Become a founding member.

Michael AubryMichael Aubry

Halloween Swap: The Wick Widow

Blueprint
Halloween Swap: The Wick Widow

Swap the person in a real clip for an original Halloween character, and keep the room real.

The board hides the actor inside a depth mask before the model sees anything. The mask carries her motion, a character sheet carries the new look, and the camera, light and room stay the original footage. The model never sees her face, so none of it leaks into the swap.

How it runs: trim your clip, mask the whole person, color the depth inside the mask, make a character sheet and an in-room anchor still, then Seedance 2.5 swaps your character in. Every step and its cost is on the board.

Source clip: Mixkit #44314 (free license). The Wick Widow is an original character.

Fork to your canvas makes a full copy in your account, every node and edge on a fresh canvas. Add to an existing workflow drops it into one of yours as a single collapsed node: its inputs become ports on the left and its outputs ports on the right, so you wire it like any other node.

Instructions

# How Real Swap works (v2: hide the person, keep the room)

Real Swap puts your character into a real clip and keeps everything else real: the camera, the light, the grain and the room. Raj is the worked example, swapped into a clip of AI Michael at his desk. Swap in your own clip and your own character.

**Why the person disappears first.** Show a video model a clear person and it copies what it sees: their face shape, their hair, their clothes. So before Seedance sees anything, this board hides the whole person inside a colored depth shape and leaves the real room around it. The shape carries the motion. Your character sheet carries the look. This is the method from Sirio Berati's open-source Genjutsu workflow (MIT), built as one board.

1. **Your clip** (Import, Video Trim). One continuous shot, one person, knees up or closer, hands mostly away from the face. Trim it to the window you want (4 to 8 s works well). The Mute node makes a silent copy for the model; set its scene length to your window (seconds x 24) and the Audio split range to match (0-7 for 7 s).
2. **The sound** (Audio split, Audio Pad). Your clip's own audio, kept on the side and put back at zero offset at the end, so nothing drifts. 0 credits.
3. **The mask** (Segment Mask). One word, person. SAM 3 tracks the whole body every frame. Scrub it: if a hand drops out for a frame, the model invents a new hand.
4. **The depth** (Depth Map, colored). Warm is close, cool is far. That is the motion without the face.
5. **The composite** (Masked Composite, free). Depth only where the person was, the real room everywhere else. Grow 30 covers hair and edges; largest region only stops specks turning into extra people.
6. **The swap** (Seedance 2.5, reference mode). @Video1 = the composite, @Image1 = your character sheet, generated audio off. The prompt is short and blunt: replace the whole person, keep the real background, take the role from the clip, not the clothes. Write the model's misses into the prompt as "never" lines and rerun only this node.
7. **The finish** (Video Editor + Side by side). The swap with your clip's original audio, and original | depth | swap to judge it.

What still breaks: hands crossing the face, two people in one shot, very fast turns. Lip sync follows the audio closely but is not frame perfect. Short clips can come back a touch short: the Seedance card warns you when it does.

Costs: every paid step shows its price before it runs. Mask + depth are a few dozen credits; Seedance is most of the bill and scales with the seconds.

Inputs and what's baked in

Overridable inputs become sockets on the left of the node. Baked ones are served as outputs, so you get the author's asset without supplying your own.

  • Import: Mixkit #44314 "Young woman exploring a haunted mansion" (Mixkit Free License, 1080x1920, no audio)
    input:image · VIDEO
    Overridable input
  • Segment prompt: person (the WHOLE person)
    input:text · TEXT
    Overridable input
  • Swap prompt: THE WICK WIDOW. Identity first, replace the whole person, keep the room. Add "never" lines for misses
    input:text · TEXT
    Overridable input
  • CHARACTER SHEET PROMPT · The Wick Widow (original character, asset sheet FIRST)
    input:text · TEXT
    Overridable input
  • ANCHOR PROMPT · paint the Widow into this exact frame, lit by the room
    input:text · TEXT
    Overridable input

See it run

Real outputs from this Blueprint's nodes.

Mute: the silent copy of your clip for @Video1 (Video Editor, 0 cr). Set its length to your window: seconds x 24
Segment Mask: the whole person, every frame (SAM 3)
Depth Map: COLORED (warm = close, cool = far). The motion, without the face
Seedance 2.5 reference mode (720p): @Video1 = the composite, @Image1 = Wick Widow sheet, @Image2 = Widow lit in this room, audio off
CHARACTER SHEET · GPT Image 2.5 Sunburst, 2560x1440 turnaround (one take)
CHARACTER SHEET · GPT Image 2.5 Sunburst, 2560x1440 turnaround (one take)
ANCHOR · GPT Image 2.5 Sunburst Edit: the Widow in the real room (frame + sheet)
ANCHOR · GPT Image 2.5 Sunburst Edit: the Widow in the real room (frame + sheet)
Start here

Ask Claude for your next ad.

Connect once. Ask any time.

  • No credit card
  • You keep the rights to what you make
  • Cancel anytime