On October 1, 2025, Krea.ai began providing access to a new text-to-video model called Wan 2.5. As revolutionary as the tech may be to some AI enthusiasts within the fetish community I belong to, it remains a helluva long way off from meeting my standards for good quality video production with regard to depicting graphic violence. Until that happens, I decided to use my Krea subscription to experiment with Wan’s built-in audio generator by creating survivor interviews.
I must admit, the automated audio syncing is quite remarkable, but dynamic vocal inflections and punctation are quickly lost if your dialogue takes up more than five seconds of speech because Wan has to fit every word within that window of time. Krea offers a way to extend videos an extra 5 seconds after initial generation, but my attempts kept returning an error, so I trimmed the dialogue as much as I could without removing the meat of what I wanted the characters to say and the images I wanted them to conjure up in viewers’ minds. The hair, makeup, outfits, shoes, and living room settings were all part of my prompts.
NOTE: All clips below were upscaled to 1080p from the 720p videos generated on Krea.
Chloroformed – 5sec Wan 2.5
Strangled – 5sec Wan 2.5
Thankfully, Krea began offering native 10-second video generation soon after the Wan 2.5 model was released, and the extra five seconds makes a huge difference in audio playback quality when characters don’t have to rush your dialogue, especially when adding dramatic pauses. It’s a significant improvement over the 5-second clips above.
Considering her ordeal, I decided to put the character in the first clip below in a more conservative blouse, and I love tie-neck blouses.
NOTE: The base images for the clips below were created with Gemini and then uploaded to Krea for reference. The second clip, First Kiss, was created for a bit of extra fun. It was inspired by a scene in a 2005 movie called Roger Dodger, which took place in a park. My clip features an AI celebrity on a talk show answering a fan question. Wan even simulated the audio quality of the microphone.
Strangled – 10sec Wan 2.5
First Kiss – 10sec Wan 2.5
