A private listening room.
Sign in to compare the recordings.
Sign in to compare the recordings.
| Model / profile | Rejected cases caught | Input | Outcome |
|---|
Five human-rejected recordings are the labeled cases. Catching a rejection means model score below 3 or an explicit reject. Replacement clips have no human acceptance label; they cannot establish false-positive rates. These correlated cases do not establish a winning model.
Effort is selected by BrightWrapper from the requested quality or speed profile. Actual routing evidence stays in each receipt; profiles are not guaranteed to produce different effort levels.
The active scene agent directs the selection. The current Puppet Sprites pipeline chooses licensed recordings from a curated Freesound / BigSoundBank catalog, trims and normalizes them, obtains a blind audio-model description, then reviews the full narration-and-music mix. Earlier effects include ElevenLabs-generated audio and local synthesis. New production direction is stock-only; generation here is an isolated experiment and changes no film. Pipeline source
| Option | Cost model | Useful for | Trade-off |
|---|---|---|---|
| BigSoundBank | Free · CC0 | Quick real-world recordings | Listen to the exact excerpt; a matching title is insufficient. |
| Freesound | Free · per-file license | Specific materials, actions and perspectives | Quality varies; retain attribution and license for each file. |
| Soundly | Free / paid plans | Search and audition a curated library | Convenience cannot guarantee the chosen effect fits the scene. |
| Pro Sound Effects | Subscription / buy & own | Broader professional coverage and variations | Choose library and license before buying; nothing purchased here. |
| Phone Foley | Existing phone + time | Exact rhythm, contact and perspective | Record multiple takes; rooms and phone processing color the sound. |
| ElevenLabs via fal | Metered generation | Rapid variations and designed textures | Prompt compliance is uncertain; compare the waveform and actual mix. |
Record in a quiet, soft-furnished room, with the phone roughly a forearm from the action and away from handling noise. Capture several gentle-to-firm takes and a few seconds of room tone; use a safe rigid plastic surface rather than striking fragile glass. Keep the original, trim a copy, and match the rhythm to the picture. Layer a low thud with a short rattle only if the result needs weight, then audition beneath the actual narration at normal listening volume. Clean, contextually convincing Foley matters more than a physically literal source.