From Sketch to Photoreal House With an AI Agent
A two-minute architectural workflow demo that proves an agent's work through tool captures and raw-to-finished render comparisons.
Field notes / why it works
A hand drawing a house in the first 3 seconds gives the workflow a tangible starting point, and the finished-house glimpse at about 0:04 establishes the destination. The middle backs up each claim with visible software stages, then paired raw and photoreal views at 1:42–1:50 deliver the payoff; the interior comparison at 1:47.1–1:48.4 is the strongest replay peak.
Plate 01 / The format in one breath
A human starts with a rough architectural sketch, then a first-person narrator shows an agent carrying that idea through site modeling, floor planning, material work and photoreal output. The story progresses by showing each tool's output on screen; a polished before/after comparison near 1:42–1:50 pays off the opening sketch.
Why it works
- The first 3 seconds show a hand completing the house drawing in close-up. The concept is visually clear before the narration promises a faster design workflow at 0:12.9.
- Each claim has visible proof: a workflow map at 0:19–0:27, Rhino geometry and plans at 0:45–1:23, Blender materials at 1:26–1:41, and raw/rendered comparisons at 1:42–1:50.
- The strongest replay peak, 1:47.1–1:48.4, lands on the interior before/after image; the result is easy to inspect and worth rewinding.
Plate 02 / Format card
| Platform · aspect · length | YouTube · 16:9 · 2:08 · measured source 640×360 at 30 fps |
| Pace | 13 measured shots; 12 cuts, 5.6–7.0 cuts/min allowing three uncertain changes; 22 within-shot steps, 15.9 total picture changes/min; mean shot 9.85 s, median 5.9 s. The opening composite lasts about 51 s. |
| Script | 252 spoken words; 125 wpm over full runtime, 166 wpm while speaking; 22 sentences averaging 11.5 words; no questions or second-person “you.” |
| Voice | Low-register, confident first-person designer/operator; conversational but precisely enunciated. No on-camera speaker. |
| On screen | Opening hand/sketch and finished-house glimpse (~9%); dark workflow diagrams and prompt/UI (~30%); Rhino/Blender screen recording and comparisons (~52%); finished environment/product close (~9%). Shares are estimates from sheets. |
| Captions | No burned-in speech subtitles. Only software UI, small step labels, diagram text, and final logo. |
| Sound | Continuous quiet tonal/ambient bed (music-like, inferred); close narration, no clear cut effects. Source: −24.4 LUFS integrated, −6.6 dBFS true peak. |
| Structure | Seven stages: idea → bottleneck/promise → inputs/agent setup → 3D form → floor plan/fixes → material and render comparison → lifestyle/product close. Main payoff 1:42–1:52; no verbal CTA; logo end card at about 2:05. |
Pro / 13 plates + recipe
The first plates are open. The full dissection is Pro.
Get the timed shots, type and sound specs, and the runnable recipe.
Open the full blueprint- Plate 03 / The hook, frame by frame (first 3 seconds)Pro
- Plate 04 / Structure (beat sheet)Pro
- Plate 05 / Shot grammarPro
- Plate 06 / Visual style guidePro
- Plate 07 / Voice and deliveryPro
- Plate 08 / Sound designPro
- Plate 09 / PackagingPro
- Plate 10 / Script templatePro
- Plate 11 / Production planPro
- Plate 12 / PromptsPro
- Plate 13 / Edit recipePro
- Plate 14 / QA checklistPro
- Plate 15 / Make it yoursPro
- Recipe / runnable promptPro