How I make cozy anime cooking shorts with AI

A simple workflow for creating 10–15 second cozy anime food videos with strong hooks, consistent characters, clean food continuity, and fewer AI glitches. It's the exact process behind my Reels, including the four prompts I give ChatGPT.

The style

My workflow

1. Start with a 1–2 second hook

   Use a relatable or visually interesting moment before the food: a blackout, rain outside, a cat interrupting, running out of snacks, being too cold to leave bed.

2. Write a tiny story

   Think hook → decision → preparation → payoff → cozy ending. The story should be understandable without dialogue.

3. Create the storyboard first

   I make a storyboard before generating the video. It locks character placement, food appearance and shot order. Treat it as the visual source of truth.

4. Keep the shots readable

   For a 15-second video, 8–12 shots works well. Give important moments 1.5–3 seconds instead of changing scenes every half-second.

5. Write a strict generation prompt

   Describe exactly what happens in each time range. Add explicit continuity rules for the character, ingredients, props, hands and utensils.

6. Generate, review, refine one problem at a time

   If something breaks (extra hands, duplicate food, a laptop in the wrong state, strange candle physics), rewrite that exact action more literally instead of making the whole prompt more complicated.

My ChatGPT workflow: four prompts

I use ChatGPT as the creative director before I generate anything. I do all four steps in one chat, so it remembers the character from start to finish. Copy each prompt, then adjust the details to your own idea.

Step 1: Create your character

I would like to create 15-second cozy anime videos for Instagram Reels. Create a character for me.

Make it a character reference sheet:
- a large cinematic portrait
- front, side and back full-body views
- a color palette
- expressions: normal, happy, curious, sleepy
- a few poses doing everyday cozy things

Give the character one signature detail (a prop or pet) that can appear in every video so viewers recognize them.

A reference sheet with front, side and back views gives every later step something concrete to match. The signature detail matters more than it looks: a white mug with a tiny black cat, or my tuxedo cat, appears in every video, and viewers start recognizing the series by it.

Step 2: Get the video idea

Give me an idea for the first video.
15 seconds, 9:16 vertical, no dialogue.
Open with a strong 1-2 second hook.
Break it into scenes with timestamps, and note any on-screen text and sound for each scene.

You'll get a short story split into scenes with timestamps, on-screen text and sound. If the opening is slow, ask ChatGPT for a stronger hook before moving on.

Step 3: Create the storyboard image

I like the idea. Create a storyboard image for that video using the same character.
The storyboard should contain 8-12 shots which will be used as the reference image in Seedance 2.0.
Label each panel with a shot number and timecode only.
Keep the character's face, hair, outfit, the room and the props identical in every panel.

Check it before moving on. If a panel shows the finished dish too early, an extra hand, or the character in a different outfit, ask ChatGPT to fix that panel now. It's much cheaper than fixing it in the video.

Step 4: Write the Seedance 2.0 prompt

Now give me a prompt that will guide Seedance 2.0 to generate the video from this storyboard. Keep it under 4,000 characters.
Use these sections: character consistency, visual style, shot sequence with timestamps, camera and transitions, audio, ending, and a final list of things to avoid.
Treat the storyboard as the primary visual reference.
No dialogue.

A good answer follows this structure. If something is missing, such as the continuity rules, ask ChatGPT to add it:

Create a 15-second vertical 9:16 cinematic cozy anime short titled "[TITLE]", following the attached storyboard as the visual guide.

CHARACTER CONSISTENCY: Keep the same [character: hair, eyes, outfit, personality] in every shot. Preserve facial features, hairstyle, outfit, body proportions and colors across all shots. [Signature detail, e.g. a white mug with a tiny black cat.]

VISUAL STYLE: Hand-painted anime aesthetic, cozy slice-of-life atmosphere, [lighting], soft shadows, detailed backgrounds, subtle film grain, [color palette]. Fluid motion and consistent anatomy.

SHOT SEQUENCE - 15 SECONDS TOTAL:
Shot 1 (0-1.5s): [Framing]. [One clear action]. [Camera move].
Shot 2 (1.5-3s): ...
[One shot per storyboard panel, 8-12 shots.]

CAMERA AND TRANSITIONS: Smooth, restrained movement, subtle push-ins, slow pans and natural cuts. Avoid abrupt transitions, excessive motion or fast editing.

AUDIO: [Music + ambient sounds, e.g. kitchen ASMR, rain, mug clink]. No dialogue or voice-over.

CONTINUITY: Same ingredient and portion throughout. One utensil at a time. Maximum two human hands visible. Do not show the finished dish before it is made.

ENDING: [Final shot. Exact on-screen text in quotes, or "no text".]

IMPORTANT: Follow the storyboard's sequence and preserve the character design. No extra characters, no costume changes, no distorted hands or faces, no duplicated limbs, food or props, no flickering, no watermarks and no unintended text.

Here's a full prompt ChatGPT wrote for a 9-shot morning-routine video, so you can see what "finished" looks like:

Create a 15-second vertical 9:16 cinematic cozy anime short titled "A Slow Morning in His Little World," following the attached storyboard as the visual guide.

CHARACTER CONSISTENCY: Keep the same young male anime character in every shot: messy dark brown hair, warm brown eyes, oversized dark olive-green hoodie, beige cargo pants, gentle expression, and relaxed personality. Preserve his facial features, hairstyle, outfit, body proportions, and colors across all shots. His signature white ceramic mug has a small black cat illustration.

VISUAL STYLE: Beautiful hand-painted anime aesthetic, cozy slice-of-life atmosphere, warm golden-hour sunlight, soft shadows, detailed backgrounds, subtle film grain, natural character movement, cinematic composition, and a muted palette of olive green, cream, amber, and sunset orange. Peaceful, nostalgic, comforting mood. High-quality animation with fluid motion and consistent anatomy.

SHOT SEQUENCE - 15 SECONDS TOTAL:

Shot 1 (0-1.5s): Medium shot. In his cozy bedroom, the boy wakes up, stretches gently, and rubs his sleepy eyes as morning sunlight enters through the window. Slow camera push-in.

Shot 2 (1.5-3s): Side-profile shot. He sits beside the window and quietly watches the sunrise over a peaceful town. Curtains sway gently in the breeze; dust particles float in the sunlight.

Shot 3 (3-4.5s): Close-up of his hands pouring hot water into his white cat mug. Steam rises naturally; warm sunlight glows on the wooden desk. Gentle pouring motion.

Shot 4 (4.5-6s): Close-up. He holds the steaming mug with both hands, takes a small sip, closes his eyes, and smiles softly. Cozy, intimate framing.

Shot 5 (6-7.5s): Medium shot. He sits at his desk, quietly studying or writing beside an open laptop. Subtle hand movements and a focused expression.

Shot 6 (7.5-9s): Close-up. Wearing black headphones, he closes his eyes and smiles as he enjoys his favorite music. Slight head sway, relaxed shoulders, warm light on his face.

Shot 7 (9-10.5s): Medium close-up. He lifts a small camera and takes a photograph of the beautiful view outside. Natural hand movement and gentle camera motion.

Shot 8 (10.5-12s): Wide rear shot. He sits by the window or balcony, gazing at a breathtaking orange-pink sunset over distant mountains and glowing town lights. Slow cinematic pullback.

Shot 9 (12-15s): Close-up profile against the sunset. He raises his gaze slightly toward the sky and gives a peaceful, contented smile. Slowly pan across his face, then gently fade out.

CAMERA AND TRANSITIONS: Use smooth, restrained camera movements, subtle push-ins, slow pans, and natural cuts. Keep each shot visually distinct while maintaining consistent lighting and geography. Avoid abrupt transitions, excessive motion, or fast editing.

AUDIO: Soft lo-fi piano with gentle ambient morning sounds, birds outside, quiet tea pouring, mug clink, faint keyboard typing, and a soft breeze. Music gradually becomes warmer and slightly fuller toward the sunset. No dialogue or voice-over.

ENDING: Display the exact text "Good things take time. ♡" over the sunset during the final shot, with clean, legible typography.

IMPORTANT: Follow the storyboard's sequence, preserve the character design, prioritize believable motion and emotional subtlety, and create a polished, relaxing Instagram Reel. No extra characters, no costume changes, no distorted hands or faces, no flickering, no random objects, no duplicated limbs, no watermarks, and no unintended text.

Generating it in Higgsfield

I generate these videos with Seedance 2.0 inside Higgsfield. If you sign up through my link, you get up to 50% off for 3 hours from the moment you log in, so have your plan picked before you click. My setup:

• Text to Video, with Seedance 2.0 selected
• Storyboard image uploaded as the reference image
• Duration: 15 seconds
• Aspect ratio: 9:16 vertical
• Motion: subtle and cinematic
• Audio generation on, if available

Model availability and settings change over time, so if you're reading this later, check that Seedance 2.0 is still available in Higgsfield.

Why I create the storyboard first

My storyboard checklist

• 8–12 shots for a 15-second video
• Strong visual hook in the first 1–2 seconds
• One main action per panel
• Same character design in every panel
• Same food shape and portion throughout
• No final dish appearing before it is made
• Simple camera angles and readable motion
• Final shot feels different from the opening

Continuity checklist

• One main food item or batch
• Same plate, bowl or pan throughout
• No duplicate cooking action
• No finished dish shown early
• No more than two human hands
• One utensil at a time
• Cat stays realistically sized
• Props keep the same state between shots

What usually causes bad generations

Too many actions in one shot. "Pick up the pan, turn, pour, stir and smile" gives the model too many chances to fail.

Complicated transitions. Simple cuts are usually more reliable than magical camera tricks.

Ambiguous object states. If a laptop must be open, say "screen upright, keyboard visible." If a match must already be lit, say explicitly that it ignites from striking the matchbox first.

Weak continuity instructions. If you need boneless chicken, specify the same boneless cut in every shot and rule out bones, wings or drumsticks.

Unplanned text. Seedance can invent captions or signs. Say "no text", or give the exact on-screen text in quotes when you do want it, like a final "Good things take time. ♡".

A good 15-second structure

Final tip