# Fusion Tables storyboard revision 2

10 concepts, 5 individually generated portrait images each. No videos generated.

## Your feedback applied

| Feedback | Revision |
|---|---|
| White table's near leg looks too long | S01/S02 use a distant, nearly broadside camera and grounded, balanced front legs. |
| Table leg crosses a bench | S03/S04 have no benches or chairs near the table. |
| S04 cover becomes L-shaped | One complete rectangular cover moves left in smaller successive increments. |
| More granular movement | Five stages per sequence, with explicit intermediate positions. |
| S05 chairs should disappear at the end | Frame04 shows a restrained chair-only dissolve. Frame05 has no chairs. |
| S06 benches are too small and distracting | Removed throughout all five frames. |
| S07 three stacked layers make no sense | Replaced by one assembled oak cover. |
| Two more transition styles | S09 foreground curtain wipe and S10 plate-to-pocket editorial match cut. |

## Proposed films

S01 Dinner rises, play revealed: 12 seconds.
S02 A ribbon of light: 12 seconds.
S03 Daylight lift and exit: 14 seconds.
S04 Reflected evening: 12 seconds.
S05 The Warm Reveal: 12 seconds.
S06 Cellar Dinner Sweep: 12 seconds.
S07/S08: see gallery timing.
S09 Curtain pass: 9 seconds.
S10 Plate-to-pocket match cut: 9 seconds.

Timings are editorial proposals. They do not claim any generation provider offers these exact clip durations. S09/S10's fifth keyframe begins at 6 seconds and holds through the 9-second ending.

## Better-video suggestions

1. Start with S09 or S10. Hide the functional switch under a real foreground curtain or make a clean shape-match cut. These are my lower-risk picks for believable product ads.
2. Build short simple shots, then edit them into a 9–14 second film. Keep each shot to one subject action and at most one camera move. OpenAI's [Sora prompting guide](https://developers.openai.com/cookbook/examples/sora/sora2_prompting_guide), checked 2026-09-09, recommends concise clips and simple, clearly timed motion. This is creative guidance, not a claim about current account access.
3. Keep the table base and room fixed. Separately control the cover, dining props and chairs where possible. Do not ask the whole frame to transform. A rigid cover should retain its thickness, rectangle, seams, orientation and shadow.
4. Start the reveal within two seconds. Let the audience recognize both functions, then hold a clean endpoint for a short call to action. Test this pacing rather than assuming longer is better.
5. Add restrained sound and typography in editing. One music accent, a subtle ball click or a soft fabric pass can make a modest transition feel cinematic. Use properly licensed audio. Do not generate logos or claims into the product footage.
6. For an orbit, keep the table's function fixed during a small camera arc. Do the function switch on a cut. Combining a large orbit, cover removal, moving furniture and changing rooms creates unnecessary failure points.
7. Validate every motion frame for changing leg lengths, furniture intersections, pocket closures, drifting rails, cover deformation, texture flicker and returning props. Approved stills are not proof of correct generated motion.

## Fidelity and scope

Built-in ChatGPT image generation used the supplied original Fusion Tables photos and newly generated anchors. No external generation API, Higgsfield run, video generation, purchase, publishing or ad launch was performed.

One assembled moving cover is a simplified advertising treatment. The actual manufacturer's panel count, storage and removal procedure have not been demonstrated. Suspended surfaces, optical wipes and chair fades are editorial VFX, not automatic product features.

Conspicuous reported errors were corrected and all selected frames visually inspected. These are photoreal concepts, not engineering-verified renderings. Fine hardware, seam locations and exact pixel registration can still vary. Before product-accurate motion, use a verified product plate or manufacturer model. A short real conversion reference would materially improve fidelity.

## Files

- index.html: offline-capable review gallery
- gallery-data.js / gallery-data.json: ten sequences with five stages
- manifest.json: full image prompts, source paths and generation provenance
- white/, black/, warm/, pergola/, new/: selected PNGs and per-batch QA
- Fusion-10-storyboards-v2.zip: portable gallery, selected frames and documentation

The previous revision remains at ../storyboards-20260909/. No original product photo or generated original was deleted.