Our sample videos were made by the Zombie AI Video team with ByteDance's Seedance 2.5 from the two AI-generated portraits shown on this page. Both people are fictional. Each example below shows the input photos, the finished video and the settings we used.
Examples
AI zombie love story — couple version, 16:9
Couple version in 16:9: Leo holds the gun, Mia turns. The reveal by the snowy window, the lowered gun, her run across the room — then a hard cut mid-stride, and she is still running, alive, into his arms in a golden meadow. 24 seconds, made with Seedance 2.5 as a cold clip and a warm clip joined on the same step; original score.
AI zombie love story — vertical 9:16 version
The same story framed for phones in 9:16: the reveal by the snowy window, the lowered gun, her run across the room — then the hard cut lands mid-stride and she runs on, alive, into his arms in the meadow. Seedance 2.5, original score.
AI zombie pet video — dog version, 9:16
Pet version in 9:16: the zombie dog in the doorway, the brass tag, the dropped bat — then a hard cut mid-stride, and the same run carries on across a sunset beach into her arms. 16 seconds, made with Seedance 2.5 as a cold clip and a warm clip; original score.
How these were made
- We generated fictional portraits with an image model.
- We wrote the shots from our shot list (copy the prompts).
- Couple versions: an 11-second cold clip and a 13-second warm clip from Seedance 2.5, joined where she is mid-stride; the warm clip starts from an edited copy of that frame.
- Sound: an original score in the key of the trend's soundtrack, with the breathing, scream and laughter separated from the model's own audio.
- Compressed for the web (H.264). No other edits.
What we measured while making them
- Join two clips on one step. Generating the cold and warm halves as two clips, and starting the warm clip from an edited copy of the cold clip's mid-stride frame, kept the leap moving in the same direction across the cut. Our first one-take 24-second version flipped the couple's left-right positions at the cut.
- “No music” is ignored. Writing “no music, only natural sound” in a Seedance 2.5 prompt did not remove the model's own score: an AudioSet sound classifier still rated the soundtrack 70% music. Keep the sound on, split the voices out with an audio separator, and put your own track under them.
- Weapon words can block frame edits. In Dreamina's image model, one of our two frame-edit prompts containing “no gun” was rejected as a policy violation; with the weapon words removed it went through. The other passed with them, so moderation is not consistent — leave weapon words out of frame edits.
- What it cost in Dreamina. Seedance 2.5 in Dreamina charged about 64 credits per second at 1080p (832 credits for a 13-second clip) and returned 1080p video at 60 frames per second.
- The music key matters. The trend's original soundtrack sits in B-flat minor / D-flat major. Moving our AI-generated cues into that key raised their audio-similarity score (CLAP) to the original from 0.62 to 0.70 for the cold half and from 0.78 to 0.80 for the warm half.
Measured October 5, 2026.
Frequently asked questions
Were these AI zombie video examples made from real people?
No. Both people and the dog were generated as fictional portraits first; the videos were then generated from those portraits.
Which model made these examples?
Seedance 2.5. The couple versions are two clips (11 s cold, 13 s warm) joined on the same running step; the pet version is two clips too (9 s cold, 7 s warm). The music is an original score.
Can I make a video like these?
Yes — copy the exact prompts from our prompt page or use the two-photo generator on the home page.