Floom launch film
Nothing is muxed into the film yet. These are just decision surfaces: pick the voice you want, and tell me whether the extra SFX earn their place. Same 0–13s excerpt, loudness‑matched, so you’re judging cleanliness and timbre, not volume.
The current VO — your raw recording with only an ffmpeg EQ/de‑ess/compress chain. This is the one you said sucks: room tone, a little boxy.
Your exact same take, run through ElevenLabs’ voice isolator. It re‑derives the voice from the speech itself, so the room coloration/reverb riding on the words (not just in the pauses) comes off — that’s the part that made the raw read feel boxy. Delivery and identity untouched. The honest “your voice, cleaner” fix.
Speech‑to‑speech re‑synthesis targeting a clone of your own voice (high stability/similarity). Cleaner and more even, but it’s a re‑generation — listen for whether it still sounds like you or drifts into a studio‑narrator version.
Same music/VO bed, but with 5 restrained, generated SFX added at spots that were bare: a pre‑reveal air whoosh before the name (21s), a spawn pop on “single prompt” (30s), a trigger pulse on “event triggers” (37s), a soft lock on “secure and sandboxed” (41s), and a warm sparkle on “open source” (47s). All ~−22 to −25 dBFS — punctuation, never cheese. Full length; skip to those timestamps to A/B.
VO stem processed: v4/v4_voice_clean.wav (your assembled raw take, 63.7s) → ElevenLabs /v1/audio-isolation, then /v1/speech-to-speech toward a clone of your own voice.
SFX generated via ElevenLabs /v1/sound-generation. Beds built on top of the finished mix_v5 — the film comps are untouched. Nothing shipped; awaiting your two picks.