Guide: AI Comedy Scene in 90s Style(Midjourney → Claude → Seedance)
What we're making
In 3 steps you get a 20-second AI comedy scene in 90s style: two women in fur coats in Courchevel pretend to read newspapers while actually checking out every man who walks by.
The whole process:
- Midjourney: create the characters, using our moodboard code for the 90s look.
- Claude: get a precise Seedance prompt for the scene.
- Higgsfield (Seedance 2.5): upload the character images as references, paste the prompt and generate the video.
What you need:
| Tool | What for |
|---|---|
| Midjourney | Character images (references) |
Claude + the higgsfield-seedance-prompt skill | Writing the video prompt |
| Higgsfield, Seedance 2.5 model | Generating the video with sound and dialogue |
Step 1: Characters in Midjourney
Create each character as a separate image, always ending the prompt with the moodboard code --profile uxo17vd . The code adds the 90s film look, color and grain, so you don't need to write them in the prompt.
Character 1: brunette (sits left)
1990s fashion editorial photo at a luxury ski resort in Courchevel, a model with long voluminous dark-brown blowout hair, oversized black oval sunglasses, ivory fur hat, long ivory faux fur coat, cream leather gloves, gold earrings, white knee-high boots, seated on a wooden director's chair in the snow with a newspaper, cool confident expression, snowy pine trees, ski slope and wooden chalets in the background, bright sunny day, ivory and cream tones --ar 9:16 --profile uxo17vd

Character 2: blonde (sits right)
1990s fashion editorial photo at a luxury ski resort in Courchevel, a model with platinum-blonde hair in a sleek low bun, small gold-rimmed tinted sunglasses worn low on the nose, long cream shearling coat with white faux fur collar, black trousers, black patent boots, gold hoop earrings, seated on a wooden director's chair in the snow holding a newspaper, dry amused expression, snowy pine trees, ski lift and wooden chalets in the background, bright sunny day, cream white and black tones --ar 9:16 --profile uxo17vd

Tips:
- The face must be clearly visible. Pick an image where the character faces the camera and nothing covers the face.
- If Midjourney blocks the prompt (“AI Moderator is unsure”), remove ages, skin and lip descriptions and words like “young woman”. Write it like a fashion editorial: “a model with…”. Sometimes simply running the same prompt again works.
- Note what Midjourney actually made (lipstick color, earrings, trousers or boots). You'll need it in step 2.
Step 2: The Seedance prompt in Claude
Upload both images to Claude and describe the scene, and Claude writes the full Seedance prompt. We use the higgsfield-seedance-prompt skill: it writes the prompt in blocks (characters, location, timed shots, camera, acting, light, sound, dialogue) and locks the faces and the characters' positions. You import the skill into your Claude once, then use it for every video.
Importing the skill (one time only):
- Download the skill's ZIP file: higgsfield-seedance-prompt.zip.
- In Claude, open Customize > Skills (on some plans Settings > Capabilities) and make sure Code execution and file creation is turned on.
- Click Upload skill (or “+”, then “Upload a skill”) and select the downloaded ZIP. Don't unzip it.
- Turn on the
higgsfield-seedance-promptskill in your skills list.
The message to send Claude (replace the parts in brackets):
Use the higgsfield-seedance-prompt skill. Write a Seedance 2.5 prompt for a comedy scene in 90s style, 20 seconds, vertical 9:16, with sound. The characters are in the attached images: the first image is @image1 (sits left), the second is @image2 (sits right). Scene: [describe what happens, e.g. two women in Courchevel pretend to read newspapers and comment on the men walking by, and at the end someone catches them]. The dialogue is in English. End the scene with a punchline or a cliffhanger.
Before moving on, check the prompt:
- Character descriptions match the images (hair, sunglasses, lipstick, clothes). If not, tell Claude what's different and ask for a fix.
- The language is locked to English in the ACTIVE REFERENCES, AUDIO and POSITIVE LOCKS blocks. Otherwise Seedance may pick another language on its own.
- Tag order: @image1 and @image2 match the order you'll upload the images in.
Step 3: Generating on Higgsfield (Seedance 2.5)
On Higgsfield, open video generation and choose the Seedance 2.5 model. Then upload the images as references in the exact order, paste the prompt and set the format.
- Upload the references in order: brunette first (becomes @image1), blonde second (becomes @image2). Order matters, because the prompt's tags are assigned by upload order.
- Paste the whole prompt from Claude, without shortening it.
- Set: 9:16 format, 20 s duration, sound on.
- Generate and review the result.
Don't add music in Seedance. Add it in editing, so it doesn't drown out the dialogue.
What the prompt Claude writes looks like
The prompt isn't one paragraph but a series of blocks, and each block controls one part of the video. That's how Seedance knows exactly who is where, what happens in each second and what must stay the same.
| Block | What it controls |
|---|---|
| SCENE CONTEXT | What happens, where, and who sits where |
| ACTIVE REFERENCES | Which image is which character (@image1, @image2) and their voice |
| LOCATION MAP | Foreground, midground and background of the scene |
| FORMAT MODE + CAMERA | Number of shots, cuts and camera movement in each shot |
| ACTION | What happens in every second, with the dialogue |
| PERFORMANCE + PHYSICS | Acting, facial expressions and how objects behave (newspapers, fur, snow) |
| LIGHTING + COLOR GRADE | Light and color |
| AUDIO | Scene sound and dialogue language |
| POSITIVE LOCKS | What must stay the same: faces, positions, exact lines |
Here's what one shot looks like in the ACTION block:
0.0s to 2.5s — Shot 1: both women sit hidden behind raised newspapers. A beat of stillness. Then, in perfect sync, both newspapers lower 20 cm, revealing two pairs of sunglasses peering over the top edges. 2.5s HARD CUT














