# Selected photos to a generated trailer

Three public-domain USFWS archive photos stand in for a small selected camera-roll album. The original JPEGs, credits and source pages are preserved in `sources.json`. No private photo library or messaging conversation was accessed.

`brief.txt` is an authored single-message example. `generation-request.json` is the actual expanded request submitted through Vaaya to Seedance 2.0 Fast reference-to-video. `generation-result.json` contains the actual successful response with account balances omitted. `output.mp4` is the unchanged downloaded provider output, with no local assembly or re-encoding. `video-metadata.json` contains its ffprobe measurements.

The requested scene order was lake, forest, then bison, with approximate four-second shots. A generative model is not a frame-accurate editor: compare the result against the original inputs. This is invented motion, not recorded travel or wildlife footage. The source agency did not create or endorse it.

Observed charges in this workflow: three Files imports at 1 cent each, two new fal image uploads at 1 cent each (the lake's existing staged source was reused), and 134 cents for the successful silent generation: 139 cents total for this trailer task. These are run receipts, not future price quotes. The prior lake staging belonged to the earlier photo-to-video article.

Failure history retained for transparency: the attempted CueFrame media imports and project creation were declined by the merchant's payment route, with no successful assembly. We used the supported reference-to-video generation route instead. An initial audio-enabled generation was rejected by the provider for potential copyright in generated audio and charged zero cents; the final request asks for a silent video. No audio from that failed job was used. No local ffmpeg assembly was used.

`editorial.webp` is a separate original conceptual illustration, not a frame from the generated video. File integrity hashes are in `manifest.sha256.json`.