The decisions I need from you
Ten gating questions live in the Input Form. Partial answers unblock the build — any subset of responses moves Phase 1 forward. The three highest-leverage decisions:
The 9-stage production pipeline
Your team's existing Lumion + LiDAR + SketchUp/Rhino feeds the pipeline at Stage 1. The integration layer (Stages 2-7), AI variant layer (Stage 5), and cinematic + show control (Stages 8-9) are net-new — operator-owned with a freelance UE5 partner for the cinematic finale.
W0-W8 Phase 1 build
Eight-week sprint. Day-30 mid-point: variant pipeline working end-to-end (drop a Lumion render in Syncthing → 4 on-brand variants in your portal in 20 minutes). W8: first real S1 session with a friendly UHNW pilot client.
What's yours, what's net-new
Your firm already markets photorealistic 3D and LiDAR-driven precision (per mhudsondesign.com). The studio vision isn't 0→1 — it's level-up. Honoring existing capacity shrinks PDF 4's 7-9 freelance bench to 2 named outside contractors (UE5 generalist + sound designer per cinematic).
Already in-house · REUSED
- SketchUp + base 3D modeling (you / draftsman)
- Lumion rendering — already in portfolio
- LiDAR + 360 photo capture — your associate
- Construction management discipline
- Visual taste + portfolio language — the substrate for the portfolio LoRA
Net-new this engagement
- Gaussian splatting (Nerfstudio + PostShot)
- AI variant pipeline (ComfyUI + Flux + ControlNet + LoRA)
- UE5 cinematic + Path Tracing — freelance partner
- Show control automation (OBS + Companion + Tailscale)
- Client portal stack (Outline + Easy!Appointments + Umami)
- Voice narration + sound design (Chatterbox / ElevenLabs + DaVinci)
- Orchestration backbone (Windmill + NocoDB + Speckle + Syncthing)
- Stage 0.5 Vibe Walk — pre-capture GTM lever
Cinematic fidelity options
Your call said "98% real visual fidelity" for the narrated movie finale. UE5 Path Tracing is the only place in the validated stack where that ceiling is computed, not approximated. Four options trade fidelity against cost and operator-ramp risk.
Option A — Full UE5
Hits 98% photoreal verbatim
$8-15K Flagship sprint
~$2-3K/mo steady-state
Matches call commitment
Option C — Hybrid Lumen
Credible fallback · UE5 as in-house
$3-5K narrow SOW
Lower steady-state cost
"Real-time cinematic" framing
Option B — AI-Composited
High-velocity MVP option
$0 contractor
~$1-2K per cinematic
Temporal artifacts visible
Option D — Defer cinematic
Surfaced for transparency · not recommended
$0 Phase 1
Misses call commitment
Highest engagement-thesis risk
The 5-7% photoreal gap between best real-time engines and path-traced UE5 lives in glass+water reflections, subsurface scattering (backlit leaves, limestone, marble), volumetric atmospherics (god rays through palm canopies), and camera realism (depth-of-field, motion blur, lens distortion). To a $25M-house buyer these are not subtle differences. Sprint cost ($8-15K) is rounding error against $75K/yr builder membership × one closed builder.
Phase 1 budget envelope
Lean-OSS stack saves ~$5,500-6,200/yr in software vs PDF 3's $6,513 procurement, plus $899 one-time on show control. Net Phase-1 external spend lands at ~$9-19K all-in (Option A baseline) — recoups against any single closed builder.
| PDF 3 line item | Substitute | Annual Δ |
|---|---|---|
| D5 Render Pro ($360/yr) | Drop — Lumion already in-house | −$360 |
| Luma AI Pro ($900/yr) | Nerfstudio + PostShot ($0 software) | −$900 |
| Midjourney + Veras ($1,210/yr) | ComfyUI + Flux + ControlNet + LoRA | −$1,210 |
| Resolume + DisplayFusion ($899 one-time) | OBS + obs-websocket + Companion | −$899 once |
| Monday.com ($684/yr) | Keep for your client-team workflows; add Windmill + NocoDB layer underneath | $0 net |
| Total recurring savings | −$2,470/yr + $899 once | |
What I'm tracking
Seven named risks with mitigations in place. R6 (Flux license) and R7 (scope-shift mental anchor) are the two new since 2026-05-20.
| Risk | Mitigation | When I'll know |
|---|---|---|
| R1 · Portfolio LoRA quality regression | Train at rank 16; 20-40 hand-picked images; regression test on held-out set; your verbal sign-off | End of W1 |
| R2 · Outline self-host needs SSO IdP | Validate IdP availability in W0; static-site fallback (Hugo/Astro on Cloudflare Pages) | W0 Q-list |
| R3 · AV vendor blocks Tailscale UDP | Q-list to AV vendor W0; fallback ZeroTier/NetBird/RustDesk; local-only ultimate fallback | W0 Q-list |
| R4 · UE5 partner sourcing slips | Upwork brief in W0 not W2; 3 candidate interviews; ArchVizPRO Discord backup | W2 SOW deadline |
| R5 · UHNW reality test (W8) | You select friendly first pilot; 48-hr written incident log | W8 |
| R6 · Flux.1-dev non-commercial (new) | Q4 forced choice with three commercial-clear paths; recommend Flux.1-Pro API for Phase 1 | Q4 answer |
| R7 · Scope-shift mental anchor (V12) | Q10 written confirmation before W2 contractor SOW signs | Q10 answer |
Opportunity surfaces: productization rights retained (case-zero pattern reusable across non-competing geographies); Stage 0.5 Vibe Walk = differentiator no competing landscape studio has; existing-capacity dividend; your 30-year design taste encoded as portfolio LoRA = "they understand my aesthetic" before first session.
Stage 0.5 — pre-capture Vibe Walk
Approved 2026-05-17 as additive Phase-1 addendum. Converts a sales-call site photo + 3-sentence brief into a walkable portal artifact within 30 minutes for <$1.50 per walk.
| Use case | Trigger | Time-to-artifact | Cost |
|---|---|---|---|
| Prospect-to-builder pitch | Builder forwards UHNW site photo + brief | 30 min | <$1 |
| In-discovery aspiration capture | Live call; prospect says "Tuscany meets coastal Florida" | Within call | ~$1.25 |
| Builder-membership renewal lever | Builder needs evidence of capability | Half day | <$1 total |
If §1 matches what you remember from the call
I want to start W0 work this week. The first concrete thing I'll do is post the UE5 partner brief and begin the conversation with your render lead on Lumion Additional Output enablement.
Reply with as much or as little as you have time for. The 10 questions are designed so any subset of answers moves the build forward. The shape-confirmation (Q10) is the one I need most.