The Seedance 2.5 zombie trend: prompts and tips

What Seedance 2.5 is, and why it suits the trend

Seedance is ByteDance’s family of AI video models, and Seedance 2.5 is the version this site uses for every zombie couple, pet and romance video. Searches for it rose alongside the zombie trend, for good reason:

• Reference images keep faces. You can attach photos and point to them in the prompt, so the survivor and the zombie partner each keep their own face, hair and build through the whole clip. The zombie hug only works if both people are recognisable.
• Timed beats. It follows a prompt split into seconds, which is how you get the hesitation, the leap and the hard cut to a memory inside 15 seconds.
• Sound included. Music and ambience come out with the picture, so the clip is ready to post.

Seedance zombie prompt 1: the original zombie hug

Attach the survivor as the first image and the partner who turns as the second.

@Image1 is the survivor and stays fully human; keep their face, hair, skin tone and build. @Image2 is the survivor’s partner, who has become a zombie: same face, hair and build, with pale grey skin, faint blue-grey veins, clouded white eyes, torn dusty clothes and slow stiff movement; their expression is tender, they recognise the survivor and are not aggressive. Abandoned city street at dusk, overturned cars, weeds in the asphalt, warm backlight and floating dust. 0–4 s: the survivor aims a pistol at the approaching partner, in tears, hands shaking, and never fires. 4–8 s: the partner stops, their eyes meet, the survivor slowly lowers the gun and opens their arms. 8–11 s: the partner runs and leaps into the hug; cut on the embrace. 11–15 s: a bright memory in a golden summer field, both healthy and human, hugging, laughing and spinning. Cinematic, realistic, shallow depth of field, light film grain, emotional piano and strings. No blood, wounds or violence; no on-screen text or watermark.

Seedance zombie prompt 2: the zombie dog

Attach a photo of you first and your dog alone second.

@Image1 is the owner and stays fully human; keep their face, hair, skin tone and build. @Image2 is the owner’s dog, who has become a zombie: same breed, colour, markings, size and collar, with matted, patchy fur, clouded white eyes and slightly stiff movement; gentle and not aggressive, it recognises its owner. Abandoned suburban street at dusk, warm backlight, leaves and dust drifting, dusty parked cars. 0–4 s: the owner aims a pistol at the approaching dog, in tears, hands shaking, and never fires. 4–7 s: the dog stops, tilts its head and wags its tail; the tag on its old collar catches the light; the owner lowers the gun, kneels and opens their arms. 7–10 s: the dog runs into the owner’s arms and is held tight; cut on the hug. 10–15 s: a bright memory in a sunny park, the healthy dog bringing back a ball, the owner laughing and rolling in the grass with it. Cinematic, realistic, emotional piano and strings. No blood, wounds or violence; no on-screen text or watermark.

More stories (wedding, rooftop, rain, ballroom) are in the zombie couple AI prompts, and the pet version is explained on zombie pet AI.

Tips for better results

• One face per reference image. Crop a couple photo into two. Group shots and small faces make the model mix people up.
• Keep it non-gory. Leave out blood, bites, wounds and attacks, and never let the gun fire. Gentle, sad wording passes moderation far more often.
• Write 15 s beats. Four short timed beats are about the limit; more and the action gets rushed or skipped.
• Spell out who stays human. Say it in the prompt for both images, or the model may zombify both people.
• Clear, front-facing photos. Even light, no sunglasses, no heavy filters.

Doing the zombie trend in Higgsfield or Google Flow

Higgsfield and Google Flow both came up in searches during the trend. At the time of writing, Higgsfield gives access to several third-party video models, and Flow is Google’s filmmaking tool built on its own Veo models. Which models, reference-image features, clip lengths and free allowances each offers changes often, so check their current terms. The general approach is the same: upload the two reference images, adapt the image tags in the prompts above, generate, and expect a few tries before the faces and the hug both land.

Or skip the prompt here

Every story on this site is a Seedance 2.5 prompt like the ones above, already tuned and tested. Upload your two photos, pick a story and get a 15 second MP4 with music and no watermark; failed videos return their credits automatically. Start with The Last Embrace or Good Dog, Still, or read the zombie AI trend guide first.

Make it without a prompt