AI Halloween trend videos: how to make the zombie, Thriller and costume clips

Some AI Halloween guides focus on photo edits. Fello AI's guide, for example, covers photo prompts for ChatGPT and Gemini. This guide is for the video side: three formats, a starting prompt for each, and what a clip costs.

What is the AI Halloween trend in video form?

The video version of the AI Halloween trend turns a single photo of a person into a short spooky clip, and this guide covers three formats.

• Zombie transformation. The person in the photo slowly turns into a zombie.
• Thriller-style dance. The person, or a zombie version of them, does a group dance in the street.
• Costume reveal. The person changes into a costume, such as a vampire, witch or skeleton, in one move.

How do I make an AI zombie video?

An AI zombie video is made by giving an image-to-video model one clear photo and a prompt that describes a slow change, a fixed camera and a short length.

1. Pick a sharp, front-facing photo of an adult who agreed to be in it.
2. Describe the change as makeup and costume: "grey skin, sunken eyes, torn collar".
3. Say what the camera does: "locked camera, no cuts".
4. Keep the clip short, then extend only the take you like.

Starting prompt:

The person in the photo stands still in a dim street at night. Over five seconds their skin turns grey, their eyes sink and their jaw slackens into a slow zombie stare. Locked camera, no cuts, soft fog.

Fello AI reports that naming film characters or asking for gore can lead to refusals in image tools. Describing the costume gets around it.

How do I make a Thriller-style AI dance video?

A Thriller-style AI dance video is best made with a motion reference: you upload a short clip of someone doing the moves and the model applies them to your photo.

Starting prompt for the photo:

The person in the photo performs the dance from the reference video on a foggy street at night, zombie makeup, torn jacket, arms bent at the elbows, shoulders hunched. Locked camera, full body in frame.

Kling's Motion Control guide sets the limits for the reference clip as of 2026-10-26:

• 3 to 30 seconds long
• one person recommended; with several, Kling uses the one filling the most frame
• full body and head visible
• one continuous shot, no cuts, no camera moves
• steady, moderate speed

These limits are for Video 2.6 Motion Control. For the full method, see Kling Motion Control: how to copy a dance onto a photo. For pets, see AI cat dance videos and AI dog dance videos.

No reference clip? A prompt can describe the dance, with looser timing. Film your own reference with a friend and you use nobody else's performance.

How do I make an AI costume transformation video?

An AI costume transformation video is made by describing one costume change in one action, with a fixed camera, from a photo that shows the person's face and upper body.

Starting prompt:

The person in the photo raises a black cape across their face. When it drops, they are dressed as a vampire with pale makeup and a high collar. Locked camera, warm candlelight, one continuous shot.

Keep the costume description short and specific. Check the first result against your photo before you pay for more tries.

How much does an AI Halloween video cost?

On Runway's API, one credit costs $0.01, and a 10-second clip lists from $0.50 to $1.30 depending on the model, and Veo 3.1 tops out at 8 seconds for $3.20 (Runway pricing page, checked 2026-10-26).

Three tries is a planning allowance. These are list prices before tax and do not cover app subscriptions. Veo 3.1 on Runway accepts 4, 6 or 8 seconds.

Worked budget for all three formats

One 10-second clip of each format, with three tries per format, is nine generations.

• On Gen-4 Turbo: 9 × $0.50 = $4.50.
• On Gen-4.5: 9 × $1.20 = $10.80.
• Mix: zombie and costume on Gen-4 Turbo (6 × $0.50 = $3.00), dance on Gen-4.5 (3 × $1.20 = $3.60), total $6.60.

Check Runway's plan pricing and the AI video cost calculator before you budget. Rates change.

Which models can I still use for AI horror video?

Google shut down the Veo 3.1 preview models in the Gemini API on 22 October 2026, so older guides that point to them are out of date.

Google's page points to Gemini Omni Flash and to generally available models on its enterprise platform. For other current options, see Sora alternatives. Runway lists several models on one rate card, which is why the table above uses it.

Can I use someone else's face in an AI Halloween video?

Use only your own face or the face of an adult who has clearly agreed, because realistic AI video of real people needs platform disclosure, and harmful uses get removed even when labeled.

• TikTok requires labels on AI content that shows realistic-looking scenes or people. Unlabeled content may be removed, restricted or labeled, and harmful content is not allowed even with a label (TikTok, checked 2026-10-26).
• YouTube requires disclosure when realistic content makes a real person appear to do something they did not do. Clearly unrealistic content, such as someone riding a unicorn, does not need it (YouTube Help, checked 2026-10-26).
• Facebook and Instagram add an "AI info" label when they detect industry-standard AI signals or when you disclose (Meta, checked 2026-10-26).

Do not use celebrities, children or people who did not agree. A zombie version of an ex, a boss or a stranger is the kind of use these rules are meant to stop. Each video tool also has its own content policy.

What music can I use for a Thriller-style clip?

Use a track you made or have licensed, because TikTok asks creators to confirm there is no copyright-protected music in the post.

TikTok says that if you cannot confirm that, or that you hold all licenses for the songs and recordings, you should decline, and that the post might be taken down. Audio generated by a video model is not a licensed song, so check each track on its own terms.

How often will this page be updated?

Prices, models and platform rules are dated, and each update changes the date at the top. Facts were checked on 2026-10-26. For other formats, see the Learn hub.