# Pepe2Pepe — The other side

The 20-second V2 remake is **video-v2.mp4**. The original `video.mp4` is preserved.
`hero.png` is the exact opening poster, 1920 × 1080. The remake uses five
principal compositions, original official artwork, genuine hydrated browser
captures and a new continuous narration performance.

## Editorial approach

| Time | Composition |
| --- | --- |
| 0–4 s | Bespoke navy/teal cinematic stage; intact large official two-Pepe logo, immediate URL and “A question. Two sides.” Subtle layered camera drift, then settle. |
| 4–9 s | One continuous real homepage view, drifting down into open Robinhood markets #12 and #10. Both questions and odds remain readable; the move settles and holds. “Your question. Someone else’s take.” |
| 9–14 s | Real market #10, question and genuine Bet NO / Bet YES controls. One restrained push, then a hold. “Pick a side. Fixed odds.” |
| 14–17 s | Clearly illustrative IMD typographic motif on the same sculptural stage. “Powered by the IMD Oracle.” Secondary copy explicitly acknowledges operator review. This is not a transaction or settlement interface. |
| 17–20 s | Return to the opening stage with “Take the other side.” and large URL. The original logo resolves gently by 17.47 s; the complete end card then holds for more than 2.5 s. CTA and URL are fully opaque for all three seconds. |

The first three edits are deliberate cuts after camera motion has settled.
Website text never cross-dissolves over other website text. The Oracle lettering
and returning brand artwork fade separately through the same clean background,
so they do not ghost over one another. No sweep lines, wipes, sliding panels,
progress bars, flashing effects or rapid caption changes remain. Typography
stays fixed; the story also works muted.

## Sources and product accuracy

Fresh Chromium captures were acquired **2026-10-02 UTC**, after hydration,
without changing page styling, text, balances or market state. The homepage's
real Crypto filter and responsive two-column layout expose two actual markets.
No wallet was connected, no stake was entered and no transaction was sent.

- [Homepage](https://pepe2pepe.fun/)
- [Public product guide](https://pepe2pepe.fun/guide)
- [Robinhood market #12 — Quotrons](https://pepe2pepe.fun/robinhood/market/12)
- [Robinhood market #10 — IMD](https://pepe2pepe.fun/robinhood/market/10)
- [Official two-Pepe logo](https://pepe2pepe.fun/brand/pepe2pepe-v2.png), retained from `production/assets/logo.png`, preserving its proportions and original lettering.

The current guide describes public question creation and funding a side at
fixed odds; others can back that side or take the opposite side. Earlier backing
matches first and accepted returns remain fixed. Funds are held on-chain on
Ethereum or Robinhood. No NFT is required. The IMD Oracle researches results,
and operator review is the current default before settlement. The film makes
no guarantee of profit, immediate payout or autonomous settlement. The markets
shown are open, never presented as wins. Capture metadata, raw page text and
expanded FAQ evidence are preserved in `production/v2/captures/` and
`production/v2/product-sources.md`.

## Voice, music and artwork provenance

New narration: **ElevenLabs Eleven v4 through the installed fal.ai connector,
voice Roger**. A second complete take used **Chris**. Roger was selected for
longer measured pauses and fit to the edit. The entire coherent take starts at
6.20 s; it is not segmented or sped up. Processing is high-pass filtering,
loudness normalization and constant gain. Neither V1 audio nor its Edge Guy
voice is used. Both candidate recordings, generation IDs, exact settings and
source links are preserved in `production/v2/audio-notes.md`.

Script: “Your question. Someone else’s take. On Pepe to Pepe, pick a side at
fixed odds, with funds on-chain. Powered by the I M D Oracle. Take the other side.”

The instrumental is an **original 96 BPM** composition generated locally from
editable synthesis: soft damped-string keys, restrained harmonic pad, rounded
bass and sparse percussion. It uses no external songs, loops or samples. The
voice sits forward of the music, with one quiet Oracle accent and no repeated
transition whooshes. The pre-encode stereo mix measures about −18.44 LUFS and
−2.78 dBTP. Source score and the instrumental stem are included.

The text-free cinematic environment was created with the built-in **imagegen**
tool. Its exact prompt and provenance are in
`production/v2/assets/hero-environment-provenance.md`. The generator did not
draw any logo, text, character or website UI. Official artwork and Space Grotesk
typography are composited independently. The existing font's OFL license is
retained in `production/assets/SpaceGrotesk-OFL.txt`.

## Format and checks

Measured delivery: exactly 600 decoded frames / **20.000 seconds**,
**1920 × 1080**, **30 fps**, H.264 High in MP4, yuv420p, AAC-LC stereo at 48 kHz
(192 kb/s encoding target), faststart. File size: **4,121,931 bytes / 3.931 MiB**.
The complete file decodes without errors and its `moov` atom precedes `mdat`.
Final AAC mix measures **−18.47 LUFS integrated / −2.90 dBTP**.
Probe data, SHA-256 hashes and validation results are recorded in
`production/v2/verification.json`. V1's hash is unchanged.
Chromium also played the complete file muted at 1×: 600 presented frames,
zero dropped frames, normal completion at 20.000 s. The machine playback result
is saved in `production/v2/browser-playback.json`; it is a compatibility check,
not a claim of human audiovisual review. `production/v2/review.html` supplies a
local player for reviewing the delivered file.

Visual review covered the complete narrative at one-second intervals, both
sides of every cut frame by frame, every frame of the Oracle-to-hero transition,
full-resolution principal compositions, and the final CTA. The market crop
was repaired after review to preserve the token tile, tab icon, complete
question and readable odds. No overlapping website text or black gap was found.

## Honest review limitations

There is no usable audio device, audio-perception tool or continuous audiovisual
perception available to the editor in this environment. Both expressive voice
takes were generated and compared using duration, pauses and timing; **they
were not auditioned by ear**. Naturalness, pronunciation and the complete mix
are therefore not claimed to have passed listening review. Likewise, visual
review uses full-resolution frames and ordered transition frame sequences,
not a claimed normal-speed audiovisual viewing. Codec and waveform checks do
not certify creative quality. No robotic fallback has been substituted.

## Editable, offline production

`production/v2/render.py` contains the five-composition timeline and camera
paths. `production/v2/audio.py` renders the original score and mixes the saved
voice recording. All consumed media are supplied locally. Run from the project
root:

```sh
python production/v2/render.py --preview
python production/v2/render.py
```

FFmpeg is supplied by the execution environment. A compact, self-contained
Pillow runtime is included as an ordinary archive in `production/v2/vendor/`,
with the original licenses, file manifest and reproducible repacking script.
It retains the imaging, variable-font and WebP engines and their linked
libraries; unused optional engines and debug symbols are omitted. The bootstrap
verifies its checksum and extracts it into disposable `test/scratch/`, without
network access or changes to project dependencies. It targets CPython 3.14 on
Linux x86_64; other platforms need a compatible preinstalled Pillow environment.
NumPy is no longer needed: editable music synthesis uses Python's standard
library and FFmpeg. Browser capture is optional acquisition, not a render
dependency; saved sources make rendering fully offline.

The original music stem, mix master and both complete voice candidates are
retained. The compact music renderer preserves the composition, timings and
voice performance, but uses a different seeded noise generator for the brush
and room sound. Regenerated music is therefore not sample-identical to the
saved master; see `production/v2/audio-notes.md`. The delivered movie still
uses the unchanged original V2 mix.

For a non-destructive offline rebuild check:

```sh
python production/v2/render.py --output test/scratch/offline-video.mp4 --hero-output test/scratch/offline-hero.png
python production/v2/audio.py --output-dir test/scratch/audio
```

## Source bundle repair

The complete V2 source tree is **8,047,411 bytes / 7.675 MiB**, below the 8 MiB source limit. The oversized full wheels and unused research screenshots were removed. Raw
research text, timestamps, URLs, DOM evidence, editable timeline and score,
all consumed brand/media assets and required runtime libraries remain.
The cinematic background and market #10 capture use lossless WebP storage.
The homepage source retains all pixels used by its camera move, with extra
resampling margin: original rows 1700–2819 of the 1800 × 3648 capture. Decoded
RGB pixels were checked byte for byte against the original PNGs. The crop's
original position is restored on an in-memory canvas, preserving the exact
camera arithmetic. Its padding is outside every rendered viewport; no invented
website pixels appear in the film.

A complete 600-frame rebuild using `python3 -B -S` (site packages disabled)
produced a byte-for-byte identical MP4 and poster from the compact sources.
The film, poster and audio masters are byte-for-byte unchanged by this repair.
The required output files remain untracked and are delivered separately from
the source bundle. Detailed size and offline rebuild checks are recorded in
`production/v2/bundle-verification.json`.