fluvie 0.1.10

Write a video like a Flutter screen: declarative widgets in, a real video file out. No timelines to scrub, no frame math.

Changelog

The format follows Keep a Changelog and the versions follow Semantic Versioning.

0.1.10 - 2026-06-24

Added

• Generative-media prerender seam. Declare AI-generated visual and audio sources with GenerativeMedia / GenerativeAudio (GenerativeSource, GenerativeKind, GenerativeCarrier) and resolve them before the frame loop through the new GenerativeResolver contract (generativeResolverProvider, defaulting to NoGenerativeResolver). collectGenerativeSources plus the now generative-aware media, clip, clip-audio, and audio collectors fold each produced file back in as a plain MediaSource.file / AudioSource.file, so a generated video (with Veo 3 audio) and generated music mix and stay in sync like local assets. The render orchestrators run generation before media pre-resolution; the real backend lives in fluvie_ai.
• The default FFmpeg render now mixes a clip's embedded audio. render() probes every clip and the encoder stages the audio of any clip that carries a track (regular Clip or a generated video such as Veo 3), delayed to its scene window. ClipMetadata gained hasAudio (the probe reports it) so a silent clip is never given a broken audio map.
• render() and renderToSandbox() accept onGenerativeProgress to surface per-asset generation progress (GenerativeProgress) before the frame loop.

0.1.9 - 2026-06-23

Lockstep maintenance release; demo (mobile layout) and CI fixes only, no library changes since 0.1.8.

0.1.8 - 2026-06-23

Lockstep maintenance release; demo and deploy fixes only, no library changes since 0.1.6.

0.1.7 - 2026-06-23

Lockstep maintenance release; demo and deploy fixes only, no library changes since 0.1.6.

0.1.6 - 2026-06-23

Lockstep maintenance release; no functional changes since 0.1.5.

0.1.5 - 2026-06-23

Added

• Non-fatal spec diagnostics: an unknown spec property is surfaced on the default render path instead of being silently dropped, backed by the JSON-schema variants and defs.
• A pluggable ClipFramePreparer contract so a host can pre-resolve clip frames without ffmpeg.

0.1.4 - 2026-06-22

Added

• In-browser rendering paints declared image media. A dart:io-free WebImageMediaResolver (selected by mediaResolverProvider on web) resolves and decodes asset, network, and memory Image sources before the frame loop, and renderToSandbox accepts an injected resolver. Audio, snapshots, and captions are not supported on web yet and fail with a clear typed error.
• On-device video Clip support, shared across desktop and web. A ClipResolveCache mixin probes a clip, plans the source frames its scene window reads (planClipFrames), extracts and caches them, and serves them synchronously per frame; a clip pre-pass (preResolveCompositionClips) runs before the frame loop in every render entry point. The browser decodes through an injected WebClipDecoder (WebCodecs), wired by webClipDecoderProvider; with no decoder a clip fails with a clear typed error.
• Shared on-device render primitives so the mobile and web encoders stay thin and consistent: RenderPhase + RenderProgress/RenderProgressCallback, frameCountFor, runGuarded (cleanup that never masks the render error), runStage (stamps a stage onto a thrown FluvieRenderException), DisposableResolver, and resolverScope.
• FluvieRenderException.stage records whether a failure happened while capturing or encoding.
• Media resolvers (MediaRepository, WebImageMediaResolver) now dispose() their decoded images, freeing native memory instead of waiting for GC.

0.1.3 - 2026-06-22

Added

• Off-screen renders can pre-resolve a composition's declared media: an injected MediaResolver decodes and caches the Image/Clip sources before the first frame, so the on-device mobile and web renderers can paint declared media.

0.1.2 - 2026-06-21

Added

• On-device audio: audio sources and a resolved audio-track model, mixed through the deterministic encode path, including looping audio beds.
• A render-sandbox seam with file and in-memory backends, so a composition can be captured to disk or to memory for the on-device mobile and web renderers.

Changed

• The render service stages audio and frames through the sandbox and runs a dedicated frame-capture loop, in place of the previous single render loop.

0.1.0 - 2026-06-20

The first public release. The API is feature-complete and well tested; the 0.x line means it can still change before 1.0, so pin a version.

Added

• The declarative composition model: Video, Scene, Scene.sequence, Scene.centered, Background.*, and the Defaults cascade.
• The timing engine: Time extensions, Anchor, Trigger, frame resolution, and a deterministic resolved timeline.
• Animation: .animate([...]), the preset menu, Animation.from/to/keyframes, springs, Stagger, and Repeat.
• Elements: Text, Typewriter, Counter, Image, Clip, Timeline, Chart, Code, Terminal, Markdown, Mermaid, WebView, Html, annotations.
• Effects: pixel post-effects, seeded particles, maskWipe, parallax, ambient motion, and shader animations.
• Transitions, shared-element morphs, and a scene-wide Camera.
• Audio and captions: Audio.music/sfx, beat detection, reactive inputs, SRT and VTT captions.
• resolveAudioMix: the encoder-neutral ResolvedAudioMix/ResolvedAudioTrack view of a Video's audio (the same timing math the FFmpeg mix uses) for custom encoders such as fluvie_mobile_encoder and fluvie_web_encoder.
• Looping music beds in the FFmpeg path (-stream_loop), so a short bed fills the whole video on the desktop, server, and web. The dart:io-free renderToSandbox accepts audio through an injected AudioByteLoader and stageResolvedAudioToSandbox, the seam the in-browser encoder mixes through; NetworkAllowlist is now exported for on-device network audio.
• The power layer: FluvieTheme, Adaptive and multi-aspect render, templates, the FrameBuilder escape hatch, and Export.* modes.
• Content-hash media caching, so an unchanged frame is read from cache rather than redrawn.