| 1 | # The Last Visa R2V example |
| 2 | |
| 3 | This directory contains the production R2V requests used to create the selected |
| 4 | shots for *The Last Visa*. Each file under `requests/` is one request using the |
| 5 | same field names as the online `/api/r2v/generate` endpoint. |
| 6 | |
| 7 | - `condition_img` is an optional first-frame condition and does not consume a |
| 8 | memory slot. |
| 9 | - `memory_slots` is ordered and contains at most seven entries. |
| 10 | - Every local memory slot has an `image_url`; it either has an `audio_url` or |
| 11 | explicitly sets `audio_mode` to `empty`. |
| 12 | - Paths are relative to the request JSON, so the example remains portable. |
| 13 | - The selected output videos are intentionally not copied into this source |
| 14 | repository. `index.json` preserves their original sequence and provenance. |
| 15 | |
| 16 | The six audio files are exact inputs from the online Last Visa case. They are |
| 17 | voice-filtered again during condition encoding, matching the online R2V path. |
| 18 | |
| 19 | Prepare all text/image/audio conditions before loading the DiT: |
| 20 | |
| 21 | ```bash |
| 22 | python inference.py \ |
| 23 | --config configs/inference.bf16.yaml \ |
| 24 | --condition-encode \ |
| 25 | --conditioning-cache-dir conditioning_cache/the_last_visa |
| 26 | ``` |
| 27 | |
| 28 | Run one request from that cache: |
| 29 | |
| 30 | ```bash |
| 31 | python inference.py \ |
| 32 | --config configs/inference.bf16.yaml \ |
| 33 | --request examples/the_last_visa/requests/009_01_shot_008_nathan_replies_to_elena_r2v.json \ |
| 34 | --conditioning-cache-dir conditioning_cache/the_last_visa |
| 35 | ``` |
| 36 |