Zombie love story AI. Two faces, one last embrace.

A real test: two photos, one reunion

We used the same two original, AI-generated fictional adult portraits for two Seedance 2.5 submissions through Kie. The first take missed the prop direction and dissolved between locations. The second removes the prop and simplifies the direction to two shots. The player shows the second take, without editing, compositing, upscaling or an added soundtrack.

Download original reference image

Download original reference image

Download either image to use the exact input file. Upload the survivor first and the partner second; these are identity references, not first and last frames.

Completed (UTC)

2026-10-15 15:54:17 UTC

Model / provider

Seedance 2.5 / Kie

Input workflow

Two image references · survivor, then partner

Requested settings

10s · 480p · 9:16 · audio on

Actual file

10.08s file · 480 × 854 · 24 fps · H.264 + AAC

Submissions / retries

2 submissions · 1 prompt revision · 0 edited versions

Provider-recorded time

210 seconds

Actual provider charge

280 Kie credits · US$1.40 at listed credit value

Total production consumption

560 Kie credits · US$2.80 at listed credit value · both submissions

Recreate in this workspace

60 site credits · about US$3.96 from Starter

The displayed take consumed 280 Kie credits. Including the rejected first take, production consumed 560 Kie credits, worth US$2.80 at the public list price; this is not a cash invoice. No site user account credits were debited. Running the displayed recipe here once is a separate quote: 60 site credits, about US$3.96 allocated from the US$9.90 / 150-credit Starter pack. The pack is paid upfront; videos are not sold individually. Reference-portrait creation is separate and its cost is excluded. Provider credit values ↗

The exact submitted prompt

Create a 10-second fictional cinematic reunion, vertical 9:16. Use exactly two separate shots with one immediate editorial cut between them. Image 1 defines the adult woman survivor in the rust-red jacket, always screen left. Image 2 defines her adult male partner in the olive-green shirt, always screen right. Preserve both distinct faces, hair, clothing and natural anatomy. Both have empty hands throughout.
[0-5s] SHOT 1: A locked, chest-height medium two-shot in an abandoned kitchen with cool blue window light. Both people are seen from a front three-quarter angle. The male partner has subtle pale theatrical zombie makeup with intact skin. They recognize each other, take one small step together, and gently embrace. Keep each face visible beside the other's shoulder, looking slightly toward the camera. Hold the embrace by the end of this shot.
[5-10s] SHOT 2: Make an immediate hard cut to a DIFFERENT LOCATION: a sunlit golden field before the outbreak. Begin already in the same gentle embrace with the same left/right positions, clothing, camera height and framing. Both are healthy, with normal skin. Keep BOTH faces visible at three-quarter angles, no back-facing head. Hold the warm memory to the end.
The transition is an abrupt film edit between two distinct backgrounds, NOT a gradual transformation. No dissolve, crossfade, morphing scenery, transparent walls, double exposure or overlapping locations. No weapons, props, injuries, attacks, extra people, text, dialogue or music. Only quiet natural ambience.

What this take achieved — and what it missed

• The revised staging is simpler. Both people are unarmed, their reference outfits stay legible, and they move into a gentle embrace. These are actual frames from the second output.
• The direct cut improved. At about five seconds, the cold kitchen changes directly to the golden field without the first take’s gradual overlap. This is one observed result, not a guarantee of exact timing on the next generation.
• Face visibility still falls short. The partner turns away in the memory despite the request to show both faces. His healthy-face appearance and identity after the cut cannot be fully assessed. Do not use this sample to promise continuous face accuracy.
• The rejected take is included in the cost. Removing the prop and reducing the direction to two shots were both changed in the second prompt, so this pair of runs cannot isolate which change caused the improvement. There were two submissions in total, not a single perfect first attempt.

Prepared for SeedanceAI.dev with AI assistance. Review used original frames, media metadata and provider task records. Two submissions are a case study, not a success-rate benchmark. The same inputs and settings can produce a different result; no fixed seed was exposed. Audio-stream presence was checked, but audio content was not independently scored.

What makes the zombie love story work?

The emotional story has four beats: recognition, a choice to approach, an embrace, and a memory. Subtle theatrical zombie styling keeps the partner recognizable. The warm scene represents the past; keep the pose and screen sides consistent so the transition reads clearly.

Original prompts to try

The revised kitchen-to-field prompt was tested in the case above. The three variations below are original, untested prompts; their outcomes are not implied by that sample.

Create a 10-second fictional reunion, vertical 9:16. Image 1 defines the adult owner; Image 2 defines their dog or cat. Preserve the owner's face and the pet's species, coat pattern, muzzle, ears and relative size.
[0-3s] In a quiet, abandoned doorway under cool evening light, the owner kneels. The pet approaches with dusty fur and subtly cloudy eyes, no wounds or aggression.
[3-5s] The owner holds still with open arms. The pet gently rests its head against the owner's chest. Use a medium side view, natural animal anatomy, no human hands on the animal.
[5-10s] Hard match cut at contact to the same pose in a sunny garden before the outbreak. Restore the pet's natural fur and healthy eyes. The owner smiles while the pet leans against them. Keep camera height and body positions. Natural garden ambience, no speech, music, weapons, injury or text.

Create a 10-second fictional zombie romance, vertical 9:16. Image 1 is the healthy adult survivor on screen left; Image 2 is the adult partner on screen right. Keep their identities distinct.
[0-3s] At an empty railway platform after rain, the survivor recognizes the partner's familiar scarf. The partner has subtle pale zombie makeup and shadowed eyes, intact skin. Amber lamp against cool blue mist.
[3-5s] They take one slow step toward each other and share a gentle hug. Chest-height locked camera; show natural hands and readable faces.
[5-10s] Cut on the embrace to the same platform on a warm autumn morning before the outbreak. Match their pose and screen positions; both are healthy and hold the embrace. Quiet rain changes to birds and breeze. No dialogue, song, blood, attacks, weapons, extra people or lettering.

Create a 10-second fictional friendship story, vertical 9:16. Image 1 defines the adult survivor; Image 2 defines their adult best friend. Keep both faces and hairstyles distinct, with Image 1 left and Image 2 right. The relationship is platonic.
[0-3s] Outside an abandoned roadside diner, the survivor recognizes their friend despite subtle pale theatrical zombie makeup, shadowed eyes and dusty clothes. Cool grey daylight, empty hands.
[3-5s] They open their arms and share one quiet, natural hug. Locked medium camera, no spinning or complex movement.
[5-10s] Hard cut at contact to the same hug at a sunny road-trip overlook before the outbreak. Match shoulder positions and camera height. Both friends are healthy and smile. Soft outdoor wind, no dialogue, song, wounds, weapons or text.

From photos to a first take

1. 01

   Assign two normal photos

   Image 1 is the survivor; Image 2 is the partner or friend. Use one adult per photo with a clear face. For the pet prompt, use the owner first and the animal second. Start without zombie makeup in the source image.

2. 02

   Choose the story and inspect the prompt

   The default is a 10-second, 9:16 Seedance 2.5 photo-reference task at 480p. Pick the couple, pet, station or friends prompt. Keep the left/right roles consistent and simplify the hug if the characters overlap.

3. 03

   Review the whole take

   Check each face before and after contact, the hands, the pet’s anatomy where relevant, and whether the ending reads as a memory. Generate separate shots and join them in an editor if the single-pass cut does not work.

Plan the match cut before generating

What does a test cost?

Seedance 2.5 photo-reference tasks use the same duration and resolution rates shown by this workspace. A higher resolution is a new paid generation, not an upscale of the previous clip.

Starter is $9.90 upfront for 150 credits. It covers two 10-second 480p tasks (120 credits), with 30 left. A completed render you choose to redo uses credits again. Provider-confirmed failed tasks return their credits.

Fix the part that failed

Questions before you generate

Sources and workflow notes

External pages explain the trend or provider inputs. Their templates, footage, pricing and results are separate from this workspace.

• Media.io — Zombie love story: roles, prompts and continuity ↗
• Kie API docs — Seedance 2.5 reference inputs and task options ↗
• Kie API docs — Task state, timing and creditsConsumed fields ↗

Plan your next video