Skip to content

Changelog

Notable changes, newest first. Full notes for each release are on GitHub Releases. Versions follow semantic versioning. The desktop app (desktop-v* tags) installs the newest engine from PyPI, so it is not listed here.

2.16.2 — unreleased

  • Fix: --llm-provider on create and draft no longer reuses the configured provider's model, server URL, token or think setting. For example, --llm-provider claude over an Ollama config sent Claude the Ollama model name and failed.
  • Live tests for Presenton and the Kokoro voice.

2.16.1 — 2026-10-17

Documentation, website and test-suite housekeeping (roadmap Phase 11). No behaviour changes.

  • Shorter README (it is the PyPI page) with links that work on PyPI.
  • New provider reference, changelog and user stories; user guide rewritten as a reference; obsolete planning notes archived.
  • Docs and website no longer claim FFmpeg must be installed.
  • Website moved to slidestream.borck.education; accessibility fixes.
  • Test files split by topic; tests for PDF and Word import; CI checks the bundled ffmpeg and a coverage floor.

2.16.0 — 2026-10-16

  • Fix: a # comment inside a fenced code block no longer starts a new slide; fenced code and tables are parsed properly. enrich keeps them.
  • Fix: video renders without a system ffmpeg, using the copy bundled with moviepy.
  • Fix: doctor no longer counts LLM calls for scripted slides.
  • Kokoro is the local voice floor: configured voice, then Kokoro (offline), then gTTS (ADR-0012). New TTS provider auto and settings.voice_floor.
  • Code on slides is drawn with syntax colouring; tables keep their rows.
  • Experimental, CLI-only Presenton designed decks: slide-stream presenton, create --template, --preserve-style, --slides-pdf, with a fidelity report and rebuilt code slides (ADR-0009, ADR-0013).

2.15.25 — 2026-09-15

  • Security: host validation on every local-mode request, registered mascot names only, authenticated uploads before reading bodies, streaming size limits, explicit trusted proxies (SLIDESTREAM_TRUSTED_PROXIES); Docker binds to loopback by default.
  • Rendering: enriched images reused only while the deck is unchanged; existing speaker notes kept by default; .pptx export and dry runs skip FFmpeg, TTS and avatar setup.
  • Editable PowerPoint output from create (narration in speaker notes).

2.15.0 – 2.15.24 — 2026-08-26 to 2026-08-31

Mostly web UI and hosted demo work:

  • Three output modes in the web UI; demo trimmed.
  • Live render progress and a render trace panel; version in the footer.
  • One-page teaser mode for the hosted demo, with idea-to-deck drafting.
  • Ollama bearer auth and a think control for reasoning models.
  • Editable enhanced slides, images without text, image style option; image and notes generated concurrently per slide.
  • Bullets beside generated artwork in video slides (2.15.24).
  • Hermetic tests; avatar polling resilience.

2.14.1 — 2026-07-25

  • Presenter and transition controls in the web UI's Configure step.

2.14.0 — 2026-07-25

  • Web UI becomes a Source → Deck → Configure → Render workflow.
  • create uses images a deck already references, so enrich → create works.
  • --avatar-slides, --reuse-avatar, --transition.
  • ComfyUI renders retry once on a GPU out-of-memory error.

2.13.0 — 2026-07-21

  • slide-stream draft: turn a text, Markdown, PDF, Word or PowerPoint file into a Markdown deck with an LLM.
  • Clean errors instead of tracebacks when an LLM cannot start.

2.12.0 — 2026-07-21

  • [all] install extra.
  • doctor reports missing provider packages with the install command.
  • Stage directions such as [pause] are removed before narration is voiced.
  • init writes a minimal starter; init --full the complete reference.
  • Documentation site published.

2.11.0 — 2026-07-19

  • enrich --notes fill|all: AI presenter notes in the PowerPoint.
  • doctor and create --dry-run: preflight estimates and warnings.
  • Web UI preflight and PowerPoint output.
  • Opt-in live end-to-end tests.

2.10.0 — 2026-07-18

  • wan-s2v avatar provider: animates mascots and photos from narration audio, with no face detector.

2.0.0 — 2025-06-25

  • Layered YAML configuration and the provider architecture (LLM, images, TTS). AI image generation and premium voices.

1.x

  • 1.1: PowerPoint input with speaker notes.
  • 1.0: first release, Markdown input.