skills/digitalsamba__claude-code-video-toolkit
1
0
Fork 0
mirror of https://github.com/digitalsamba/claude-code-video-toolkit.git synced 2026-09-18 19:41:13 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
main
digitalsamba__claude-code-v…/_internal/CHANGELOG.md
T
Conal Mullan e24eacbc27 Release v0.20.1: fix NameError crashes in the Codex/Kiro migrations and dewatermark --setup 
Patch release. v0.20.0 shipped two entry points that aborted with a NameError
before doing any work:

- `migrate_to_codex.py` / `migrate_to_kiro.py --force` — undefined `write_text`,
  dropped by the _migrate_common extraction (#85, thanks @Anai-Guo)
- `tools/dewatermark.py --setup` — undefined `get_runpod_config` (#86)

Also consolidates the duplicated migrate helpers and adds a Lint Python CI gate
for undefined names, which had no coverage before.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-08 22:08:20 +01:00

40 KiB
Raw Permalink Blame History

Toolkit Changelog

All notable changes to claude-code-video-toolkit.

Releases are automated via GitHub Actions. See .github/workflows/release.yml.

Unreleased

2026-09-08 (v0.20.1)

Patch release. v0.20.0 shipped with two tools that aborted with a NameError before doing any work; both are fixed here.

Fixed

  • Codex and Kiro migrations (scripts/migrate_to_codex.py, scripts/migrate_to_kiro.py) — write_text() was dropped by the _migrate_common extraction without being moved to the shared module, so --force aborted with NameError: name 'write_text' is not defined on the first real write. --dry-run returned before that path, which is why it went unnoticed. Restored in _migrate_common, and the two orphaned function bodies the same refactor left stranded are gone. Thanks @Anai-Guo (#85).
  • tools/dewatermark.py --setup — setup_runpod() called get_runpod_config(), which is defined nowhere in the repo, so the documented one-time setup crashed immediately. Now uses the load_dotenv() + os.getenv("RUNPOD_API_KEY") pattern tools/upscale.py already uses (#86).
  • _migrate_common.load_mapping returned raw JSON instead of the normalized shape, so any caller would KeyError on mapping["skip_commands"]. It was unreachable only because both scripts shadowed it with a local copy. The real version is now shared and the duplicates (plus a duplicate find_repo_root and an unused yaml_quote import) are gone — verified byte-identical migration output before and after (#86).

Added

  • Lint Python CI workflow — gates on undefined names (ruff --select F821,F811,F822,F823) across scripts/ and tools/. Deliberately narrow: only rules where a hit is a real defect, not style. Run against the v0.20.0 tree it reports 26 errors, including both write_text sites. There was previously no Python lint in CI (#86).

2026-08-31 (v0.20.0)

Added

  • SoulX-FlashHead talking head (tools/soulx.py + docker/modal-soulx/) — the toolkit's default talking head generator (Soul AI Lab, Apache 2.0, 1.3B), Modal-only. Preserves the input aspect ratio, so 16:9 presenter images come back 16:9 with no --preprocess workaround, and --size snaps to the model's latent grid while keeping the aspect. It is the default because identity holds over a long take: measured at 97% of frame-zero sharpness at 70s, flat across 72 segments, so per-scene generation is a choice rather than a workaround. ~$0.0024 per second of output against SadTalker's ~$0.0014. Decision table in CLAUDE.md, full detail in docs/soulx.md. (#81)

Changed

  • docker/modal-soulx/ keeps its weights in a Modal Volume, unlike the other Modal apps which bake them into the image. Measured: redeploy after a code change is ~2.3s against 79-385s baked, while cold start and generation speed are unchanged. Needs a one-off modal run …::populate_weights (14.7 GB). The settled apps stay baked.
  • Upstream repo ref and both model revisions in modal-soulx pinned by SHA — weights in a Volume aren't tied to the image, so nothing else prevents drift. (same lesson as #71/#74)
  • NarratorPiP honours its objectPosition prop, which was declared and documented but silently ignored by a hardcoded value, and gains an objectFit prop (default contain, unchanged behaviour).
  • tools/cloud_gpu.py learned soulx, including its GPU tier, so its jobs report a cost estimate.

Removed

  • EchoMimicV3 (tools/echomimic3.py, docker/modal-echomimic3/, docs/echomimic3.md), added earlier in this unreleased cycle and never part of a tagged release. Identity drift made it unusable for real narration: on a controlled A/B — same photo, same 80s audio, same 544x736 — it fell to 50% of frame-zero sharpness by 70s, collapsing to a featureless smear, where SoulX-FlashHead held 97%. The failure is absorbing rather than gradual, since each segment re-anchors on the previous one's output, so it could not be tuned out; it implied a ~30s render ceiling that shaped the surrounding design. SoulX supersedes it on every axis that mattered — drift, cost (~3.7x cheaper), wall clock (~3-6x faster) — while keeping aspect-ratio preservation. (#81, closes #80)

2026-08-27 (v0.19.0)

Added

  • uv is the install path — pyproject.toml + uv.lock; uv sync (extras: whisper, modal, youtube) and every doc/tool now says uv run tools/…. tools/requirements.txt stays as the pip fallback. Contributed by @AsharibAli (#33, rebased as #59).
  • Kiro CLI support — scripts/migrate_to_kiro.py installs the toolkit's skills/commands for Kiro with steering-file generation; see docs/kiro.md. Contributed by @homeo26 (#55). Follow-up refactor tracked in #58.
  • Version maintenance (closes #23) — scripts/check_versions.py (toolkit-wide Remotion/Python/toolkit staleness, wired into /versions); tests/render-baseline/ + scripts/render-baseline.mjs + render-baseline.yml (same-runner before/after render diff that gates Remotion bumps in CI); scripts/bump_remotion.py as /versions --bump-toolkit; grouped monthly Dependabot for the remotion group. (#63, #64, #65)
  • sync_timing.py generic mode — pairs any *AudioFile with *DurationSeconds so ticket-driven and future config layouts work without tool changes. (#62)
  • Community add-ons README section for integrations that live in their own repos (first entry: VOICEPEAK offline Japanese TTS, #42), plus CONTRIBUTING guidance for single-vendor integrations and automated/AI-generated PRs. (#60, #61)

Changed

  • Remotion 4.0.425 → 4.0.518 across all templates, examples, showcases and the baseline harness — first bump gated by the render-baseline check (all frames identical). (#67)
  • All Remotion projects now pin the same exact version (four stragglers were on 4.0.381/4.0.382/^4.0.0).

Housekeeping

  • External PR queue triaged: MiniMax video/music and Atlas Cloud bot PRs closed under the new policy; Edge-TTS (#46) has review feedback pending.
  • Stale branches removed; unmerged experiments preserved as archive/* tags.

2026-08-26 (v0.18.0)

Added

  • YouTube publishing — tools/youtube_upload.py + /publish command. OAuth login (--auth), private-by-default upload, scheduled go-live (--publish-at), thumbnails, --dry-run validation, 403s classified by API reason. Setup guide in docs/youtube-upload.md; google-* deps are opt-in in tools/requirements.txt. (#29, #37, #38, #39, #40)
  • 60db TTS provider — third voiceover.py --provider 60db backend plus standalone tools/sixtydb_tts.py (synthesize / stream / websocket transports, --list-voices). Unified 0-1 voice settings auto-convert to 60db's 0-100 scale; redub.py --tts-provider 60db. Brands carry a sixtydb block in voice.json. (#27)
  • _internal/narrative-patterns.md — Discovery Story structure for sprint reviews (reorder by narrative arc, turning-point slides, honest bridge framing).
  • docs/ltx2.md Windows gotchas + LTX-2 Modal config wired into verify_setup.py. (#51, #53)

Changed

  • Remotion skills sync now tracks upstream's new per-topic layout via skills/remotion-best-practices/ (router skill embedding markup/maps/captions/render/saas/studio/docs/upgrade). Workflow fails loudly on the next re-layout and carries the upstream skill version in PR titles. First sync from the new layout: 39 → 137 files. (#56)
  • Modal image-edit requests A100-80GB (OOMed on 40GB).
  • .gitignore covers scratch/, Playwright run artifacts, recursive assets/voices; duplicate project copies under projects/ untracked.

Fixed

  • Weekly Remotion skills sync had failed every run since 2026-07-13 (upstream directory moved).
  • verify_setup.py accepts both modal CLI JSON key casings when detecting deployed apps.

Registry

  • Added missing entries: align_captions tool, quick-spot / data-viz-chart / ds-crt-stinger / sky-blue-short examples. the-space-between marked showcase-only (no source directory).

2026-06-10 (v0.17.0)

Added

  • concept-explainer-short template — 9:16 vertical TikTok/Reels/Shorts explainers. First pure Python/moviepy template: scenes.json → gen_vo.py (per-scene TTS) → gen_captions.py (whisper forced-align) → build.py (Ken Burns stills, boomerang clips, karaoke caption pills, ducked music). (#31)
  • examples/sky-blue-short — 52s showcase of the template with all assets committed.
  • TTS pacing QC — tools/pacing.py; voiceover.py and qwen3_tts.py report per-take wpm + pacing label and accept --max-wpm for a pitch-preserving atempo clamp. Fixes cloned voices inheriting a rushed pace from their reference. (#30)

Docs

  • README restructured to lead with finished videos; "Using with Codex" moved to docs/codex.md.

2026-06-08 (v0.16.0)

Added

  • Ideogram 4 text-to-image — tools/ideogram4.py + ideogram4 skill. Structured JSON captions for legible in-image text and exact brand colours; hosted paid API. FLUX-vs-Ideogram decision rule added to CLAUDE.md. (#28)
  • Qwen3-TTS v0.2 — VoiceDesign mode (--design-instruct), shared-prompt cloning for consistent multi-scene narration, clone.design brand schema with self-caching reference; tools/align_captions.py (Scribe-based caption timing alignment).
  • Codex migration flow for toolkit skills/commands (#16, thanks @kimhoontae-gogo); thin AGENTS.md.

Changed

  • Remotion pinned to exact versions in lab examples; CLAUDE.md clarifies the <OffthreadVideo> choice vs upstream @remotion/media.
  • Official Remotion skills synced (upstream 277510e). (#17)
  • GitHub Actions bumped to Node 24-compatible versions; docker-build workflow handles multi-commit pushes and skips modal-* dirs.

Fixed

  • Python 3.9 compatibility restored across tools/ via from __future__ import annotations. (#22)
  • modal-upscale basicsr build conflict. (#19, thanks @ATECHPCS)
  • Codex migration name read from registry; no-op skip entry dropped. (closes #20)

2026-04-21 (v0.15.0)

Added

  • LTX-2 style LoRA support with crt-terminal preset.
  • examples/ds-crt-stinger — 6s CRT LoRA + grunged-logo brand stinger.
  • Animated synthwave README banner (showcase/banner).

Changed

  • Official Remotion skills synced (upstream 41a3b06); "Stsrt" → "Start" typo fixed. (#14)

2026-04-09 (v0.14.2)

Added

  • moviepy skill (.claude/skills/moviepy/) — Python video composition for overlaying deterministic text on AI-generated video (LTX-2, SadTalker). Leads with the "trustworthy text" framing: news, trailers, lower thirds, and social captions all need deterministic overlay because AI models can't guarantee readable text. Covers the PIL workaround for moviepy 2.x's TextClip ascender-clipping bug, audio-anchored timelines, common recipes (labels on LTX-2 clips, lower thirds on SadTalker heads, tinted overlays for contrast, split-screen composites, music + VO mixing), 2.x API gotchas, and a genres-where-this-shines table.
  • examples/quick-spot — runnable 15-second ad-style moviepy example. Audio-anchored timeline, PIL text rendering with cross-platform font loading, optional per-scene VO + ducked music, solid-colour backgrounds. Renders with python3 build.py and zero external assets.
  • examples/data-viz-chart — runnable animated time-series chart (real GitHub star history) demonstrating the "matplotlib for data + moviepy for trustworthy text" split that mirrors real news-graphics pipelines. Cache-aware matplotlib step.
  • examples/hello-world — minimal Remotion sprint-review example finally committed. The root README's quick-start callout has been pointing at cd examples/hello-world && npm install && npm run render since Feb 23, but the example had zero git history on any branch. Fixes the broken link for anyone cloning the repo.
  • LTX-2 skill: Stylized Character Cameo — new use case in .claude/skills/ltx2/SKILL.md documenting LTX-2 image-to-video as a SadTalker alternative for non-photoreal faces (fantasy characters, heavy beards, masks, helmets) where lip-sync precision matters less than motion + atmosphere.
  • CLAUDE.md: Audio-Anchored Timelines — new subsection in the Video Timing guide complementing the existing reactive sync_timing.py flow with a prevention-first pattern (generate audio first, anchor visuals to absolute timestamps). Especially useful for single-file moviepy build.py projects.

Changed

  • tools/requirements.txt adds Pillow>=10.0, moviepy>=2.0, matplotlib>=3.7. One install command (python3 -m pip install -r tools/requirements.txt) now covers every Python feature in the toolkit: AI voiceover, image gen, music gen, and the new moviepy examples.
  • Root README quick-start switches to python3 -m pip install — the PyPA-recommended invocation that works identically on macOS, Linux, and Windows and guarantees pip and python3 resolve to the same environment.
  • docs/sadtalker.md adds a "When NOT to Use SadTalker" section pointing at LTX-2 image-to-video for stylized characters, heavy facial hair, masks, and non-frontal angles.
  • examples/README.md splits "Using Examples" into Remotion (npm install flow) and moviepy (python3 build.py flow) sections and adds both new examples to the listing table.

Fixed

  • Hardcoded macOS font paths in examples/quick-spot/build.py and examples/data-viz-chart/build.py — both examples used /System/Library/Fonts/Supplemental/Arial Bold.ttf directly, which would crash with FileNotFoundError on Linux and Windows. Replaced with a platform.system()-keyed fallback chain that tries OS-appropriate sans-serif fonts (Arial on macOS and Windows, DejaVu Sans on Linux, with multiple candidate paths each) and falls back to ImageFont.load_default() if nothing matches, so the examples never crash on unusual setups. Both examples now genuinely run cross-platform as their READMEs promise.
  • Latent Pillow gap in tools/flux2.py and tools/image_edit.py — both tools imported PIL inside a try/except ImportError guard with a friendly error message, but Pillow was never actually declared in tools/requirements.txt. Users had to install Pillow manually after hitting the runtime error. Free-rider fix via the moviepy declaration.
  • Guarded imports added to both moviepy example build.py files — matches the existing flux2.py / image_edit.py friendly-error pattern. Prevents bare ModuleNotFoundError tracebacks for users who run the examples before installing dependencies.

Docs

  • README examples table adds q2-townhall-longarm-ad (Super Bowl-style launch ad with LTX-2 animated Lugh cameo) and q2-townhall-stars (GitHub star history time-lapse) as finished demo entries dated 2026-04-08.
  • moviepy skill added to the README's skills table and _internal/toolkit-registry.json.

2026-04-04 (v0.14.1)

Added

  • tools/chain_video.py — Chain LTX-2 clips with visual continuity (each scene uses previous clip's last frame as input). Supports per-scene prompts via JSON, auto-resume of interrupted runs, and structured progress reporting.
  • Structured progress reporting (--progress json) for all cloud GPU tools
  • OpenClaw skill: yieldMs polling — agent stays in poll loop for live progress instead of going silent with background:true
  • OpenClaw skill: style drift prevention — rules to prevent anime/Asian style drift in chained LTX-2 sequences (per-scene prompts + negative prompts)

Changed

  • acemusic cloud API is now default music generation provider (official API, XL Turbo 4B model with 5Hz LM thinking)
  • SadTalker timeouts increased: 20x multiplier + 300s buffer for size=512+gfpgan on A10G
  • Synced official Remotion skills (upstream d5d3955)

Fixed

  • chain_video: glob matching for manually-renamed clips (e.g. chain-05-brigid.mp4)
  • chain_video: 20-min subprocess timeout, stderr in error messages, JSON validation, temp file cleanup
  • chain_video: registered in toolkit-registry.json
  • R2 recovery note for SadTalker client timeouts

2026-03-26 (v0.13.2)

Changed

  • Default cloud provider switched to Modal — All 6 cloud GPU tools (flux2, image_edit, music_gen, qwen3_tts, sadtalker, voiceover) now default to --cloud modal instead of --cloud runpod. RunPod remains available via --cloud runpod.

Fixed

  • Remove unused imports in tools/ltx2.py (json, requests) — flagged by code quality bot

2026-03-25 (v0.13.1)

Added

  • tools/ltx2.py — AI video generation using LTX-2.3 22B DiT model
    • Text-to-video generation from prompts (~5s clips at up to 1024x1536)
    • Image-to-video — animate still images with motion prompts
    • Joint audio+video generation (synchronized ambient audio)
    • Quality presets: standard (30 steps) and fast (15 steps)
    • Valid frame counts: 25-193 frames ((n-1)%8==0), 24fps default
    • Modal deployment on A100-80GB with baked weights (~55GB)
    • Estimated cost: ~$0.23 per 5-second clip
  • LTX-2 skill (.claude/skills/ltx2/) — prompting guide, parameters, video production use cases
  • Openclaw skill updated — LTX-2 added as section 4d (video clips), setup instructions, cost table

Technical Notes

  • torch.inference_mode() required around pipeline call — without it, Gemma text encoder retains ~37GB of autograd activations causing OOM (upstream bug: Lightricks/LTX-2#152)
  • Weights hosted on Lightricks/LTX-2.3 HuggingFace repo (separate from the code repo Lightricks/LTX-2)
  • transformers pinned to 4.57.6 (5.x breaks Gemma3TextConfig.rope_local_base_freq)
  • HuggingFace token required for both LTX-2 (rate limiting) and Gemma 3 (gated model)

2026-03-22 (v0.12.0)

Added

  • tools/music_gen.py — AI music generation using ACE-Step 1.5
    • Text-to-music with precise BPM, key, time signature, and duration control
    • Vocal music with lyrics in 50+ languages
    • Cover / style transfer from reference audio (--cover)
    • Stem extraction — isolate vocals, drums, bass, etc. (--extract)
    • 8 scene presets for video production: corporate-bg, upbeat-tech, ambient, dramatic, tension, hopeful, cta, lofi
    • Brand-aware generation — --brand loads style hints from brand.json
    • --setup creates RunPod template + endpoint via GraphQL
    • Docker image: ghcr.io/conalmullan/video-toolkit-acestep:latest (CUDA 12.8, baked model weights)
    • MIT licensed model — free alternative to ElevenLabs Music with more control
    • ~2-3s inference on GPU (turbo mode, 8 steps)
  • ACE-Step skill (.claude/skills/acestep/) — prompt engineering patterns, lyrics formatting, scene preset guide, video production integration

2026-03-22 (v0.11.1)

Fixed

  • Queue timeout safeguard — All 6 RunPod tools now auto-cancel jobs stuck in queue for >5 min, preventing runaway billing from GPU unavailability
  • Cancel on exit — Jobs are cancelled on RunPod when client polling times out (previously left orphaned jobs in queue indefinitely)
  • R2 upload support — flux2.py and image_edit.py now pass R2 config to handlers (was already supported server-side but never triggered)

Changed

  • flux2 default GPU — Changed from ADA_24 (RTX 4090, frequently throttled) to AMPERE_24,ADA_24 (fallback chain)
  • flux2 setup — --setup-gpu now accepts comma-separated GPU types for fallback
  • RunPod skill — Added API reference (GraphQL queries/mutations, REST endpoints, GPU IDs, R2 CLI patterns)
  • R2 docs — Added operations guide (listing, cleanup, --region auto gotcha) to runpod-setup.md

2026-03-15 (v0.11.0)

Added

  • tools/flux2.py — AI image generation and editing using FLUX.2 Klein 4B
    • Text-to-image generation from prompts (~2.5s fast mode, ~8s quality mode)
    • Image editing with reference images (--input photo.jpg --prompt "...")
    • Multi-image compositing (up to 3 reference images)
    • 8 scene presets for video production: title-bg, problem, solution, demo-bg, stats-bg, cta, thumbnail, portrait-bg
    • Brand-aware generation — --brand digital-samba reads brand.json and injects color palette into prompts
    • Preset + prompt layering — preset provides style/mood, --prompt adds subject context
    • --setup creates RunPod template + endpoint via GraphQL
    • --list-presets shows all available scene presets
    • Docker image: ghcr.io/conalmullan/video-toolkit-flux2:latest (baked model weights, ~15GB)
    • Apache 2.0 licensed model (commercial OK)
  • "The Space Between" — showcase video demonstrating end-to-end AI video creation
    • Avatar generated with flux2, voiced with Qwen3-TTS, animated with SadTalker, composed in Remotion
    • Concept, script, voice, visuals — all AI-generated
    • New video format: video essay (beyond sprint reviews and product demos)

Changed

  • Baked Docker images rebuilt — both flux2 and qwen3-tts images now include model weights baked in, eliminating cold-start model downloads (~30s cold start vs minutes)
  • Updated _internal/toolkit-registry.json with flux2 tool entry, cloud endpoint, scene presets, and showcase example
  • Updated README with flux2 tool, Docker image, demo table entry, and Cloud GPU tools table

2026-02-25 (v0.10.1)

Added

  • tools/sync_timing.py — Audio-to-config timing sync tool
    • Measures actual per-scene audio durations via ffprobe
    • Auto-detects config file and template type (sprint-review v1/v2, product-demo)
    • 3-pass audio-to-scene matching: audioFile field → index → name
    • Comparison table with delta indicators; skips changes < 0.3s
    • --apply updates durationSeconds in-place (creates .bak backup)
    • --voiceover-json accepts voiceover.py output directly
    • Suggests playbackRate adjustments for demo scenes

Changed

  • CLAUDE.md slimmed 44% (861 → 480 lines)
    • Removed catalog data duplicated in toolkit-registry.json (skills, commands, components, transitions, presets, Docker images, duplicate CLI examples)
    • Added cross-references to registry for structured data
    • All workflow guidance, timing knowledge, code patterns, and tool-specific gotchas retained
  • Integrated sync_timing.py into Video Production Workflow (step 7) and TTS drift feedback loop
  • Updated toolkit-registry.json with sync_timing tool entry

2026-02-24 (v0.10.0)

Added

  • Qwen3-TTS integration — Self-hosted TTS via RunPod as free alternative to ElevenLabs
    • tools/qwen3_tts.py standalone CLI with 9 speakers, tone presets, voice cloning
    • voiceover.py --provider qwen3 for per-scene generation
    • Docker image: ghcr.io/conalmullan/video-toolkit-qwen3-tts:latest
    • Temperature/top_p params for expressiveness control
  • /voice-clone command — Record, test, and save a cloned voice to a brand profile
  • sprint-review-v2 template — Composable scene-based architecture for sprint reviews
  • FilmGrain component — SVG noise overlay for cinematic film texture
  • hello-world example — Minimal 25s video, zero config, renders in 2 minutes
  • runpod skill — Cloud GPU setup, Docker images, endpoint management, troubleshooting
  • Remotion skills sync update — 6 new upstream rules (audio-visualization, ffmpeg, light-leaks, subtitles, transparent-videos, voiceover)

Improved

  • Onboarding & developer experience — All API keys now optional, videos render with just Node.js
    • .env.example reorganized with sections, all keys commented out by default
    • README and getting-started.md restructured with "Try It Now" quick start
    • All Python tools show actionable guidance when API keys are missing (add key, use alternative, or skip)
    • voiceover.py catches missing elevenlabs pip package with helpful message pointing to Qwen3-TTS

Changed

  • Updated README with prerequisites table, Qwen3-TTS, Docker images, voice-clone command
  • Updated CLAUDE.md with FilmGrain component and checkerboard transition
  • Updated roadmap metrics and skill status table
  • Fixed cho-oyu demo link; added cortina sprint to demos table

2026-02-19 (v0.9.3)

Added

  • Official Remotion skills — Synced 33 rule files from remotion-dev/skills into .claude/skills/remotion-official/
  • Weekly sync workflow — GitHub Actions checks upstream every Monday and opens a PR if files changed
  • Sync documentation — docs/remotion-skills-sync.md explaining the split and sync process

Changed

  • Remotion skill split — Custom remotion skill trimmed from ~470 to ~160 lines, now covers only toolkit-specific patterns (transitions, components, conventions). Core framework knowledge deferred to remotion-official
  • Updated CLAUDE.md skills table with both remotion skills

2026-01-25 (v0.9.2)

Added

  • Per-scene voiceover generation - New recommended workflow for audio

    • voiceover.py --scene-dir processes all .txt files in a directory
    • voiceover.py --concat joins scene audio for SadTalker narrator
    • Each scene's <Audio> starts at frame 0 within its Series.Sequence
    • Regenerate individual scenes without re-doing the whole video
    • Scene durations match audio naturally (no manual offset calculations)

  • Sprint-review template per-scene audio support

    • audioFile field on DemoConfig, info, overview, summary
    • SprintReview.tsx renders per-scene <Audio> elements
    • Backward compatible: global voiceover track still works

  • Updated /generate-voiceover command

    • Detects public/audio/scenes/*.txt and offers per-scene mode
    • Per-scene is now the default when scene scripts exist
    • Concat option for SadTalker integration

Changed

  • Updated documentation (CLAUDE.md, README.md, getting-started.md)

2025-12-30 (v0.7.0)

Added

  • tools/dewatermark.py - AI-powered watermark removal using ProPainter

    • Cloud GPU processing via RunPod serverless (~$0.05-0.30/video)
    • Local processing for NVIDIA GPU users (8GB+ VRAM)
    • Auto resize-ratio: 1.0 for <30s, 0.75 for <1min, 0.5 for longer
    • Chunked processing for long videos with overlap stitching
    • Cloudflare R2 for reliable file transfer
    • Automated --setup for one-command RunPod configuration
    • Presets: notebooklm, tiktok, sora, stock-br, stock-bl, stock-center

  • tools/locate_watermark.py - Helper tool for finding watermark coordinates

    • Coordinate grid overlay for visual identification
    • Region verification across multiple frames
    • Preset support for common watermarks
    • Requires ImageMagick (brew install imagemagick)

  • tools/notebooklm_brand.py - Post-processing for NotebookLM videos

    • Trims NotebookLM visual outro while preserving full audio
    • Adds custom branded outro with logo and URL
    • Handles freeze-frame bridging when audio exceeds video

  • Docker infrastructure for RunPod serverless (docker/runpod-propainter/)

    • Dockerfile, handler.py, README
    • Public image available for quick deployment

  • Documentation

    • docs/optional-components.md - ML component installation guide
    • docs/runpod-setup.md - Cloud GPU setup guide

Changed

  • ElevenLabs TTS: Added eleven_v3 (alpha) model option
  • Updated CLAUDE.md with dewatermark and locate_watermark documentation
  • Updated .env.example with RunPod and R2 configuration

2025-12-28 (v0.6.0)

Added

  • redub.py --sync - Word-level time remapping for TTS synchronization
    • Uses ElevenLabs convert_with_timestamps API for character-level alignment
    • Uses Scribe word-level timestamps from original audio
    • Builds segment mapping (configurable via --segment-size, default: 15 words)
    • Applies variable speed per segment via FFmpeg filtergraph
    • Fixes audio drift when TTS voice speaks at different pace than original
    • Solves: TTS often starts fast and ends slow, causing 3-4+ second drift

Changed

  • Updated CLAUDE.md with Redub Sync Mode documentation
  • TTS duration now measured via ffprobe (more accurate than timestamp data)

2025-12-28 (v0.5.0)

Added

  • tools/addmusic.py - Add background music to existing videos
    • Generate music via ElevenLabs or use existing audio file
    • Options: --music-volume, --fade-in, --fade-out
    • Works on any video file without requiring a project structure

Changed

  • Updated CLAUDE.md with addmusic documentation

2025-12-28 (v0.4.0)

Added

  • /redub command - Redub existing videos with a different voice

    • Guided workflow for voice selection and transcript handling
    • Supports transcript review/editing before TTS generation
    • Works on any video file without requiring a project structure

  • tools/redub.py - Voice replacement utility tool

    • Extracts audio from video (FFmpeg)
    • Transcribes audio to text (ElevenLabs Scribe STT)
    • Generates new audio with target voice (ElevenLabs TTS)
    • Replaces audio track in video (FFmpeg)
    • Options: --save-transcript, --transcript, --keep-temp, --speed
    • Warns about duration mismatches between original and new audio

  • Utility tools concept - Documented distinction between project tools and utility tools

    • Project tools (voiceover, music, sfx): Used during video creation workflow
    • Utility tools (redub): Quick transformations on existing videos, no project needed

Changed

  • Updated CLAUDE.md with utility tools documentation
  • Updated Python Tools section to include redub

2025-12-20

Added

  • GitHub Actions release automation (.github/workflows/release.yml)

    • Tag-triggered releases (v* tags)
    • Manual dispatch option with version input
    • Auto-generates changelog from commits since last tag
    • Reads version from toolkit-registry.json

  • /versions command - Check dependency versions and toolkit updates

    • Detects Remotion package version mismatches in projects
    • Compares local toolkit version against GitHub releases
    • Offers to fix version mismatches by pinning and reinstalling
    • Documents common issues and prevention strategies

Fixed

  • Remotion version mismatch in digital-samba-free-intro project
    • Pinned all Remotion packages to 4.0.387 (was mixing 4.0.383 and 4.0.387)
    • Removed ^ prefix to prevent future drift

2025-12-10

Added

  • /scene-review command - Dedicated scene-by-scene review with Remotion Studio
    • Starts Remotion Studio for visual verification
    • Walks through scenes one by one (not summary tables)
    • Generic - works with any template's config
    • /video now delegates to /scene-review when phase is review
    • /generate-voiceover warns if review not complete
    • Fixes: Review kept getting skipped because /video command was too long

Changed

  • Consolidated tracking files - Simplified from 4 files to 3:
    • ROADMAP.md - What we're building (removed duplicate "Next Actions" and "In Progress" sections)
    • BACKLOG.md - What we might build (removed all implemented items, now only future ideas)
    • CHANGELOG.md - What we built (historical record)
    • Removed FEEDBACK.md - Evolution principles moved to docs/contributing.md
  • Created docs/contributing.md with evolution principles and contribution workflow

Fixed

  • Slash commands not loading - Renamed /skill to /skills to avoid conflict with built-in Skill tool. The naming collision was silently preventing ALL custom commands from loading. Bug reported to Anthropic.

Removed

  • /review command - Clashed with Claude Code's built-in PR review command. Replaced by /scene-review.

Added

  • Animation components (lib/components/)
    • Envelope - 3D envelope with opening flap animation, configurable message
    • PointingHand - Animated hand emoji with directional slide-in and pulse effect
  • /contribute command - Guided contribution workflow
    • Report issues via gh issue create
    • Submit PRs for improvements
    • Share skills and templates
    • Safety checks to exclude private project work
  • Evolution narrative across all commands
    • Consistent "## Evolution" section in each command
    • Local improvement workflow
    • Remote contribution links (GitHub issues + PRs)
    • Command history tracking
  • Template evolution guidance in /template command
    • How to add features to existing templates
    • Template maturity indicators
    • Pattern extraction to shared lib/
  • Product demo template README (templates/product-demo/README.md)
    • Quick start guide, configuration, project structure
    • All 7 scene types documented with examples
    • Narrator PiP and demo chrome options

Changed

  • projects/ now fully gitignored - User video work stays completely private
    • Only .gitkeep is tracked to preserve directory structure
    • Safe to contribute without exposing project content
  • New examples/ directory for shareable showcase projects
    • Configs, scripts, and docs are tracked
    • Large media files (mp4, mp3) are gitignored
    • Each example includes ASSETS-NEEDED.md documenting required media
  • /contribute now supports example projects (Option 5)
    • Guides copying from projects/ to examples/
    • Auto-generates ASSETS-NEEDED.md
    • Creates README with quick start instructions
    • Contributor recognition with backlinks to website/org
  • CONTRIBUTORS.md - Recognition for organizations and individuals who share examples
  • Documentation updates
    • docs/getting-started.md - Updated for /video command, added multi-session workflow
    • docs/creating-brands.md - Updated for /brand command integration
  • Renamed skills-registry.json → toolkit-registry.json
    • Consistent format across all entries (path, description, status, created, updated)
    • Added components section for shared lib/components
    • Synced with actual commands (7), templates (2), components (9)

2025-12-09

Added

  • Multi-session project system (lib/project/)

    • types.ts - TypeScript schema for project.json
    • README.md - Documentation for project lifecycle and phases
    • Projects now track: phase, scenes, assets, audio, session history
    • Filesystem reconciliation (compares intent vs reality)
    • Auto-generated CLAUDE.md per project for instant context

  • Unified commands - Context-aware entry points that list existing items or create new:

    • /video - Replaces /new-video. Scans projects, offers resume or new
    • /brand - Replaces /new-brand. Lists brands, edit or create new
    • /template - Lists templates or creates new ones (copy, minimal, from project)
    • /skill - Lists installed skills or creates new ones

Changed

  • Command pattern unified - All domain commands now scan first, then offer actions
  • Commands integrate with project.json for state tracking
  • README.md updated with new commands and multi-session workflow

Removed

  • /new-video - Replaced by /video
  • /new-brand - Replaced by /brand

Notes

  • After creating/modifying commands or skills, restart Claude Code to load changes

2025-12-09

Added

  • Shared component library (lib/)
    • lib/theme/ - ThemeProvider, useTheme, createStyles, type definitions
    • lib/components/ - Reusable video components:
      • AnimatedBackground (supports subtle, tech, warm, dark variants)
      • SlideTransition (fade, zoom, slide-up, blur-fade)
      • Label (floating badge with optional JIRA ref)
      • Vignette (cinematic edge darkening)
      • LogoWatermark (corner branding)
      • SplitScreen (side-by-side video layout)
      • NarratorPiP (picture-in-picture presenter - needs refinement)
    • lib/index.ts - Unified exports for templates

Changed

  • Templates now import from shared library
    • sprint-review: Uses lib components via re-exports
    • product-demo: Uses lib components (keeps local NarratorPiP for different API)
    • Both templates use shared ThemeProvider from lib/theme
    • Deleted duplicate component files from templates
  • Updated ROADMAP.md - Shared component library marked complete
  • Updated BACKLOG.md - Added NarratorPiP API refinement task

2025-12-09 (earlier)

Added

  • Product demo template (templates/product-demo/)
    • Scene-based composition (title, problem, solution, demo, stats, CTA)
    • Config-driven content via demo-config.ts
    • Dark tech aesthetic with animated background
    • Narrator PiP (picture-in-picture presenter)
    • Browser/terminal chrome for demo videos
    • Stats cards with spring animations
  • /new-brand command - guided brand profile creation
    • Extract colors from website URL
    • Manual color entry with palette generation
    • Logo and voice configuration guidance
  • Digital Samba brand profile (brands/digital-samba/)
  • Template-brand integration
    • Brand loader utility (lib/brand.ts)
    • brand.ts in each template (generated at project creation)
    • project.json for tracking project metadata
    • Brand generator script (lib/generate-brand-ts.ts)
  • /new-video command - unified project creation wizard
    • Choose template (sprint-review, product-demo)
    • Choose brand from available brands
    • Scene-centric workflow:
      • Content gathering (URLs, notes, paste)
      • Claude proposes scene breakdown
      • Interactive scene refinement
      • Scene types: title, overview, demo, split-demo, stats, credits, problem, solution, feature, cta
      • Visual types: [DEMO], [SCREENSHOT], [EXTERNAL VIDEO], [SLIDE]
    • Generates VOICEOVER-SCRIPT.md with narration and asset checklist
    • Creates project.json with scene tracking
    • Guides through asset creation phase
    • Ongoing project support (resume any project)

Changed

  • Replaced /new-sprint-video with /new-video - single entry point for all templates
  • Templates now load theme from brand.ts instead of hardcoded values
  • Updated CLAUDE.md with new workflow and commands
  • Updated ROADMAP.md - Phase 3 template-brand integration complete
  • Added Video Timing section to CLAUDE.md - pacing rules, scene durations, timing calculations
  • Removed /convert-asset from backlog - FFmpeg skill handles this conversationally
  • Removed /sync-timing from backlog - timing knowledge now in CLAUDE.md

2025-12-08

Added

  • Open source release - Published to GitHub at digitalsamba/claude-code-video-toolkit
  • Brand profiles system (brands/)
    • brand.json for colors, fonts, typography
    • voice.json for ElevenLabs voice settings
    • assets/ for logos and backgrounds
    • Default brand profile included
  • Documentation (docs/)
    • getting-started.md - First video walkthrough
    • creating-brands.md - Brand profile guide
    • creating-templates.md - Template creation guide
  • Environment variable support
    • ELEVENLABS_VOICE_ID - Set voice ID via env var
    • Falls back to _internal/skills-registry.json if not set
  • /generate-voiceover command - guided ElevenLabs TTS generation
  • /record-demo command - guided Playwright browser recording
  • Interactive recording stop controls (Escape key, Stop button)
  • Window scaling for laptop screens (--scale option, default 0.75)
  • FFmpeg skill (beta) - common video/audio conversion commands
  • Playwright recording skill (beta) - browser demo capture
  • Playwright infrastructure (playwright/) with recording scripts
  • Python tools: voiceover.py, music.py, sfx.py
  • Skills registry for centralized config
  • README.md, LICENSE (MIT), CONTRIBUTING.md
  • .env.example template

Changed

  • Directory restructure for open source:
    • templates/ - Video templates (moved from root)
    • projects/ - User video projects (moved from root)
    • brands/ - Brand profiles (new)
    • assets/ - Shared assets (consolidated)
    • _internal/ - Toolkit metadata (renamed from _toolkit/)
  • Updated /new-sprint-video command paths
  • tools/config.py reads from _internal/ and supports env vars
  • Playwright recordings output at 30fps (matches Remotion)

Fixed

  • Removed hardcoded voice ID from committed files
  • Proper .gitignore for secrets and build artifacts
  • FFmpeg trim command syntax (use -to not -t for end time)
  • Playwright double navigation issue
  • Recording frame rate mismatch (was 25fps, now 30fps)

2025-12-04

Added

  • Sprint review template (templates/sprint-review/)
    • Theme system with colors, fonts, spacing
    • Config-driven content via sprint-config.ts
    • Slide components: Title, Overview, Summary, EndCredits
    • Demo components: DemoSection, SplitScreen
    • NarratorPiP component for picture-in-picture narrator
    • Audio integration (voiceover, background music, SFX)
  • /new-sprint-video slash command for guided project creation
  • Initial workspace setup
  • Remotion skill documentation
  • ElevenLabs skill documentation
  • First video project: sprint-review-cho-oyu
  • Voice cloning workflow with ElevenLabs
Reference in New Issue View Git Blame Copy Permalink