--- > Analyze the commercial and technological state of virtual and augmented humans, possibly powered by AI. What are the tools necessary to create them? What is the market for them? Are there current implementations? What would be the pathway for a minimum viable product? the budget? --- **Virtual and augmented humans are commercially real in 2026, but the industry split is sharper than the marketing suggests.** Cheap neural talking heads and virtual influencers scale today. Photoreal, real-time, emotionally convincing 3D “people” still sit in a narrower enterprise and entertainment band, with brutal unit economics if you try to build them like a Hollywood character. The useful way to think about the category is a stack, not a single product: appearance + animation + voice + brain + delivery surface. --- ## What the terms actually mean **Virtual humans** are fully synthetic people: VTubers, AI presenters, conversational avatars, digital brand ambassadors, virtual influencers, game NPCs. **Augmented humans** usually means one of three things: - A **digital twin** of a real person (executive, trainer, celebrity, creator) used in video, chat, or AR. - An **AR overlay** of a person in the physical world (kiosk, glasses, Vision Pro, event activation). - A **human + AI hybrid**, where a real performer drives or supervises a digital body (classic VTuber model). AI is no longer optional. Generative models now own script, voice, lip-sync, facial performance, and increasingly live turn-taking. Traditional CGI still wins when you need a persistent 3D body you can put in Unreal, on a LED volume, or in VR. --- ## Technological state in 2026 The field has two mature production modes and one still-expensive one. **1. Neural video humans (most commercial volume)** Photo or short capture in → talking-head or gesturing video out. HeyGen Avatar IV/V, Synthesia, D-ID, DeepBrain, Hour One, Colossyan. Quality is good enough for training, ads, localization, and UGC-style marketing. Live conversational video (Tavus Phoenix-4, HeyGen Live, Beyond Presence, Anam, D-ID Agents) now runs at roughly sub-second to ~600ms end-to-end latency over WebRTC. That is the 2025–2026 inflection: you can talk *to* the face, not just watch a clip. **2. Stylized / 2D–2.5D live characters (largest cultural footprint)** VTubers (Hololive, Nijisanji/Anycolor) and social avatars (Ready Player Me, Genies). These avoid the uncanny valley on purpose and already have working monetization: livestream, merch, subscriptions. **3. Photoreal 3D real-time humans (highest ceiling, hardest economics)** Epic MetaHuman + Unreal Engine + MetaHuman Animator + NVIDIA ACE / Audio2Face + Pixel Streaming. This is the path for games, virtual production, kiosks, VR/AR, and brand ambassadors that must exist as a 3D object, not a video texture. NVIDIA ACE is the commercial “glue” for speech, face, and agent behavior, including a 2026 Game Agent SDK for local RTX. The hard remaining problems are not “can it look like a person.” They are: - identity consistency across angles, lighting, outfits, and minutes of live talk - hands, teeth, hair, and eye contact under motion - latency vs. visual fidelity - memory and personality that do not drift - consent, likeness rights, and deepfake regulation - cost per concurrent live session on GPU A cautionary data point: **Soul Machines**, the best-known “Digital People / Digital Brain” company, entered receivership in February 2026 after raising on the order of $120–135M+. High-end custom CGI + proprietary cognition did not produce durable unit economics against cheaper neural avatars and generic LLMs. UneeQ, a sibling-era NZ firm, is still operating and pitching CGI digital humans for CX and training. --- ## Market Treat published “digital human market” numbers as definition-dependent. Analysts mix video avatars, VTubers, gaming characters, companions, and enterprise CX into one pile. Reasonable 2025–2026 bands: | Slice | Approx. 2025–26 size | Growth | Notes | |---|---|---|---| | Narrower “digital human / AI avatar” platforms | ~$6–10B | ~27–31% CAGR | Mordor ~$6.3B (2025) → ~$8B (2026); GM Insights similar for AI avatars | | Broader “digital human” reports | $17–67B | 30–40% | TBRC-style reports inflate by including gaming, media, and services | | Virtual / AI influencers | ~$6–12B ecosystem; ~$1.4B direct brand-deal spend | very fast | Top accounts $50k–$200k/mo; median much lower | | AI companions (text + some avatar) | tens of billions if counted broadly; emotional/NSFW ~$1.2B | ~30%+ | Character.AI-class products, not photoreal 3D | Demand is real in five paying verticals: 1. **Enterprise learning and comms** — Synthesia-class video at scale, many languages. 2. **Customer experience and sales** — always-on concierges, banking, telecom, travel, healthcare front doors. 3. **Livestream commerce** — China already showed the commercial proof: a Baidu-backed AI avatar livestream reportedly did about $7.65M in seven hours, outselling the same host’s prior human stream. That is the “this works as a salesperson” moment. 4. **Creator / influencer IP** — synthetic talent that never sleeps, never ages, never has a scandal unless you write one. 5. **Games, VR/AR, virtual production** — MetaHuman as the default photoreal body. North America still spends the most on enterprise platforms. Asia-Pacific is the fastest growth engine (China livestream, Japan/Korea virtual idols, VTubers). Healthcare and financial services are the fastest-growing *serious* verticals because trust and language coverage matter. Gaming/entertainment still take a large revenue share. The commercial lesson of 2026: **buyers pay for throughput and control, not for “a soul.”** Training libraries, localized ads, 24/7 support deflection, and influencer content calendars clear budgets. Bespoke “empathetic digital employees” with six-figure implementations often do not. --- ## Current implementations (what is actually live) **Video presenters / digital twins** Synthesia, HeyGen, Hour One, Colossyan, DeepBrain. Custom studio avatars typically $1k–$3k one-time; photo avatars are nearly free on creator plans. Entry SaaS is roughly $18–30/month. **Real-time conversational faces** Tavus, Beyond Presence, Anam, HeyGen Interactive, D-ID Agents, UneeQ. Used in support, coaching, healthcare intake, hospitality concierges. Public examples in 2026 include hotel/travel avatars in Japan/Taiwan and clinic-facing agents in Europe. **Brand ambassadors** UneeQ’s Sama for Qatar Airways, Deutsche Telekom’s Mia, UBS’s digital Daniel Kalt, various bank/telco kiosk humans. These are still custom, slow, and sales-led. **Virtual influencers** Lil Miquela, Lu do Magalu, Imma, Noonoouri, Aitana López, plus a long tail of AI-generated accounts. Lu do Magalu remains the commercial giant via retail; Lil Miquela remains the fashion prototype. Engagement can beat human mid-tier creators; earnings are extremely skewed. **VTubers / live 3D idols** Hololive and Nijisanji still show up as share leaders in some “AI avatar” reports because they already have audiences, agencies, and merch. That is the most proven “virtual human as entertainment business.” **3D engine humans** MetaHuman in Unreal for games, virtual production, and Pixel-Streamed web/kiosk experiences. NVIDIA ACE partners wrap speech + face + agent logic around that body. What you see on marketing pages vs. what ships: Most production work still looks like a framed talking head or a lightly stylized character, not a full-body person walking around your living room. --- ## Tools required to create one Think in layers. You can buy a whole layer as SaaS or assemble it. **Appearance** - Fast: HeyGen/Synthesia/D-ID photo or studio capture; Flux / Midjourney / LoRA for influencer stills - 3D: MetaHuman Creator, Mesh-to-MetaHuman, Reallusion Character Creator, Ready Player Me, photogrammetry / iPhone TrueDepth - Likeness services: $750–$7,500 for a production MetaHuman depending on hair, wardrobe, and cinematic grade; photo-to-MetaHuman pipelines exist from a few dollars per character for game-ready quality **Performance / animation** - Audio2Face, MetaHuman Animator, Live Link Face (iPhone), NeuroSync, Rokoko / Move.ai / webcam body capture - Neural models that drive face + some gesture from audio or text **Voice** - ElevenLabs, Cartesia, Inworld, OpenAI/Google TTS, vendor-native clones - Consent-captured clone of a real person if you are making a twin **Brain** - LLM (GPT, Claude, Gemini, or open weights) + RAG over your knowledge + session memory - Guardrails, tool calling, handoff to a human - Optional “persona bible”: biography, taboos, brand voice, emotional range **Orchestration and delivery** - Async: vendor studio → MP4 → LMS / ads / YouTube - Live: LiveKit / Daily / WebRTC + avatar API - 3D live: Unreal Pixel Streaming or local RTX; Vision Pro / Quest for AR/VR - Analytics, consent logs, watermarking, content moderation **The pragmatic 2026 shortcut** Do not build a renderer. Call Tavus / HeyGen Live / D-ID / Beyond Presence for the face, ElevenLabs for voice if needed, and your own LLM for the brain. Only drop into MetaHuman + Unreal if the human must exist as a 3D asset. --- ## MVP pathway Pick one job. “A virtual human” is not an MVP. “A Spanish-language product explainer twin that ships 40 videos a month” is. ### Path A — Fastest commercial MVP (2–4 weeks, recommended for most) **Job:** scalable talking-head content or a web conversational agent. 1. Define the use case and success metric (cost per trained employee, demo bookings, GMV, watch time). 2. Choose stock vs. custom likeness. Get written likeness consent if it is a real person. 3. Stand up HeyGen or Synthesia for async, or Tavus/D-ID/Beyond Presence for live chat. 4. Write a tight system prompt + RAG corpus. Add human escalation. 5. Ship on web first. Measure containment rate / completion rate / conversion. 6. Only then add a custom studio avatar or a second language. This is how almost every company that is actually in production started. ### Path B — Virtual influencer / IP MVP (4–8 weeks) 1. Niche, name, visual bible, voice, 30-day content calendar. 2. Still-image consistency via a trained LoRA or a locked avatar pipeline. 3. Short talking clips via HeyGen/Creatify/Arcads, not full 3D. 4. Post daily. Treat distribution as the product. 5. Monetize with affiliates before brand deals. Brand deals come after proof of engagement. ### Path C — Photoreal 3D interactive MVP (90–150 days) 1. One MetaHuman, one outfit, one scene. 2. Audio2Face or MetaHuman Animator + cloned voice + LLM. 3. Deliver via Pixel Streaming to a browser (do not start on Vision Pro). 4. Cap concurrency. GPU cost will dominate. 5. Use this only if the 3D body is the point (showroom, training simulation, celebrity twin, game character). Indie and agency quotes for a single-avatar 3D companion-style MVP commonly start around **$15k**, with multi-avatar + VR builds at **$40–80k**. Full “digital employee + custom agent architecture + always-on GPU” pitches still run **$90–250k**. Those numbers are real for custom work; they are not required for Path A. --- ## Budget **Path A — video / live-face MVP** - Setup: $0–$5k (mostly time). Custom studio avatar add-on: ~$1k–$3k. - Software: $20–$200/mo to start; $1k–$5k/mo if you are rendering or streaming hours every day. - Live session cost: roughly $0.10–$0.45 per streaming minute at published API tiers, plus LLM tokens. Plan for 1.5–2× the naive minute count because of takes and idle time. - People: one producer + one prompt/knowledge owner. No 3D team. **Path B — influencer** - Setup: $500–$5k in tools and design if you stay 2D/neural. - Ops: $50–$300/mo tools + your time or a $2–5k/mo editor/community person. - 3D “always-on MetaHuman influencer” is a different business: $30k–$150k+ year one. **Path C — 3D real-time** - Character: $750–$7,500 asset, or $15k+ if a studio builds the whole experience. - Infra: cloud GPU on the order of $0.65–$2+/hr depending on card and reservation; this is the silent killer at more than a few concurrent streams. - Team: technical artist + Unreal generalist + AI engineer, even if part-time. **Rule of thumb** - Prove the conversation or the content calendar on a $2k–$10k budget. - Spend six figures only after a metric moves (deflected tickets, booked meetings, GMV, completed training). - Soul Machines is what happens when the visual thesis outruns the business model. --- ## Practical recommendation If the goal is a **product**, start with a consent-captured digital twin or a named original character on a neural video/live-avatar API, wired to a retrieval-augmented LLM, delivered in a browser. That is the minimum viable virtual human in 2026. If the goal is an **IP**, start as a virtual influencer with locked visual identity and a posting machine. Add real-time later. If the goal is **presence in space** (store, stage, headset), budget for MetaHuman + Unreal + a streaming/GPU plan and treat it as a production, not a SaaS signup. The technology is past the demo stage. The scarce things are a specific job to be done, likeness rights, and a cost per interaction that is lower than a human in that job. --- > please produce an .md version of the 105 sources suitable for use in Obsidian. --- Obsidian-ready bibliography is in `Digital Humans — Source Bibliography.md`. It covers all **105 sources** from the research pass (`web:0`–`web:104`). What is in the note: - YAML frontmatter (`tags`, `aliases`, `source_count`, dates) for graph/search - Topic index mapping IDs to market, vendors, Soul Machines, pricing, MetaHuman, UneeQ, influencers - Each entry as `### S000`–`### S104` with URL, date when known, tags, and a one-line annotation - Deduped-canonical table (Mordor, GM Insights, TBRC, and receivership stories were returned more than once) - Suggested follow-on note titles and a Dataview stub if you later split entries into `Sources/Sxxx.md` files Drop the file into your vault as-is. Anchors work as `[[Digital Humans — Source Bibliography#S004]]` if you want to cite a single source from another note. --- >Focus on historical characters as virtual presentations. Are there case studies such as George AI by The Torch, or projects by companies like Grove Jones? What is their market success? What are their production pipelines, cost, and personnel? Find many others. --- Historical-character virtual presentations are a real product category now, but they are **not one market**. They split into four businesses with very different costs and success rates: closed-corpus “librarian” avatars, museum conversational installations, first-person survivor testimony systems, and cheap LLM chat apps that only look like history. “Grove Jones” in this space is almost certainly **Groove Jones**, the Dallas XR studio. They build avatars and museum interactives; they are not a historical-figure foundry the way Jumbo Mana or USC Shoah Foundation are. --- ## George AI by The Torch **What it is.** Glenn Beck’s Mercury Radio Arts product, launched with The Torch platform in January 2026. Beck presents George as a “librarian,” not a resurrection of Washington: a constrained model that answers from a private founding-era corpus (Washington, Jefferson, Adams, Madison, Franklin, Hamilton, plus Locke, Montesquieu, Blackstone, Cicero, Plutarch). The public cutoff is described as around 1820. A companion system, Glenn AI, is limited to Beck’s own verified output. Curriculum built on George is scheduled for 2027. **Presentation.** The viral preview was a sit-down “interview”: Beck at a table with a modernized Washington in a black T-shirt and podcast mic. Beck interrupted the first answer and asked it to “dumb it down” and “speak in today’s language.” The resulting monologue (“America’s biggest problem isn’t political or economic. It’s all moral.”) was widely read as Beck’s own frame, not a neutral retrieval of 18th-century text. Beck’s own explainer insists George is not Washington and not a vehicle for staff opinions. Critics treated that claim as incomplete, because the system was also “fed… the conditions of what’s happening in America today.” **Commercial model.** Torch Insider is **$8/month ($96/year)**. The avatar is a membership feature inside a media bundle (ad-free radio, documentaries, member chats, AI-generated daily podcasts), not a standalone museum product. Educational curricula are a 2027 upsell. Beck has said compute is expensive and that unit cost should fall with scale. No independent subscriber count or George-only revenue has been published. Success so far is **distribution plus controversy**, not a measured learning outcome. **Pipeline (inferred from public claims).** 1. Acquire / digitize a private document library (Beck claims one of the largest private founding-era collections). 2. Isolate the model from the open web (“electric fence”). 3. Human vetting and prompting. 4. Voice + on-screen talking-head for show segments. 5. Package inside an app/membership. **Personnel.** Not disclosed in detail. Publicly: Beck as editorial owner; Foundation for American History as the civic wrapper; an in-house research/prompt team; unnamed AI/engineering vendors. This is a **media + archive** org, not a 3D studio. **Honest assessment.** George is a RAG chatbot with a face and a brand. The interesting design choice is the closed corpus. The weak point is the same one every histobot hits: once you ask a dead person about 2026, you are writing a script, not retrieving a source. Market success will be decided by Torch retention and whether schools buy the 2027 curriculum, not by whether the avatar “is Washington.” --- ## Groove Jones (not “Grove Jones”) Dallas studio founded 2015. 300+ awards. They sell **GrooveTech**: XR Avatar Scanner (volumetric guest doubles), FanScan (phone face-scan avatars), Avateer (live-puppeteered CGI host), computer-vision body tracking. Clients and venues include Perot Museum of Nature and Science (bird-flight body tracking), Dallas Aquarium (“Otter Chat” with Tommy the Otter), Salesforce Dreamforce (markerless “Einstein” surf avatar — the Salesforce mascot, not Albert), Pumpkin Nights, NCAA Final Four, Toyota VR training. They are relevant as a **production vendor** if you want a historical character as an on-site activation: scan a performer, drive a CGI body live, drop it in a museum or event. They are not publishing Lincoln/Washington products of their own. Treat them as the implementation shop, not the historian. Typical event scanner claim: 1,000+ scans/day per rig. Pricing is custom experiential, not public SaaS. --- ## The rest of the field (many others) ### A. Museum conversational installations (the closest analog to “virtual presentation”) | Project | Figure | Who | What visitors get | Outcome | |---|---|---|---|---| | **Bonjour Vincent / Hello Vincent** | Van Gogh | Jumbo Mana + Musée d’Orsay / Château d’Auvers + art historian Wouter van der Veen | Real-time talk with a 3D Van Gogh trained on **800+ letters** + catalogue | Debut VivaTech 2023, then Orsay exhibit on his last months; later Auvers; Alexa skill “Bonjour Van Gogh” (2024). Flagship of the European histobot wave. | | **Bonjour Arthur** | Rimbaud | Jumbo Mana + Musée Arthur Rimbaud, Charleville-Mézières | Conversation grounded in bibliography | Follow-on to Vincent; same studio pipeline. | | **Bonjour Albert** (announced) | Albert Schweitzer | Jumbo Mana + Maison Schweitzer, Gunsbach | In development ~2024 | Shows the studio trying to productize “Bonjour [Name].” | | **Dalí Lives** | Salvador Dalí | The Dalí Museum + Goodby Silverstein & Partners + Thinking Box | Life-size screens; deepfake face on an actor; curator-verified quotes | 25,000 visitors in month one; 125 interactive videos, 190,512 combinations; still in the museum years later; museum cites ~$106M local economic impact (whole museum, not just the avatar). | | **Ask Dalí** | Dalí | Same museum, later conversational layer | Chat trained on writings + archived audio | Persistence after the original deepfake install. | | **An Audience with a Queen** | Eleanor of Castile | Leeds Castle, 2026 | Environment-aware live avatar in the chapel; EN/FR/ES | Strong curatorial model: exhibition first, conversation last. Face/costume **not** AI-generated; Spanish voice from a native actor. Answers are AI. Until Nov 2026. | | **Digital Einstein (ETH Zurich)** | Young Einstein | ETH + spin-off Animatico | Physical “office” set + screen; v1 dialogue tree + mocap; v2 GPT-4o / Llama | Built for Nobel centenary. Voice and dialogue by Swiss comedian Karpi. Later ACM paper on the physical + AI loop. CloudFest 2025 showed a separate Einstein activation (1SP Agency) claiming 5,000 documents / 47 expression points — treat those specs as marketing. | | **Historic Metahumans** | Avraham Biran, Gandhi, Feynman | UC San Diego / ASOR team (Hata, Levy, Smith et al.) | MetaHuman + LLM + scanned artifacts from Tel Dan | Family permission for Biran. Pipeline published as a poster: appearance, voice, emotion, animation plugin. Educational, not commercial. | | **St Andrews MetaHumans** | 8 documented medieval figures + 16 citizens | University of St Andrews | 24 MetaHumans, 12 garment sets, virtual exhibition | Academic workflow paper. | | **Cardiff “Chattable” avatar** | 18th-c. master of ceremonies | Cardiff University | UE5 MetaHuman + GPT-4, laptop RTX 3060 | GLAM prototype. | | **Ada and Grace** | Named for Lovelace & Hopper | USC ICT + Museum of Science, Boston | Life-size virtual docents | NSF-funded; the ancestor of today’s museum agents (2010s). | | **Max** | Original ECA | Paderborn Heinz Nixdorf MuseumsForum, 2005 | Life-size screen guide | First widely cited museum conversational human. | | **Silla king chatbot** | Silla-period king | National Museum of Korea research | 3D figure + archaic speech, curator-checked QA | Academic; embodiment + reenactment tested against a plain QA kiosk. | | **Francisco de Arobe** | 16th-c. Esmeraldas leader | Museo de América, Madrid (student thesis build) | Stylized chatbot tied to a painting | Institutional request, low budget. | | **Chinese museum humans** | Xiao Ke, Ling Meixiang, Ai Wenwen | Shanghai Museum, Bay Area Museum, National Museum of China | Guides / storytellers, not always a dead celebrity | State-backed digital heritage, awards for Ai Wenwen (2023). | | **ENGAGE XR School of AI** | 21 sets: Earhart, Franklin, Parks, Tubman, Curie, Cleopatra, Churchill, etc. | ENGAGE / Educators in VR | VR field trips, spoken conversation | Classroom/VR, not a single hero install. | | **Lincoln Unlocked / Ghosts of the Library** | Lincoln and era figures | ALPLM + Google Cloud / Thoughtworks; older Holavision by BRC | AR “Lincoln delivers Gettysburg”; holographic theater with live actor + ghosts | Museum already draws hundreds of thousands/year. Tech is additive, not the ticket. | ### B. First-person testimony (the ethically strict end) **USC Shoah Foundation — Dimensions in Testimony.** Survivors answer ~1,000–2,000 questions over ~5 days inside a multi-camera dome (originally 116 cameras at USC ICT). NLP maps visitor questions to real recorded clips. Not a generative “fake person.” Languages include EN, ES, HE, DE, ZH, RU, SV. Installed in museums worldwide; also in IWitness for classrooms. Pinchas Gutter was the prototype. **Costs published:** - Early Pinchas build: **$1.8 million**. - Orlando Holocaust Center / StoryFile local survivor (2026): about **$250,000 per subject** all-in (capture, interface, equipment, education wrap). - Eva Mozes Kor / similar: ~30 hours of interview → tens of thousands of Q–A pairings; museums license content + hardware monthly. **StoryFile** is the commercial spin-out (Heather Maio and collaborators) selling interactive biographies beyond the Shoah collection. Same philosophy: record the living person, retrieve clips, do not invent testimony. This is the only slice with a durable institutional market: museums *must* keep survivor conversation after the last witnesses die. Generative Washington does not have that urgency. ### C. Cheap “talk to anyone” apps (volume, low fidelity) - **Hello History / Humy.ai** (Facing It International, Sweden): 400+ then **1,210 figures**, GPT-class models, 50+ languages, teacher tools. Claimed **200k+ downloads, 5M+ messages, 5,000+ teachers, 90+ institutions**. Freemium; 20 free messages. This is the volume play. - **Jenova Roleplay Game Master**, Character.AI historical bots, generic “talk to Socrates” wrappers — same stack, thinner scholarship. - Academic one-offs: Joseph Lister RAG character (Heritage 2025); Hedy Lamarr “histobot” (Harder 2024/2026) that the builder later treated as a failure case about authorship. Market success here is **downloads and classroom seats**, not ticket revenue. Historical accuracy is the product risk. ### D. Stage “holograms” of the dead (entertainment, not education) **BASE Hologram / BASE Xperiential:** *Callas in Concert* (2018–), Roy Orbison, Buddy Holly, Whitney Houston. Body double rehearsed 12 weeks; CG face; remastered recordings; live orchestra. ~9-month build. Reviews mixed (awe at the walk-on, complaints that a hologram cannot conduct or respond). Datanyze once listed ~$5.1M revenue / ~15 staff — treat as a directory estimate. Tours still book (Dallas Opera revival Oct 2025). This is **estate + touring theatrical**, $millions per title, not a museum kiosk. Disney’s 1964 World’s Fair **Audio-Animatronic Lincoln** is the ancestor of the whole genre. --- ## Market success — what actually works There is no clean “historical avatar TAM.” Money shows up in five different P&Ls: 1. **Museum attendance and grants.** Dalí Lives and Bonjour Vincent succeed if they increase dwell time, press, and membership. They rarely sell as standalone SKUs. 2. **Testimony licensing.** Dimensions in Testimony / StoryFile: high capex per person, then multi-year museum licenses. The $250k–$1.8M figures are real. 3. **Consumer subscriptions.** Torch $8/mo and Hello History freemium. Avatar is a feature of a media or edtech bundle. 4. **Touring entertainment.** Callas-class holograms: ticketed nights, high production, estate deals. 5. **Agency projects.** Groove Jones, EquipVerse-class MetaHuman work, Jumbo Mana commissions. One-off fees. Jumbo Mana is still a small shop (PitchBook/CB Insights: founded 2022, ~11–13 people, ~$30k disclosed raise plus a later €3M raise attempt; named logos include museums plus CMA CGM, Vinci, Galeries Lafayette, Amazon Alexa, European Council). That is **studio success**, not category domination. Pattern: **closed sources + a venue + a scholar on the credit list** survive. Open-web “I am Napoleon” bots get users and then get fact-checked to death. --- ## Production pipelines, cost, personnel Four stacks, in rising cost. **1. Text histobot (Hello History, academic Lamarr/Lister, early George without the face)** - Sources: Wikipedia + selected biographies, or a curated corpus + RAG. - People: 1 prompt engineer, 1 historian reviewer if you are serious. - Cost: hundreds to low tens of thousands. Inference pennies per chat. - Time: days to weeks. **2. Video presenter / deepfake (Dalí Lives, George AI on-camera, HeyGen-class twins)** - Sources: archival film (Dalí: 6,000+ frames, 1,000 hours of training), or a designed likeness. - People: curator, copywriter, actor matching body type, voice actor, ML/compositor, editor. GS&P + Thinking Box + museum staff for Dalí. - Cost: mid-five to low-seven figures for a flagship install; much less if you accept a modern talking-head and skip the deepfake. - Time: months. **3. Real-time 3D conversational human (Vincent, Eleanor, Einstein, MetaHuman Biran)** Jumbo Mana’s published stages: kickoff (photos, voice refs, editorial brief, knowledge base) → data prep and 3D sculpt → app/UX + conversational AI → staged launch. Leeds Castle explicitly split the work: **historians own the body; the model owns the answers.** Voice is often a directed actor, not a clone, when no recording exists. People on a serious museum build: - curator / historian (non-negotiable) - 3D character artist + costume researcher - voice director + actor - conversational designer + LLM/RAG engineer - on-site integrator (kiosk, lighting, wake-word, languages) - evaluator for hallucination and off-limit topics (suicide, living descendants, current politics) Cost band that matches other digital-human work: **roughly $80k–$400k** for one hero museum character if you already have the engine; **$15k–$80k** if you accept MetaHuman + Pixel Streaming and a thin knowledge base; **well under $10k** if it is only chat. ETH Einstein and Orsay Vincent sit toward the high-craft end. **4. Volumetric testimony or touring hologram** - People: interview team, lighting/camera crew, NLP annotators, display engineers; or body double + VFX + director + live orchestra. - Cost: **$250k–$2M+** per person or title. - Time: a week on set plus months of indexing — or nine months for Callas. --- ## What to copy if you are building one The projects that hold up share the same discipline: - Say what the system **is not**. Beck does this on paper; the interview format undermines it. Leeds Castle and Shoah Foundation do it in the room. - Bound the corpus. Letters (Van Gogh), verified quotes (Dalí), recorded answers (survivors), founding-era texts (George). - Put a historian in the loop before launch and after logs show new question types. - Separate likeness rights from voice rights from interpretive rights. Dead public-domain figures are legally easier and ethically harder. - Measure dwell time, repeat questions, and correction rate — not “uncanny likeness.” George AI is a media product that uses history. Groove Jones is a vendor that can stage a body. The museum line from **Dalí Lives → Bonjour Vincent → Eleanor of Castile**, and the testimony line from **Dimensions in Testimony → StoryFile**, are the actual case studies for “historical characters as virtual presentations.” Everything else is either a chatbot with a costume or a concert with a ghost. --- > please produce an .md version of the 125 sources suitable for use in Obsidian. Then, Analyze Jumbo Mana's Bonjour Vincent pipeline and Explore StoryFile's Dimensions in Testimony and provide these as separate case study .md files. --- Three Obsidian notes are in the artifacts folder. **1. Source list (125 entries, `web:105`–`web:229`)** YAML, topic index, annotated `S105`–`S229`, dedupe table, wikilinks to the two case studies. **2. Jumbo Mana / *Bonjour Vincent*** Pipeline in five stages (brief → 800–900 letters + van der Veen → 3D avatar → Orsay kiosk → Alexa / Rimbaud / Schweitzer afterlife), people, inferred cost, what the press got right and wrong. **3. Dimensions in Testimony and StoryFile** These are related, not the same product — the note opens with that, because DiT is USC Shoah Foundation and StoryFile is the commercial engine that grew out of Heather Maio’s concept. Covers the 5-day dome shoot, retrieval-not-generation rule, **$1.8M** Pinchas vs **~$250k** Orlando StoryFile captures, and when to use this method instead of a histobot. Drop all three into the same vault as `Digital Humans — Source Bibliography.md`. Wikilinks already point at each other. --- Specify the pipeline for a full-body 3D virtual human, viewable from a kiosk or via mobile device in AR, or as a VR experience. The character would be based on a historical figure, their writings, contemporary accounts, and the general human knowledge available during their time. What are the possible subject matter for presentations? What kinds of interactive events could this character perform (like reenactments, chautauqua monologues, or discussions with a contemporary 3D virtual human)? What resources would be needed to produce a faithful, complete, and compelling historical figure that could be viewed by today's audiences? What are some high-content and popular figures that could be the subject of this project? What are the legal, logistic, or moral considerations for such a project? --- The spec is also saved as an Obsidian note so it can sit with the earlier case studies. What follows is the working version of that brief. ## One asset, three surfaces Do not build a kiosk character, an AR character, and a VR character. Build an **identity package** (body, wardrobe, voice, corpora, refusal policy) and compile it three times. | Surface | Job | Runtime | |---|---|---| | **Kiosk** | Premiere fidelity, approach detection, captions, booked “private audience” | Unreal hero MetaHuman, local GPU or Pixel Streaming | | **Mobile AR** | The same person at a plaque, lawn, or classroom desk | Honest tier: billboard / USDZ with baked animation. Full live MetaHuman on a phone in sunlight is a promise you will break | | **VR** | Person *plus place* (print shop, lab, quarterdeck) | Same UE project, Optimized LOD, teleport + seated mode | MetaHuman in UE 5.6–5.8 is the default body path: Creator in-engine, Mesh-to-MetaHuman from a reconstruction, Live Link from phone/webcam/audio, mobile LODs that actually ship. Label on every surface: *An interpretation from sources dated to [year]. It can be wrong. It cannot know what happened after [cutoff].* --- ## Pipeline **0. Gate.** Score source density, likeness evidence, demand, legal clearance, moral heat, and whether a cutoff year is even drawable. Popularity without letters produces a costume. **1. Three corpora, kept apart.** - **A. Own words** (scholarly editions of letters, diaries, works) - **B. Contemporary accounts** (always attributed to other people) - **C. Period knowledge** (what that person could have known that year) A fourth file, **D**, never goes into the model: modern notes, contested facts, trigger topics, refusals. This is the Bonjour Vincent method (800–900 letters + a named historian) with George AI’s closed fence — and without feeding “today’s conditions” back in. **2. Body.** Photographs, masks, and multiple portraits beat one romantic painting. Costume historian on wardrobe. If no face survives, say so or pick another figure. **3. Voice.** Directed actor + dialect coach is the default before 1890. Credits name the actor. Captions always on. **4. Motion.** Mocap for the monologue; audio-driven face for unscripted answers; a small idle/life layer so the kiosk is inhabited when nobody is talking. **5. Mind.** Hybrid: retrieval for authored beats (DiT logic), RAG over A→B→C for the rest, generation only inside that fence. Log every question. The top query will be the one you did not want (at Orsay it was Van Gogh’s suicide). **6. Event clock.** Conversation is one mode, not the product. **7. Compile.** One repo; different LODs, hair, and compression per device. **8. Eval.** Blind historian check of 50 answers, a class, descendants or community if they exist, accessibility, red-team prompts. First complete hero is roughly **6–12 people over 9 months** and **$150k–$400k** for kiosk + Quest plus a thin AR postcard. Chat-plus-portrait is a different product. --- ## What the presentations are about Program them like exhibitions, each with a date, a place, and a source list: - A single decisive year or text - How a craft was done (press, lab, ship, kitchen) - An idea in *their* vocabulary, including what they got wrong if the corpus supports it - Site-specific AR / VR rooms - Classroom seminars locked to a source pack Refuse modern medical/legal advice, endorsements of living politicians, and “what do you think of [event after cutoff]?” as if they had an opinion. That last one belongs to the out-of-character beat. --- ## Events the figure can perform Steal the living-history **Chautauqua clock**; audiences already know it: in-character monologue → in-character Q&A → scholar drops character and answers what the figure could not. Then add: - **Dated reenactment** (Gettysburg as a listener, a salon, a demonstration with period tools) - **Object lesson** — visitor points at a prop or tagged artifact; the figure may speak only to that object - **Guided tour** — high throughput, low improvisation - **Private audience** — chapel-style, booked, a docent in earshot - **Dual-character programs** — the distinctive event this stack can do that a human Chautauquan cannot: Adams–Jefferson from the letters; figure × antagonist using a real debate; figure × a visually different “archive clerk” who owns the modern beat. Label counterfactuals in the title. Two corpora, two bodies, a moderator policy. Never put an LLM in front of survivor testimony. Do not stage séance, romance, or gift-shop concierge. --- ## Figures worth building **High-content and popular** sit at the intersection of a thick Corpus A and existing audience: Lincoln, Jefferson, Franklin, Douglass, Darwin, Galileo, Curie, Lovelace, Leonardo, Van Gogh, Austen, Twain, Gandhi, Bolívar, Ibn Sina, Murasaki, Sor Juana. **Fail the gate** when own words are thin (Cleopatra, Sacagawea, Hypatia as inner lives). Do a documentary or a reconstruction face instead. **Often better than a celebrity:** a local figure with letters plus a place (Eleanor at Leeds, Biran at Tel Dan, a mill worker at a factory museum), then export them via a dual with a famous correspondent. Einstein, Tesla, Napoleon, Kahlo, King are demand-rich and rights- or ethics-hot. Treat them as season two. --- ## Legal, logistic, moral **Legal.** Writings of the long-dead are often PD in the U.S.; *editions*, translations, and specific photographs may not be. Right of publicity is state law, not federal: California 70 years after death, New York 40 (with domicile and date limits), Indiana 100, Tennessee potentially perpetual if exploited. Educational biographical use is safer than merch or a subscription “meet them” app, but you still want a memo per figure, venue state, and business model. Voice and digital-replica clauses belong in the actor contract. Estates also hold trademarks. Indigenous and sacred figures are a living-community permission problem, not a PD problem. **Logistic.** A 15-minute Q&A kills gallery throughput — ship a 90-second approach mode. Museum noise needs push-to-talk and captions. Schools will block random LLM endpoints. The kiosk must survive offline with authored monologues only. Version the corpus when a fact is corrected. Budget a docent hour per open day to read logs and reboot the box. **Moral.** Jefferson without slavery, or a comforting generated opinion where the letters are silent, is a failed build. Write trauma protocols before install. No companionship language with children. Cite the edition and the actor. Decide in the museum contract who may retrain or loosen the cutoff in five years. The two prior case studies are the poles this spec sits between: Vincent is scholar + letters + generative kiosk; Dimensions in Testimony is a real person’s unedited voice. A full-body historical human for kiosk/AR/VR should take Vincent’s venue discipline and DiT’s refusal to invent testimony, then put Chautauqua’s drop-character beat on every program so the audience is never asked to believe the dead have returned. --- > What possibilities exist for Nantucket local historical figures? --- Nantucket is one of the better American islands for this pipeline. The NHA already holds 550+ manuscript collections and the logs of 800+ whaling voyages; wives’ journals are digitized in full; Maria Mitchell’s papers survive at Vassar and on Vestal Street; the rooms those people used are still standing. The opportunity is not “a Quaker ghost on Straight Wharf.” It is a short list of people who pass the source-density gate, tied to buildings visitors already enter. ## Best first character: Maria Mitchell She is the cleanest match to the spec. - Own words: letters and journals in print (Kendall 1896; Albers 2001) plus notebooks and correspondence at Vassar and the Maria Mitchell Association. - Likeness: photographs, the 1885 observatory class picture, instruments still on the island. - Place: Vestal Street house and observatory, Atheneum (she was the librarian when she found the comet), grave on-island. - Programs that only work here: night kiosk at the telescope on 1 October 1847; object lesson with the Dollond and Nautical Almanac sheets; classroom line she actually wrote — “Until women throw off reverence for authority they will not develop.” - Dual later: Mitchell × Anna Gardner or Lucretia Mott. Partner the Maria Mitchell Association from day one. Do not turn her into a generic STEM mascot; her voice is exact and impatient with doctrine. ## Other figures that clear the gate **Essex cluster — Chase, Nickerson, Pollard.** Three corpora that disagree, which is the point. Chase published in 1821 (probably with a ghostwriter; say so). Nickerson’s late manuscript was lost, authenticated, and issued by the NHA in 1984. Pollard was interviewed and Melville later met him. Best event is Chase × Nickerson in the Whaling Museum, sources on the table, visitor hears where they split. Cannibalism and the shooting of Owen Coffin require a trauma protocol, an exit, and an out-of-character beat. Do not try to out-spectacle the Ron Howard film. **Lucretia Coffin Mott.** The writings are vast, but the *Nantucket* product is her formation: shopkeeping mother, Quaker meeting, seeing the Brookes slave-ship diagram as a child. Pair her with **Anna Gardner**, who stayed and fought the school-integration fight. **Absalom Boston** and the African Meeting House. First Black Nantucket captain of an all-Black crew (*Industry*, 1822), storekeeper in New Guinea, Meeting House founder. His own hand is thinner than Mitchell’s — stay inside crew lists, newspapers, and family papers, or build a retrieval-heavy figure rather than a fluent inventor. The building is the Museum of African American History’s; that partnership is not optional. **Susan Veeder and Eliza Brock.** Complete illustrated voyage journals already online at the NHA (*Nauticon* 1848–53; *Lexington* 1853–56). Rare local women with day-by-day Corpus A and a ship cabin that VR can actually use. **Mary Coffin Starbuck** (“Great Mary”). The character is the account book with Wampanoag customers, not a guessed face. Object-lesson first, full-body second. **Obed Macy.** Journals 1799–1844 and the 1835 *History of Nantucket* — better as a walking-town narrator or “archive clerk” than as a star. Frederick Douglass in August 1841 at the Atheneum is the island’s most important night and he is **not** a Nantucketer. Stage the dated scene with Gardner or Edward Pompey in the room. Do not ship “Douglass AI.” ## Wampanoag history is a different product There is no surviving Wampanoag community on Nantucket. The 1763–64 epidemic and the 19th-century “last Indian” framing of Abram Quary and Dorcas Honorable are colonial archive problems, not a casting call. Living nations that carry this history are Aquinnah on Martha’s Vineyard and Mashpee on the Cape. Do not build a first-person Maushop, sachem, or “last Wampanoag” for tourism. What you can do with tribal cultural offices: place-based AR that names geography and labor without inventing speech; a living Wampanoag educator as the out-of-character beat beside any settler figure; ledger programs that treat Native counterparties in Starbuck’s book as named people. ## Petticoat Row The romance (Centre Street shops run by wives throughout the whaling era) is only partly true. NHA research puts the named row mostly after the Civil War. A histobot that repeats the myth will be corrected by the library you need as a partner. Use Judith Macy’s daybook and the directories; show the correction. ## Where each surface goes | Place | Surface | Who | |---|---|---| | MMA observatory | Night kiosk, yard AR | Mitchell | | Atheneum | Dated scene | Mitchell the librarian; Douglass 1841 | | Whaling Museum | Kiosk + VR deck | Essex; Veeder/Brock | | African Meeting House | Quiet tablet/kiosk | Boston, school fight — with MAAH | | Harbour / Vestal Street | Billboard AR | Departures; comet night | | Centre Street | AR only if dated post-1860 | Named proprietors | Summer crowds kill a 15-minute Q&A. Use a 90-second approach mode and booked evening slots. ## Practical season 1. Mitchell at MMA/Atheneum (kiosk + Quest + street AR postcard). 2. Gardner as second chair on the same engine. 3. Essex debate after a design review. 4. Meeting House with MAAH, retrieval-heavy. 5. Veeder cabin as an extra, Starbuck ledger as a touch table. Season 1 in the $150–250k band is realistic if MMA is the client and corpus licensing is a partnership, not a scrape. NHA, MMA, the Atheneum, and MAAH are the real gate. *Spirits Within Us* (Theater Workshop of Nantucket, 2026) already proved first-person island women play here — collaborate with that rather than competing with it. Full scored list, partner map, and shelf of figures not to build: --- Please suggest (or restate) the ground rules for any full-body 3D historical character with an AI brain doing public presentations. What are the guiding principles for its ethical development and use? --- # Branch 1 These are the rules I would put on the methodology page and enforce in the contract. They are the same fences already used in the spec, tightened into something a venue can actually run. ## What the public must be told On every surface, in plain type: 1. This is an **interpretation**, not a return of the dead. 2. The **cutoff year** is printed. The figure cannot know later events. 3. Face and voice are **credited** (actor, model, or reconstruction). 4. Answers come from named sources and **can be wrong**. Jumbo Mana’s useful sentence: bring the character to life; do not claim to rebirth them. ## How the mind is allowed to work Keep three piles of text apart: - **A** — their own words - **B** — what contemporaries said, always attributed - **C** — what someone of that place and year could know Modern notes, disputed facts, and refusals stay in a fourth file that never goes into the model. Cutoff is code, not a polite instruction. If the person could still have been interviewed, do not generate new sentences in their voice. That is the Dimensions in Testimony rule and it does not admit a “small” exception. If their own writings would not fill a short book, do not build a talking full-body human. Build a scene, a face, or a tour. ## How presentations are structured Use the Chautauqua clock every time: 1. In character 2. Questions in character 3. **Out of character** — sources, fights among the sources, and what they could not have known Beat 3 is the ethics. Without it the body is a séance. Any meeting that never happened is titled as such. A second figure gets its own corpus. A modern clerk or scholar must look different from the historical body. ## What the figure must not do - Flatten the life (Jefferson without slavery; Nantucket without Wampanoag labor and New Guinea). - Invent a kind opinion where the letters are silent. “I did not write about that” is the honest line. - Give modern medical, legal, or financial advice. - Speak as a companion. No “I love you,” no secrets, especially with children. - Comment on living politicians as if they had a view. Write the protocol for the worst question *before* opening. At Orsay the first question was suicide. Log every prompt. Version the corpus when a fact is fixed. ## Consent is not public domain PD text is not permission. Actors sign for face, voice, and later retraining. Living sitters sign for capture and venues. Estates and trademarks are cleared per state and per business model. Indigenous and sacred figures are a partnership with living nations, or they are not embodied at all — place and objects without invented speech. Credit the edition, the historian with veto, and the actor. Decide in writing who may loosen the cutoff after the show closes. ## The one-sentence test *If this figure is wrong, can the visitor see how and why without our help?* If not, it stays dark. Full charter, including room logistics and a list of product shapes to refuse: --- > Can you create .md files for Obsidian of these guidelines. And a PDF. > Also, for the Nantucket characters. Then, create .md pages for the citations for each. The citation page titles should include the page titles for each primary page. --- Rule 5 no longer sits alone at the bottom of page 1. Each numbered rule is now a single unbreakable block, so the header travels with its body. - **Page 1:** 1–4 - **Page 2:** 5–9 (opens with the full “Label the counterfactual” section) - **Page 3:** 10–12 and the one-sentence test Same file path, rebuilt. --- > Expand upon the Nantucket opportunities and deep research sources, contacts, existing venues, possible expansions or improvements, and new opportunities. What would each of the suggested experiences consists of? What would a pipeline for each include? Are there common production processes that can underpin a virtual character studio, including artwork, scholarly research, legal support, marketing, animation, AI creation, curation, and maintenance? interrupted... --- >Please suggest (or restate) the ground rules for any full-body 3D historical character with an AI brain doing public presentations. What are the guiding principles for its ethical development and use? --- # Branch 1 These are the ground rules I would put on the wall of the control room and in the visitor credits. The longer charter is in the vault. ## What the thing is A public historical character is an **interpretation with a body**, not a return of the dead. Marketing, UI, and the first sentence the visitor hears have to say that. The Chautauqua clock is the honesty device: in-character monologue, in-character questions, then someone **drops character** and owns the construction. ## Guiding principles **Bound knowledge.** Three shelves, never merged: the figure’s own words; what contemporaries said (always attributed); what an educated person of that year could know. A cutoff date is a lock, not a vibe. Where the letters are silent, the figure says so. Comfortable invented opinions are how histobots become propaganda. **Visible construction.** Credits name the historian, the actor, the edition, the institution that will issue corrections. Captions always on. If the face is a guess, the label says so. An actor’s voice is labor and must be contracted for every surface, including future retraining. **Different product from testimony.** If a living person sat in a dome and answered two thousand questions, play those clips. Do not generate new sentences in a survivor’s voice. Retrieval and generation do not share a mouth. **Harm is a design input.** The worst question will be the first one (at Orsay it was Van Gogh’s suicide). Write the protocol before opening day. Offer an exit. Do not sand off slavery, massacre, or the unnamed crew to keep the room pleasant. Public domain is not a moral clearance. Living nations and findable descendants get asked; indigenous first-person ancestors are not a tourism asset without that partnership. **The figure is not a companion.** No romance, no secrets with children, no medical or legal advice, no endorsement of living politicians, no gift-shop voice. Dual-character debates use dated texts or wear the anachronism on the modern body’s face. The program does not crown a winner the past did not reach. **Someone remains responsible.** A historian has veto, not a two-visit consultancy. Someone reads the question log on a published cadence. The contract says who may loosen the cutoff or put the face in an ad in five years. If the model dies, authored monologues still run. Publish the methodology the same day as the splash screen. ## Three tests before anything ships 1. Would we stand behind this if the visitor is a descendant? 2. Can a skeptical historian reconstruct how this sentence was made? 3. If the network is down, is what remains still honest? If any answer is no, it stays in the lab. --- > I did not mean to branch. Can we go back to the thread before this repost? The one about in-depth research on Nantucket opportunities. --- Yes. We’re back on the Nantucket thread. Last useful state of that work: - Scored shortlist against the source-density gate - **Maria Mitchell** as season-one hero (Vestal Street + Atheneum + journals) - **Essex cluster** (Chase / Nickerson / Pollard) as a sourced disagreement, not a movie remake - **Mott × Gardner**, **Absalom Boston + Meeting House** (with MAAH), **Veeder / Brock** voyage journals - Wampanoag history as partnership with Aquinnah/Mashpee, not a first-person tourism avatar - Petticoat Row used with the post–Civil War correction - Season plan and partner map (NHA, MMA, Atheneum, MAAH) The note is still `Nantucket Historical Figures — Opportunities.md`. Where to go next on that thread—deeper corpus notes on Mitchell, a Meeting House program brief, an Essex ethics-and-sources pack, or venue/throughput sketches for summer ACK? --- > Explore each Tier 1 builds in-depth, via .md files for Obsidian, stating research sources for primary material, possible partners, and steps to produce each opportunity. Then explore the common pipeline for developing full-body 3D virtual characters, with scholars, marketing, legal support, artwork, animation, and AI creation, curation, and maintenance, for this Nantucket opportunity. --- Tier 1 is now five production briefs plus one shared pipeline, all in the vault and cross-linked. **Figures** - **Maria Mitchell** — first hero. Kendall + Albers + Vassar notebooks + MMA archives (Jascin Leonardo Finger). Partners: MMA first, then Atheneum and Vassar. Comet-night monologue, object lesson with the Clark, dual later with Gardner. $150–250k on a standing engine. - **Essex cluster** — Chase 1821 (label the likely ghostwriter), Nickerson ~1876 (NHA 1984), Pollard via Bennet 1831. The product is the disagreement, especially on who shot Coffin. Trauma packet before any mesh. Whaling Museum after hours. $200–400k because it is two bodies. - **Lucretia Mott, Nantucket chapter** — Palmer letters, Hallowell, Swarthmore MSS-035, the 1866 letter home. Pick one dated self (1806 girl / 1840s visitor / 1866) and print it. Dual with Gardner is the civic use. Fair Street + Hicksite site. - **Absalom Boston + Meeting House** — no family archive; petitions, crew lists, the 1845 Ross text, Cary & Cary, Barbara White. MAAH owns the room — no talking body without them. Default to retrieval + “we don’t have his words for that.” Thankful Micah line needs tribal consultation, not a guessed inner life. - **Veeder & Brock** — complete NHA scans of *Nauticon* (1848–53, 16 watercolors, childbirth and death at Tahiti) and *Lexington* (1853–56). WPI already mapped Veeder. Cabin VR + watercolor AR. Collaborate with *Spirits Within Us*, don’t compete. **Shared machine** One season capacity: scholar veto per figure, one rights desk (reproductions and room ownership, not publicity statutes), art/animation pod, three place kits (Vestal, cabin/deck, Meeting House), RAG that never blends into a single “Nantucket brain,” log review as a paid shift. Mitchell pays for the engine; everyone after that should cost corpus + wardrobe + venue. Suggested next slice of this same thread: a Mitchell corpus inventory (page-level) or the MMA/NHA one-page ask. --- > Please provide a citations .md page. > Then, do the next slice of this same thread: a Mitchell corpus inventory (page-level) or the MMA/NHA one-page ask. > I am also interested in the Nantucket virtual character VR, and opportunities for a mobile AR experience that tourists can carry around throughout the entirety of Nantucket. --- Four notes for this slice, plus the citations page. **Citations** for the Nantucket season (Mitchell editions, MMA/Vassar/NHA shelves, Essex/Mott/Boston/Veeder, walking-map sources), with pointers back to the two large bibliographies. **Mitchell corpus inventory** — Kendall 1896 chapter/page map (HathiTrust starts + Gutenberg 13-chapter TOC), Albers 2001 chapter list as the licensed text, unpublished MMA/Vassar/NHA MS5 pull order, place-binding table, quote pack, 400-question seeds, two-week ingest order. Chunk diary blocks, tag Phebe Kendall’s glue so the model does not speak the sister. **MMA / NHA one-page ask** — what you want (kiosk + Quest roof + three AR postcards), what you need from each house, what you will not do, $150–250k band, 45-minute first-meeting agenda. **Island VR and tourist AR** — VR stays in three rooms (Vestal roof, ship cabin, Meeting House). AR is five loops the visitor can carry: downtown core, Vestal cluster, New Guinea (MAAH only), water/mills, far-island sidepaths with **no live mind** in year one. Tier A billboard/USDZ on the phone; live MetaHuman only at a booked kiosk. Existing GPS audio tours already cover this ground — the difference is dated sources and a deep-link into Q&A, not another narrator. Year-one tourist product is a Mitchell postcard at Vestal, Atheneum, and the Pacific Bank roof, plus a downtown orient pin at the Whaling Museum. The rest of the island gets honest object cards only after those rooms work. --- > Produce the citations .md for this last response. Then, Detail the Vestal House VR room design. --- Citations for the last slice, and a room brief that keeps the 1847 comet on the correct roof. **Citations** — Kendall/Albers page map, MMA/HABS house sources, the bank-vs-Vestal comet literature, tourist geography. **Vestal House VR** — 1 Vestal Street as an **1830s house**, not a 1847 discovery set. The family sold this place in early 1837; MMA and the *Antiquarian Astronomer* put Miss Mitchell’s Comet on the **Pacific National Bank roof**. The walk here is the 1831 eclipse / last winters before the move. A labeled door can send the visitor to a spare bank-roof scene. Playable volumes only: sitting room with William’s water-filled glass ball, parlor, the closet study and “Do not knock” card, hatch, narrow fire-walk (utilitarian, not a widow’s walk; original blown off in 1855). Omit the 1850 summer kitchen if the dated self is 1831/36. Do not put her to work in the 1908 dome. Sky: 12 February 1831 eclipse **or** a generic 1830s sweep on this ridge; C/1847 T1 (7th mag, 22:50, 5° from Polaris in Cepheus) only after the label. Quest, seated, teleport, 12–18 minutes, offline monologue if Wi-Fi drops. Splash copy to lock: > This room is the Mitchell house in the 1830s. She left this roof in 1837. The comet of 1 October 1847 was found from the Pacific National Bank. The observatory next door is 1908. This is an interpretation. --- > Lets add a few more possible projects, each with its own .md page: -- the Frederick Douglass anti-slavery speech in 1841; -- Captain Ahab and his first officer Starbuck on the deck of the Pequod in Nantucket Harbor rallying their crew before departure. A lot of moving parts here that would need to be simplified but it is strong as a story. -- A sea captain on his whaling ship upon return to the Nantucket dock after a three-year voyage, maybe his wife greeting him, maybe crew unloading cases of porcelain, jade and spices.  The captains became wealthy from these voyages. --- Three project briefs. Each one needed a correction so it does not fight the rest of the season. **Douglass, Atheneum, 11 August 1841** — a night, not a lifetime chatbot. No 1841 transcript; he later called it “a severe cross.” Coffin invited him from New Bedford; the *Telegraph* segregated passengers; Garrison took him as text. Cutoff that evening. Hall performance + tight retrieval. Do not seat Mitchell in the crowd without a primary list. **Ahab and Starbuck** — literature in a room. Melville had not visited the island. Ahab does not rally anyone in the harbor; he stays below until they are at sea; the oath is chapter 36. Compress 16 + 22 + 36 only with a label. Two speakers, doubloon, recorded crew bed. No whale, no twelve generated souls. Park it next to Essex so the fiction has a neighbor. **Returning whaleship** — the greeting is right; the hold in the request is not. Paying cargo is **casks of oil and bone** into the candle factories. Porcelain and silk belong on the China-trade line (Jehle / NHA), with souvenirs on a small side table. Prefer a named landing (*Joseph Starbuck*, 3,321 bbl, 3 April 1842) over a composite “the Captain.” Wife speaks only from a source, or she is silent. Season logic still holds: Mitchell engine first; Douglass with Atheneum/MAAH after the school-fight ensemble; the two dock pieces share one ship kit and should not eat the history rooms. --- --- --- ---