FreyaVideo Hotel Lobby AI turns two performer photos into a short orange studio rap performance using a built-in reference video. Users do not upload a reference video or write a prompt. This page records our own tests on September 30, 2026; it is a first-party product report.
We used two original, AI-generated fictional portraits: a man in a red shirt and a woman with a blue shirt and short black hair. The original studio reference was 4.064 seconds, with two performers and rap audio. These samples do not establish accuracy for every real person's photograph.
The tested editing endpoint was fal-ai/kling-video/o3/pro/video-to-video/edit. The updated workflow supplies two separate identity elements and a video reference, with explicit left/right mapping. The dedicated page handles these settings on the server.
| Test | Observed result | Provider processing time |
|---|---|---|
| Earlier image-reference edit, blue portrait left and red portrait right | Left performer changed, but the right performer retained the source identity. Visual quality check failed even though the provider reported a completed task. | 625.177 seconds |
| Updated two-element identity edit, same left/right assignment | Sampled frames at 0, 1, 2 and 3 seconds showed the blue-shirt performer on the left and red-shirt performer on the right, without an observed identity swap or extra performer. | 247.047 seconds |
The updated sample was a 4.042-second, 1920 × 1080 MP4 with H.264 video and AAC audio. Processing times above measure these individual provider runs; queueing, storage and download preparation can add time. They are not a typical-time guarantee or a benchmark against competing tools.
This original FreyaVideo test output shows the red-shirt performer on the left and the blue-shirt performer on the right. It comes from an earlier successful test, not the updated two-element run described above.
We checked visible identity placement in sampled frames. We have not measured facial similarity across a representative population or phoneme-level lip synchronization. Exact choreography and mouth movements are not guaranteed. A completed provider task can still fail a visual quality check.
Use one well-lit, unobstructed face per photo. Review the quoted credits before generating and check the result before sharing. See pricing, refund rules and support for billing questions.