The author tested several AI video generation models via the Bothub platform to create a talking avatar from a static photo. The goal was to generate a short video of the user dressed as a pirate inviting people to a Neptune festival.
Wan 3.0 Prime provided dynamic image transformation and good facial expressions, though it altered the input text. Grok Imagine Video 1.5 produced a 1-second clip at a cost of 11.84 RUB per second, ignoring the prompt's text and showing a 'plastic' appearance.
PixVerse V6 successfully performed a pirate costume transformation on video and followed the text instructions, though lip-syncing was imprecise. Veo 3.1 resulted in inconsistent video orientation and strange facial gestures, costing 355.38 RUB for 4 seconds.
LTX Video, the cheapest option at 14.53 RUB for 3 seconds, failed to produce a usable result, generating only visual hallucinations. Meanwhile, the Kling model rejected the user's photo due to safety filters regarding the use of real human images.
