NOTE: There is no on-screen violence in this clip. It's all dialogue.
This is my first attempt using an image "seed" to maintain character consistency. The results were acceptable enough that I felt comfortable completing the project and sharing it.
Seeds are numerical values that control the randomness in image generation, acting as a starting point for the algorithm. Using the same seed with the same prompt and settings will consistently produce the same or a very similar image, while changing the seed introduces variations. By default, open-source apps like Forge and Fooocus use a value of -1 to produce random images.
For this project, I used Flux to create the base image for the brunette reporter working in a news room, and then used the seed from the base image to create a corporate headshot of the same character by changing the last digit in the seed. I also used Flux to create the base image for the blonde reporter. Next, I animated the brunette with Wan 2.5, and the blonde with Google's Veo 3.1 because Wan kept bugging out afterward - and I barely got the blonde's dialogue past Google's safety systems. Finally, used DaVinci Resolve to put the whole thing together.
Unfortunately, I couldn't get the blonde to sound mournful or distraught, and both models are too damn expensive for me to have kept coaxing more applicable emotion from her, so, after six attempts I thought, "Fine. Fuck it. She's emotionally detached."
View attachment Bay City Strangler.mp4
This is my first attempt using an image "seed" to maintain character consistency. The results were acceptable enough that I felt comfortable completing the project and sharing it.
Seeds are numerical values that control the randomness in image generation, acting as a starting point for the algorithm. Using the same seed with the same prompt and settings will consistently produce the same or a very similar image, while changing the seed introduces variations. By default, open-source apps like Forge and Fooocus use a value of -1 to produce random images.
For this project, I used Flux to create the base image for the brunette reporter working in a news room, and then used the seed from the base image to create a corporate headshot of the same character by changing the last digit in the seed. I also used Flux to create the base image for the blonde reporter. Next, I animated the brunette with Wan 2.5, and the blonde with Google's Veo 3.1 because Wan kept bugging out afterward - and I barely got the blonde's dialogue past Google's safety systems. Finally, used DaVinci Resolve to put the whole thing together.
Unfortunately, I couldn't get the blonde to sound mournful or distraught, and both models are too damn expensive for me to have kept coaxing more applicable emotion from her, so, after six attempts I thought, "Fine. Fuck it. She's emotionally detached."
View attachment Bay City Strangler.mp4
Last edited:
