Zombie Hug Notes
When a trend blows up, the first question in every comment section is "what app is this?" For the AI zombie trend, there's no single official app. The clips online come from many different video models and tools. This post explains what one specific tool, Zombie Hug, uses, and why the choice of AI model for zombie videos matters more than people think.
Zombie Hug uses Seedance by ByteDance to generate every video. The model receives your two photos and a fixed story text for the scene you choose, then makes a 15-second, 9:16, 720p video with sound. You can't write your own prompt, which keeps results consistent. The details are on Zombie Hug's page about the AI models it uses.
Free-form prompting sounds more flexible, but for a trend like this it's usually worse. The zombie hug has a specific structure: the dark opening, the recognition, the run, and the cut to sunlight. A fixed story text means every video hits those beats. With three scenes to choose from (ruined street, pumpkin patch or marigold field), you still get variety, and every scene uses the same model at the same price.
Orders go to Seedance through a main API provider. If that fails or is slow, the same order is sent to the same model through a second provider. In other words, the backup doesn't quietly switch you to a weaker model.
An automated NSFW classifier checks uploaded photos and generated videos. That's standard practice, and it's worth asking about for any tool that handles photos of real people.
People often ask whether they can skip a dedicated tool and use a general chatbot or video app. Some can produce zombie images, and a few can make short clips, but getting two specific faces, the recognition beat and a clean match cut in one go is hard. Zombie Hug wrote an honest comparison of whether ChatGPT, Gemini or Sora can do the zombie trend, which is worth reading before you spend time prompting.
Written by Eason. Disclosure: I'm on the Zombie Hug team.