Changelog

v1.14.0

Cinematic 16:9 gallery, a new image and motion pipeline, two new heroes.

• Widescreen gallery. Scene images are now 16:9 cinematic, matching the video.
• New look. A more detailed cel style (Legend of Vox Machina) with real cinematic camera framing per scene.
• Two new party members. Cartman Sandiego (white dragonborn paladin) and Wrathgar (mountain-man ranger) join the roster; Shirley retired after his heroic death.
• First animated scene. Scene 3 of the July 2 gallery plays a motion clip on hover, our first storyboard-animated card.
• July 2 session with a full gallery and a "Previously on..." recap over a new graveyard-exploration music track.

v1.13.4

Changelog page rendering fixes.

• Entries without bullet items (like v1.13.3) now render their lede line as a subtitle.
• Inline bold and ` code ` markdown renders properly instead of showing literal asterisks.

v1.13.3

Oggy crop touch-up. Tighter zoom + a bit of breathing room above his dreadlocks.

v1.13.2

Portrait crop final pass.

• Caelum's hair no longer clips against the top of his card.
• Orion and Oggy zoomed a touch tighter so their faces match the rest of the roster.

v1.13.1

Homepage portrait retune + hero Kimber-sword fix.

• Uniform face size across all 6 portraits. First pass kept Shirley's face dominant while the others looked zoomed-out. Crops retuned so all faces fill a similar share of each card.
• Hero regenerated. v3 gave Kimber a sword; v4 explicitly reserves the Shadow Blade to Oggy. Kimber's hands now swirl bardic sigils.

v1.13.0

Oggy locked, uniform homepage portraits, mobile hero fix.

• Oggy is officially LOCKED (Jesse-approved). The canonical reference is now the full-body pantless variant because the no-pants gag is canon and full-body framing gives scene gen more to anchor on. Illusion-pants variant kept as a secondary for town scenes.
• Homepage portraits unified. All 6 characters now use the same tight Shirley-style bust crop. Caelum noticeably tighter. References stay full-body, homepage portraits are decoupled bust crops via a new scripts/crop-homepage-portraits.mjs.
• Removed portrait card bottom-fade gradient. The overlay was eating detail without adding anything.
• Mobile hero autoplay fixed. iOS Low Power Mode auto-reports reduced-motion, which used to hide the cinemagraph entirely. The hide rule is gone (the ambient loop isn't motion-reduce-worthy) and a JS .play() safety call handles iOS Safari's autoplay race.
• Hero regenerated with new Oggy. New group-action-v3.webp + new cinemagraph loop. Oggy visible in the back-left with his purple-black Shadow Blade.
• May 21 gallery refreshed. Scenes 4, 6, and 7 (the three with Oggy) regenerated as both still images and videos using the new locked reference.

v1.12.0

6-character cinematic hero + cleaner party grid.

• Homepage hero now features all 6 main party characters (was 4) in a 16:9 cinematic banner. Cinemagraph: Voltedge lightning crawls, Destone runes pulse, Kimber's bardic sigils swirl, embers drift, fur and cloth micro-drift. image == end_image loop anchor keeps the seam invisible.
• Hero container: 4:3 → 16/9. More banner-y, less of a square block.
• Diego (Maxi) hidden from party grid while on hiatus. Clean 3×2 of the 6 active characters.

v1.11.0

Portrait-forward party roster + image-generation style consistency.

• Homepage party section redesigned: full 2:3 portrait cards replace the tiny avatar circles (3×2 grid on desktop, 2-col tablet, 1-col mobile). Each card features the character's portrait with accent-colored top border, hover lift + scale + glow keyed to their accent color.
• New /portraits/ directory: WebP portraits for Caelum, Orion, Kimber, Idril, Shirley, Oggy. Diego placeholder card while his portrait is pending.
• Image generation style consistency: scripts/generate-images.js now auto-includes locked character portraits as input refs to GPT Image 2's images.edit endpoint whenever a scene mentions a registered character. Anchors both style AND likeness across sessions. New scripts/lib/style.js is the canonical source of the v4 style block and locked-portrait registry.
• Orion portrait LOCKED (David-approved). Oggy v2 generated using locked Orion + Idril as style refs (pending Jesse's approval).

v1.10.0

Animated hero.

• Homepage hero swapped from the static d20 image to a looping cinemagraph of the four canon-locked party members mid-combat. Source: GPT Image 2 group portrait → Seedance 2.0 i2v with image and end_image both pointing at the same frame to anchor the loop. Cinemagraph motion-only prompt (lightning crawling on Voltedge, embers, magical sigils swirling) keeps the seam invisible.
• Video assets: /video/hero-loop.mp4 (2.2 MB, audio stripped, H.264 CRF 26, +faststart), /hero-poster.webp (179 KB first-frame poster).
• Hero markup: <video autoplay muted loop playsinline preload="metadata"> with the poster image as fallback. prefers-reduced-motion: reduce users see the static poster only.
• New canon: player-approved visual descriptions locked for Caelum, Kimber, and Shirley.
• New scripts: scripts/generate-portraits.js and scripts/generate-group-portrait.js.

v1.9.1

Cinematic recap on the Recap page.

• Recap page now embeds the stitched session video at the top as a "Cinematic Recap" section (no autoplay, uses the first scene image as poster). The "Previously on..." narration audio stays below it, both sections labeled and described.
• Gallery page gets a "Watch the full cinematic recap →" link below the grid that deep-links to the recap page.
• New optional recapVideo field on the player-summary frontmatter (string path). When unset, both surfaces hide gracefully.
• May 7 session is the first session with a published recap (/video/sessions/2026-05-07/recap-full.mp4, 49 MB, 1:40).

v1.9.0

Animated session gallery.

• Session gallery cards now play their Seedance i2v video clips on hover (muted, looped) and open the full clip with audio in the GLightbox lightbox on click. Backward-compatible: cards without a video: frontmatter field still render as still images.
• "Click for sound" badge in the top-right of each animated card. Keyboard focus, touch tap-and-hold, and mouse hover all trigger the preview.
• New scripts/stitch-video.js for local recap stitching: node scripts/stitch-video.js YYYY-MM-DD concats every scene-XX.mp4 into a single .video/stitched/YYYY-MM-DD-recap-full.mp4 via ffmpeg stream-copy (re-encode fallback if codecs ever drift). Local-only, not shipped to the site.
• May 7 session shipped as the first fully animated gallery (10 scenes × 10s at 720p with synchronized audio via infsh app run falai/seedance-2-i2v).
• Schema: optional video field added to the image entry in site/src/content.config.ts. .gitignore extended to cover .video/ and *.mp4/*.mov/*.webm so video assets ship to Vercel via the CLI like audio.
• CLAUDE.md Future Improvements updated: Seedance i2v marked VALIDATED, and a new "narrated video recap" idea captured for a stitched + voiceover cinematic recap.

v1.8.0

Pipeline infrastructure overhaul.

• New /process-audio slash command wrapping scripts/process-audio.sh for consistency with /process-session and /narrate
• Mono 16kHz audio compression in process-audio.sh and transcribe-cloud.sh (down from 48kHz stereo). 4x smaller files, more reliable for gpt-4o-transcribe, native format for both Whisper and GPT-4o
• Transcription research doc (docs/mechanics/transcription-research.md), full A/B testing of whisper-1 vs gpt-4o-transcribe vs gpt-4o-mini-transcribe vs WhisperX. Documented gpt-4o-transcribe's repetition hallucination loops, WhisperX's prompt-injection hallucinations, and silent failure modes. Reverted default to whisper-1.
• Open SDK migration: openai 6.29.0 → 6.38.0, dotenv 17.3.1 → 17.4.2
• Google SDK migration: deprecated @google/generative-ai → @google/genai ^2.5.0
• OpenAI image model: gpt-image-1 → gpt-image-2 in scripts/generate-images.js (released April 21, 2026)
• Animation style R&D complete (.audio/test-style/README.md): v4 prompt validated for western fantasy animation key-frame aesthetic via GPT Image 2 + Seedance 2.0 i2v
• Character reference workflow introduced: per-character locked refs in docs/characters/image-references/. Idril locked as the first (April 16, 2026, Cary's approval). Process documented for the remaining characters.
• May 7 session shipped with the new animation style as the first session in the new aesthetic
• Mexoz Valaryn lore canonized: corrected from "Mexos" + added the full DM-read backstory: Serath-Kul empire origin, Black Pilgrimage 40-year wandering, The Ebon Gospel fragments with missing final pages
• Session gallery cleanup: removed the NB2 / GPT Image 2 model toggle from site/src/components/ImageGallery.astro. Sessions now ship a single set of images in the animation style. The --model nb2|pro flags remain in scripts/generate-images.js for ad-hoc use. Legacy src_nb2 field kept on the schema as optional so older session frontmatter (Mar 12, Apr 9) still parses without error.

v1.7.1

• Background ambient music mixed into narration via ElevenLabs Music API + ffmpeg
• New mix-narration.js script with configurable volume, intro/outro delay, fade in/out
• Voice-only narration preserved as backup (*-narration-voice.mp3)
• April 9 narration updated with mixed version (dark medieval ambient track at 30%)
• ElevenLabs client timeout bumped to 120s for longer scripts

v1.7.0

• Narration recap promoted to first-class UI: Recap tab in session navigation (conditional, like Gallery)
• "Previously, on Mostly Mortals..." featured card on homepage above session listing
• Headphones badge on session cards linking directly to recap page
• Recap page now uses Session layout with full tab navigation (was standalone Base layout)
• Recap page is now discoverable (removed noindex, shareable URL)
• New LatestRecap.astro component auto-selects most recent narrated session
• Session card markup refactored to support independent badge link

v1.6.2

• Structured image prompt recipe (INPUTS, CHARACTER REFERENCES, CORE COMPOSITION, VISUAL EFFECTS, COLOR SYSTEM, STYLING, LAYOUT DETAILS, CONSTRAINTS, NEGATIVE)
• Recipe documented in docs/mechanics/image-prompt-recipe.md
• generate-images.js updated to parse recipe format (backwards compatible with legacy)
• All 9 session images regenerated with recipe prompts (GPT + NB2)
• Storytelling captions replace technical descriptions in gallery
• /process-session updated to use recipe format and separate captions from prompts

v1.6.1

• Phased validation in /process-session (3 phases: Combat & Events, Items & Quests, Best Moments & Images)
• Deep banter pass: best moments now explicitly extracted as Quotable Quotes, D&D Highlights, and Table Culture
• Audio pipeline auto-detects garbled Whisper chunks and retries with WhisperX fallback
• Unified audio pipeline script (process-audio.sh) with .audio/ directory structure
• WhisperX venv rebuilt for current project path

v1.6.0

• Added Session Stats bar to player summary page (crits, fails, kills with Lucide icons)
• Stats extracted from transcript during /process-session with validation pass
• Stats stored as optional frontmatter in player-summary files
• DM stats section shown only when DM rolled crits or fails
• Cleaned up duplicate site/openspec directory and added working directory rules to CLAUDE.md

v1.5.0

• Every present character now guaranteed inclusion in Player Summary, Session Gallery (8-10 images), and narration recap
• Image prompt count bumped from 6-8 to 8-10 to ensure full party coverage
• Pipeline instructions updated to flag missing characters during validation

v1.4.2

• Added collapsible "Read the narration" transcript on recap page
• Narration text cleaned of Audio Tags and SFX markers for readable display

v1.4.1

• Fixed dark mode audio player visibility on recap page (inverted controls for contrast)

v1.4.0

• Added "Previously on Mostly Mortals..." AI narration via ElevenLabs v3
• Custom "Old Adventurer" voice designed in ElevenLabs Voice Design for v3
• /narrate command generates dramatic recap scripts with v3 Audio Tags ([excited], [sarcastic], [curious], [sighs], [whispers], [clears throat])
• Three voices available: Old Adventurer (default), Clint, Oxley via --voice flag
• Hidden /recap page for sharing narrations with players (noindex, not linked)
• NarrationPlayer component with Grimoire-themed audio player
• Script review and approval step before audio generation

v1.3.2

• Extended left page edge effect (spine line + gradient) through footer for seamless visual flow
• Updated site URL to custom domain mostly-mortals.bydom.io

v1.3.1

• Added MM shield logo to site navigation bar
• Added favicon, apple-touch-icon, and PWA icons generated from logo
• Added web manifest for mobile "Add to Home Screen"
• Added OG and Twitter card meta tags using hero image for link previews

v1.3.0

• Added Quest Log with accordion UI, Lucide icons, status badges, progress fractions
• Added "Current" quest highlighting with gold glow
• Added manual sort order for quests via sortOrder field
• Added Hooks & Leads section with CircleHelp icons
• Added section navigation (Party / Quests / Hooks / Sessions) with smooth scrolling
• Added completed quests section at 50% opacity
• Migrated 7 quests from knowledgebase to structured content collection
• Updated /process-session with quest extraction, validation, and dual-file updates

v1.1.0

• Added AI image generation pipeline (scripts/generate-images.js)
• Added GPT Image 1.5 (default), Nano Banana 2, and Nano Banana Pro model support
• Added --compare flag for side-by-side model testing
• Added Session Gallery page with GLightbox for full-size viewing
• Added GPT Image 1.5 / Nano Banana 2 model toggle on gallery
• Added image prompt extraction with framing types (wide/closeup/extreme) to /process-session
• Updated content schema with optional images array
• Generated 6 images for March 12 cemetery/Mexoz session

v1.0.0

• Initial release of the Mostly Mortals campaign website
• Astro static site with "The Grimoire" parchment theme (dark/light toggle)
• Homepage with hero banner, party roster, and session list
• Player Summary, Best Moments, and hidden DM Summary pages
• Session Timeline and Detailed DM Timeline sections
• Content collections with Zod schema validation
• Manual deploy to Vercel via CLI