• Multi Source / Multi Sink — Workflows can now contain multiple source and sink nodes in a single graph. Each source routes to its own input track, each sink shows its own per-sink stats (FPS / bitrate), and per-node recording lets you capture each sink independently. Works in local, cloud, and headless modes.
• Audio In Node — A new graph node for audio inputs. WAV / MP3 / FLAC / OGG files can be uploaded to the asset library, picked from a new audio media picker, and forwarded into pipelines via an audio complex field.
• Bug Fixes — Many fixes around graph mode, cloud relay routing, NDI / Syphon sources, recording timestamps, and parameter sync.

What's Changed

• Multi Source / Multi Sink across WebRTC, cloud, frontend graph editor, MCP, and headless mode by @leszko in #797, #798, #799, #800, #809, #854, #855, #863, #866, #869, #884, #885, #887, and #888
• Audio-In node and audio assets by @BuffMcBigHuge in #803
• Headless MPEG-TS streaming endpoint and PTS fix by @j0sh in #851 and #852
• Keyboard shortcuts, fullscreen button on Sink node, and live output clarity in onboarding by @hthillman in #847 and #848
• Sync external parameter updates (OSC/REST/MCP) to graph mode, persistent permission grant, and string primitive auto-send by @gioelecerati in #833, #865, and #867
• LoRA merge strategy view, integer handling, and prompt management in graph mode by @BuffMcBigHuge in #860, #861, and #862
• Unify cloud connect and surface connection errors by @j0sh in #873
• Refresh workflow builder pipelines after cloud connect and resolve/plugin listing perf by @leszko in #853 and #857

Full Changelog: v0.2.1...v0.2.2

v0.2.1

This release upgrades the default starter workflow to LTX 2.3, adds new Workflow Builder features including a scheduler node and improved canvas UX.

Highlights

• LTX 2.3 Default Workflow — The starter workflow now uses LTX 2.3 text-to-video (replacing Paint Blobs), so new users can generate video without a camera input
• Scheduler Node — A new browser-side scheduler node that fires named trigger outputs at configurable time points, with orange trigger connectors and edge flashing on fire
• Canvas UX Improvements — A floating [+] button and an empty-state placeholder make it easier to add nodes when starting from scratch in the Workflow Builder

What's Changed

• Replace Paint Blobs with LTX 2.3 and enrich onboarding_completed event by @hthillman in #816
• graph: Add frontend-only scheduler node with trigger type system by @gioelecerati in #812
• graph: Add [+] button and empty state placeholder to canvas by @hthillman in #822
• fix: pipelines getting stuck in graph mode by @gioelecerati in #832
• cloud: Pre-install ltx-2 plugin in cloud Docker image by @leszko in #828
• Livepeer network improvements: payments, text mode, local recording, and LoRA support by @j0sh in #769, #810, #776, #774, #787, and #802
• feat: add MCP server recording support and cloud connection tools by @leszko in #794
• feat: add local cloud dev server for testing cloud relay flow by @leszko in #793
• feat: pass analytics tokens to Docker and Electron CI builds by @leszko in #792

Full Changelog: v0.2.0...v0.2.1

v0.2.0

This release introduces the Workflow Builder, a new graph-oriented UI for building and customizing video pipelines, along with a guided onboarding experience for new users.

Highlights

• Workflow Builder — A new graph-based visual editor for building and customizing pipelines, now the default mode. Connect and configure nodes for models, processors, audio, and tempo with a redesigned toolbar and sample video previews
• Onboarding Experience — A guided flow for new users covering account setup, model downloading, and pipeline configuration

What's Changed

• Workflow Builder: graph-based frontend, toolbar redesign, audio/tempo nodes, prompt blending, sample videos, play/pause UX, and bug fixes by @gioelecerati and @leszko in #702, #740, #742, #745, #748, #763, #764, #775, #777, #778, and #779
• Add onboarding flow for new users by @hthillman, @gioelecerati, and @leszko in #697, #755, and #766
• Livepeer Cloud Mode by @j0sh in #738
• feat: analytics opt in by @hthillman and @gioelecerati in #728 and #768
• fal: Set request timeout to 120 minutes by @j0sh in #741
• fix: recording not working with remote inference by @gioelecerati in #753
• Fix security issues by @leszko in #762
• feat: descriptive model load & minor changes by @gioelecerati in #767

Full Changelog: v0.1.9...v0.2.0

v0.1.9

This release adds audio output support for pipelines and fixes a VACE regression introduced in 0.1.8.

Highlights

• Audio Out Support — Pipelines can now return audio alongside video, streamed over WebRTC with a mute/unmute toggle in the frontend by @BuffMcBigHuge, @ryanontheinside, and @leszko in #534, #718, and #719
• Fix VACE Regression — Fixed tempo sync modulation interfering with VACE noise scale, a regression introduced in 0.1.8 by @BuffMcBigHuge in #716

What's Changed

• Audio Support for Scope Rework by @BuffMcBigHuge in #534
• Fix: tempo sync modulation interfering with VACE noise scale by @BuffMcBigHuge in #716
• fix(pipeline_processor): idle backoff and prepared state for audio-only outputs by @leszko in #718
• Flush audio buffer on prompt change for immediate TTS playback by @leszko in #719

Full Changelog: v0.1.8...v0.1.9

v0.1.8

This release adds beat-synced parameter modulation with Ableton Link and MIDI clock support, an MCP server for AI-assisted pipeline control, and DMX Art-Net input.

Highlights

• Tempo Sync & Beat Modulation — Lock pipelines to Ableton Link or MIDI clock. Quantize parameter changes to beat/bar boundaries, modulate parameters (noise scale, denoising steps, etc.) with waveforms synced to the beat, and automatically cycle prompts on beat or bar boundaries by @BuffMcBigHuge, @ryanontheinside, and @leszko in #703
• MCP Server — AI assistants can now programmatically manage pipelines, control parameters, capture frames, and drive headless sessions via the Model Context Protocol by @ryanontheinside in #660
• DMX Art-Net Input — Real-time DMX input with configurable channel-to-parameter mappings, import/export, and live SSE updates by @thomshutt in #657
• FPS Fix — Removed FPS-based throttling from the input source receiver loop, fixing frame delivery stalls and improving streaming smoothness by @leszko in #701

What's Changed

• Tempo Sync Modulation by @BuffMcBigHuge in #703
• Add MCP server for AI-assisted pipeline control by @ryanontheinside in #660
• DMX Art-Net input support by @thomshutt in #657
• fix: remove FPS-based throttling from input source receiver loop by @leszko in #701
• Add graph mode backend by @leszko in #664
• Use daydream API to check if plugin is allowed by @mjh1 in #646
• fix: invalidate plugin cache when (un)installing by @mjh1 in #666
• feat: improve cloud button discoverability and header icon visibility by @emranemran in #654
• Make Settings dialog box size match Plugins by @thomshutt in #661
• Remove local filesize limits by @thomshutt in #658
• Fix displaying new loading text by @mjh1 in #659
• Prefix logs with connection ID by @mjh1
• ci: speed up PR deploys by decoupling from standard image build by @mjh1 in #649

Full Changelog: v0.1.7...v0.1.8

v0.1.7

This release adds MIDI controller support for real-time parameter control, native Workflow export to daydream.live, WebRTC stability improvements, and performance optimizations.

Highlights

• MIDI Controller Support - Full WebMIDI integration with device discovery, mappable pipeline and plugin parameters, persistent profiles, and learning mode by @gioelecerati and @jamesdawsonWD in #537

What's Changed

• feat: MIDI support & MIDI mappable parameters by @gioelecerati and @jamesdawsonWD in #537
• Add native Workflow export to daydream.live by @thomshutt in #636
• Cache pipeline schemas and plugin list responses by @leszko in #645
• fix: batch-level FPS tracking for smooth frame delivery by @ryanontheinside in #617
• Fix WebRTC VP8 decode errors and PLI keyframe requests by @emranemran in #614
• Close WebRTC session when cloud WebSocket disconnects by @emranemran in #638
• Fix vace_context_scale not sent in initial params without ref images by @ryanontheinside in #630
• Fix download UI blocking on empty model files by @ryanontheinside in #610
• Optional logging of OSC messages by @thomshutt in #629
• Account for not found error in fal app delete by @mjh1 in #634
• Output app logs from e2e as a github artifact by @mjh1 in #608

Full Changelog: v0.1.6...v0.1.7