Anime food is its own genre. The trembling egg yolk, the steam that catches the light, the sauce that looks lacquered — Ghibli and Shinkai spent decades making drawn food look more delicious than real food. It turns out Seedance 2.0 can do this too, but only if you prompt it the way anime food is actually directed.
We collected the 11 anime cooking prompts that produced the best results on scenic.sh. They cluster into five techniques: style-first ambience, per-second montage with a recurring chef character, storyboard-to-anime conversion, continuity-locked fast cuts, and vertical ASMR. Every prompt is linked and ready to copy.
1. Makoto Shinkai Strawberry Cake — Style-First Baking
See the full prompt on scenic.sh →
"Anime cinematic style inspired by Makoto Shinkai, warm sunlight, soft glow, high detail food textures, shallow depth of field, smooth transitions. 15 seconds, multi-angle (top-down + side + close-up), woman's face and hands always visible.
A young anime woman cooking calmly, expressive face, soft smile, natural movements.
Sequence: Eggs + sugar whisked fluffy → butter + vanilla added → flour sifted and folded → batter into pan → oven glow → sponge out and cooled → sliced into 3 layers → cream piped → strawberries added…"
Why this works: This is the most-saved anime food prompt in the gallery, and it's also the shortest. Naming a director — "inspired by Makoto Shinkai" — does more work than twenty lighting keywords, because the model already knows what Shinkai light looks like: warm backlight, lens glow, saturated golden hour. The whole recipe is compressed into a single arrow chain ("eggs + sugar whisked fluffy → butter + vanilla added → …"), which reads like a storyboard margin note and keeps all fifteen steps in order without eating your token budget on prose.
The other trick is the constraint "woman's face and hands always visible" plus the negative "no faceless shots, no only top-down." Food-only montages drift into stock-footage territory; keeping the cook's face reactions in frame is what makes it feel like a scene from a film instead of a recipe video.
2. Tonkotsu Ramen — Anime Culinary Realism with Smash Cuts
See the full prompt on scenic.sh →
"Style & Mood: Anime culinary realism — creamy whites and deep ambers, steam-filled kitchen atmosphere, warm candlelight-style practicals, dramatic backlight through rising vapor. Visual language from Japanese food animation: macro textures, translucent noodle strands, trembling yolk close-ups, milky broth depth.
Dynamic Description: Wide stabilized overhead — wooden prep board with fresh ramen noodles, rolled pork belly, halved soft-boiled egg… Hard smash cut to tight macro — hands rolling pork belly with twine… Smash cut to extreme close-up static — pork belly hitting a cast iron pan fat-side down, oil spattering outward in slow motion…"
Why this works: Every cut in this prompt is written as "[cut type] + [camera position] + [one action]": "Hard smash cut to wide low-angle — a large pot on roaring flame." That three-part grammar is why the edit feels intentional instead of random — the model treats each sentence as one shot and executes the transition it was given. "Smash cut" specifically produces the aggressive, rhythm-driven editing that anime cooking sequences (think Food Wars) are known for.
Steal the color direction too: "creamy whites and deep ambers" is the entire tonkotsu palette in five words. Defining a two-color palette up front keeps an eleven-shot sequence looking like one film rather than eleven stock clips.
3. Shoyu Ramen — Storyboard Reference, Full-Color Output
See the full prompt on scenic.sh →
"Use rough pencil sketch storyboards as visual references. Use the storyboards only for composition, camera angles, hand movements, food placement, and cooking continuity, and convert the final video into a full-color, high-quality Japanese 2D anime cooking video. Do not retain the appearance of black-and-white pencil sketches in the final video.
[0:00 - 0:02] The Teaser: Extreme macro close-up of a finished bowl of shoyu ramen. Thick, volumetric steam billows gracefully… [0:03 - 0:04] The Boil: Low-angle dynamic shot. A teppo (noodle basket) filled with raw yellow noodles is plunged into a roaring pot of boiling water…"
Why this works: This is the storyboard workflow — you attach rough pencil sketches as reference images, then explicitly scope what the reference is for: "composition, camera angles, hand movements, food placement, and cooking continuity" — and what it is not for: "Do not retain the appearance of black-and-white pencil sketches." Without that second sentence, Seedance happily renders your video in pencil-sketch monochrome. Scoping the reference is the difference between borrowing a storyboard's directing and inheriting its art style.
Note the tempo clause: "fast-paced editing, quick-tempo cuts. However, not fast cooking, not sped-up cooking, and not time-lapse cooking." Fast cuts and fast motion are different things, and if you don't separate them the model speeds up the hands into blurry time-lapse. This one sentence fixes the most common failure in AI cooking video.
4. Peperoncino — Ten Numbered Cuts with Strict Continuity
See the full prompt on scenic.sh →
"Japanese full-color anime, ultra-detailed cinematic food animation. Fast rhythmic peperoncino cooking montage in a bright modern Japanese home kitchen… Movie-quality anime food commercial style, Studio Ghibli-meets-modern anime energy. Warm indirect lighting, soft steam rim light, 35mm film grain, subtle anamorphic flare… No neon, no AI glow, no plastic textures.
Fast cuts:
- Macro knife slices garlic in one smooth motion, tiny fragments bounce, sharp chopping sound.
- Top-down still macro of sliced garlic on wooden board, tiny oil sheen, brief silence.
- Low pan close-up: olive oil pours into silver frying pan, glossy backlit reflections…
Maintain strict continuity: same hands, utensils, pasta shape, garlic/chili, kitchen layout, pan texture, oil color, lighting, and pasta strands across all cuts."
Why this works: Ten numbered cuts, each one shot, each with its own sound cue baked into the visual description ("aggressive sizzling," "brief silence"). Writing sound into a numbered cut list paces the montage the way a real editor would — cut 2's "brief silence" after cut 1's sharp chop is genuine editing rhythm, not just a shot list.
The closing continuity block is the part most people skip and shouldn't. Listing every element that must stay identical — "same hands, utensils, pasta shape, garlic/chili, kitchen layout" — is what stops the pan from changing color between cut 3 and cut 7. In multi-cut food prompts, ingredient morphing is the number one giveaway of AI; one explicit continuity sentence largely eliminates it. Also steal "No neon, no AI glow, no plastic textures" — the anti-AI-look negative that keeps anime food from going uncanny.
5. Teriyaki Cheese Donburi — Sizzle Verbs over Adjectives
See the full prompt on scenic.sh →
"The star of the video is a guilt-inducing teriyaki chicken donburi, topped with melted, stretchy cheese and a glossy, tempting egg yolk.
…Cut to the chicken being placed skin-side down into a hot frying pan. Convey a sharp, sizzling impact sound, tiny droplets of oil splattering, and the air warping from the heat. Cut to soy sauce, mirin, and sugar being poured into a small saucepan, swirling and blending together into a rich amber-colored sauce… Pour the rich teriyaki sauce into the frying pan. Show the sauce bubbling, boiling down, and coating the chicken in a lacquer-like, glossy shine."
Why this works: Look at the verbs: swirling, blending, bubbling, boiling down, coating, quivering. Anime food appeal lives in physics, not adjectives — "the air warping from the heat" gives the model a concrete visual phenomenon to render, where "delicious-looking chicken" gives it nothing. Every shot in this prompt describes something moving: droplets splattering, sauce reducing, cheese stretching, yolk trembling.
"Lacquer-like, glossy shine" is the signature phrase — that's the exact finish anime sauce has, and naming the material (lacquer) is more precise than stacking "shiny glossy beautiful." It also repeats the anti-timelapse clause from prompt 3, which by now you should treat as mandatory boilerplate for any cooking montage.
6. Aji Fried Cutlet — The 12-Scene Discipline
See the full prompt on scenic.sh →
"Generate a 15-second, 16:9 horizontal video, Japanese 2D animation cooking video, showing the process of making golden Aji fried fish cutlets. Make the cooking process itself the sole protagonist. Start from the first actual cooking step; do not show the finished dish at the beginning. All 12 scenes must be different cooking steps. Scenes 1 through 11 are limited to cooking steps, preparation, heating, tool preparation, or atmospheric deep-focus shots. Only Scene 12 shows the finished dish. No repeated plating shots, no repeated different angles of the finished dish.
Scene 1, 0.0–1.0 sec: On a stable cutting board, place an opened raw Aji fish, with a hand sprinkling salt. Fine grains fall onto the fish's surface…"
Why this works: This prompt solves the "hero shot spam" problem — left alone, video models pad cooking sequences with repeated beauty shots of the finished dish. The fix is structural: "Scenes 1 through 11 are limited to cooking steps… Only Scene 12 shows the finished dish. No repeated plating shots." You're not asking nicely; you're allocating scene slots, and the model respects the allocation.
Second-level detail worth copying: every scene gets a timestamp with a decimal ("Scene 1, 0.0–1.0 sec") and a stated composition ("a wide, slightly oblique 45-degree kitchen-counter composition"). Twelve scenes in fifteen seconds only stays readable because each scene is one action shot from one declared angle.
7. Tonkatsu — Match Cuts from a Storyboard
See the full prompt on scenic.sh →
"Please turn the storyboard sequence into a fast-paced cooking video following the order of the scenes.
Editing:
Use fast-paced, rhythm-driven cutting as the foundation, ensuring the cooking process is intuitively easy to understand. Apply match cuts (matching shape, motion, composition, and texture) to transition smoothly between scenes.
Scene 1 — Pounding the pork loin
Close-up of anime-style hands gripping a wooden meat mallet, firmly pounding a thick pink pork loin cutlet on a wooden cutting board. The meat flattens slightly with each strike, fibers visibly loosening…"
Why this works: This is the only prompt in the set that names an editing theory: "Apply match cuts (matching shape, motion, composition, and texture)." A match cut on shape — round cutlet to round plate, pouring egg wash to pouring sauce — is how professional food commercials glue scenes together, and Seedance actually attempts it when asked with the definition included in parentheses.
The prompt is also formatted in markdown (## Style, ## Editing, ### Scene 1) with bilingual scene titles, which works fine — Seedance doesn't need prose. Structure your prompt like a production document and it behaves like a crew reading one. Each scene ends with the same style stamp ("Hyperrealistic anime style, sharp kitchen lighting, cinematic shallow depth of field"), re-anchoring the look every scene so cut 5 can't drift away from cut 1.
8. Takoyaki Street Stall — Per-Second Steps with a Chef Character
See the full prompt on scenic.sh →
"Use @Image1 as the chef character reference. Keep the chef's face, outfit, hairstyle, proportions, and anime style consistent throughout.
Create a 10-second horizontal 16:9 anime cooking video showing takoyaki being made at a Japanese street food stall in a fast cinematic step-by-step montage. Each second shows a new cooking step. Style: premium Japanese anime food commercial, warm festival stall lighting, glowing lantern ambience…
0:00–0:01 Close-up of the hot takoyaki grill. The chef lightly brushes oil across the round molds. 0:01–0:02 The chef pours smooth batter into the takoyaki pan, filling each mold as the batter spreads and sizzles…"
Why this works: "Each second shows a new cooking step" is the cleanest montage recipe in this collection: ten seconds, ten steps, one line each. The rigid grid means you spend zero tokens on transitions — the model cuts on every second boundary automatically — and takoyaki is the perfect subject because the real dish genuinely has ten discrete steps (oil, batter, octopus, crumbs, first turn, second turn, browning, tray, sauce, bonito flakes).
The other half is @Image1 character referencing: attach a chef illustration once and lock it with "Keep the chef's face, outfit, hairstyle, proportions, and anime style consistent throughout." That's what makes the next two prompts possible — same chef, different dish, instant series.
9. Shrimp Tempura — Same Chef, New Dish
See the full prompt on scenic.sh →
"0:05–0:06 Close-up of hot deep-fry oil. The chef uses chopsticks to drip a little batter into the oil to test the heat. Small bubbles appear immediately. 0:06–0:07 The chef gently places the shrimp into the hot oil. Oil bubbles actively around them. … 0:08–0:09 One shrimp is cut in half, revealing a juicy inside and crunchy coating. 0:09–0:10 Final plating shot: shrimp tempura arranged beautifully on a plate with shredded cabbage, dipping sauce, and lemon. The chef squeezes lemon over the shrimp."
Why this works: Same author, same template as the takoyaki prompt — and that's exactly the lesson. Once the per-second grid and chef reference are locked, producing a new episode means swapping the ten steps. Notice which steps made the cut here: the batter-drip heat test at 0:05 and the cut-in-half reveal at 0:08. Those are the two moments a real tempura chef would call the money shots — the craft beat and the texture beat.
The cross-section reveal ("revealing a juicy inside and crunchy coating") deserves special attention: interior texture shots are the highest-appetite-appeal frames in food video, and models only produce them when you explicitly cut the food open in the prompt.
10. Fluffy Pancakes — The Breakfast Episode
See the full prompt on scenic.sh →
"Create a 10-second horizontal 16:9 anime cooking video showing pancakes being fried in a fast cinematic step-by-step montage. Each second shows a new cooking step. Style: premium Japanese anime food commercial, warm kitchen lighting, cozy breakfast mood, beautiful food close-ups, golden textures, butter shine, soft steam…
0:04–0:05 Small bubbles appear on the pancake surface as it cooks. Soft steam rises. 0:05–0:06 The chef slides a spatula under the pancake and flips it cleanly in one smooth motion. … 0:08–0:09 The chef adds a platform of butter on top and drizzles syrup slowly over the pancake stack."
Why this works: Third episode of the chef series, and the style line is where the variation happens: "warm festival stall lighting, glowing lantern ambience" (takoyaki) became "warm kitchen lighting, cozy breakfast mood." Two swapped phrases relocate the entire scene. Template prompting isn't lazy — it's how you get a coherent content series instead of eleven unrelated clips.
The dish-specific physics still get their own beats: surface bubbles before the flip (the visual cue every pancake video needs) and the slow syrup drizzle, which is the single most-watched motion in breakfast food content. One flip, "cleanly in one smooth motion" — remember that anime cooking hands fail on repeated motion far more than on single decisive actions.
11. Mint Mojito ASMR — Vertical Slice-of-Life
See the full prompt on scenic.sh →
"15-second vertical anime-style animation (9:16). No background music. Only natural sound effects and soft ASMR audio. A cozy summer kitchen with mint-green cabinets and soft natural sunlight. Flat anime illustration style, pastel colors, transparent sparkling glass, fresh mint leaves, lime slices, fizzy soda bubbles…
Scene 3: Close-up of lime and mint being crushed with a wooden muddler inside a glass. Sound: soft crushing sound, mint leaves moving, juicy citrus sound.
Scene 4: Ice cubes falling into the glass. Sound: sharp glass clinks and ice sounds."
Why this works: The only vertical prompt in the set, built for Reels/Shorts, and the only one where audio leads: "No background music. Only natural sound effects and soft ASMR audio." Every scene gets a paired Sound: line — knife on wood, leaf rustle, glass clinks, fizz. Seedance 2.0 generates audio, and separating the sound direction from the visual direction per scene is how you get ASMR that syncs instead of a generic soundtrack.
The palette is also a lesson: "mint-green cabinets," "pastel colors," "transparent sparkling glass" — the drink's colors become the room's colors. Anime slice-of-life does this constantly (the scene is lit in the mood of the dish), and it's a one-line trick: paint the kitchen in your ingredient's palette.
Anime food prompt cheat sheet
Across all 11, the principles that make anime food prompts work:
- Name the aesthetic, don't describe it — "inspired by Makoto Shinkai," "Studio Ghibli-meets-modern anime energy," "premium Japanese anime food commercial." One named style outperforms twenty lighting keywords.
- Fast cuts ≠ fast cooking — always add "not sped-up cooking, not time-lapse cooking." Cut quickly between actions, but keep each action at natural speed and physically readable.
- Physics over adjectives — trembling yolk, warping air, stretching cheese, sauce boiling down to a lacquer shine. Describe what moves, not how delicious it is.
- One step per time slot — per-second grids (0:00–0:01…) or numbered scenes with one action each. Budget the finished-dish shot explicitly ("only Scene 12 shows the finished dish") or the model spams hero shots.
- Lock continuity and character —
@Image1chef reference + "keep face, outfit, proportions consistent," and a closing "same hands, utensils, kitchen layout across all cuts" clause. Ingredient morphing is the #1 AI giveaway. - Direct the sound per scene — a
Sound:line under each scene gets you synced ASMR; "no BGM" keeps it clean.
FAQ
Which AI video model is best for anime food videos? Seedance 2.0 is currently the strongest at anime-style cooking content — it handles 2D animation aesthetics, per-scene audio direction, and reference images (storyboards or character sheets) in one prompt. All 11 prompts above were made with it.
How do I stop AI cooking videos from looking like time-lapse? Add an explicit tempo clause: "fast-paced editing, quick cuts — however, not fast cooking, not sped-up cooking, not time-lapse cooking." Cut speed and motion speed are separate instructions; if you only say "fast-paced," the model speeds up the hands.
How do I keep the same anime character across multiple cooking videos? Attach a character image and reference it: "Use @Image1 as the chef character reference. Keep the chef's face, outfit, hairstyle, proportions, and anime style consistent throughout." Reuse the same template with a new dish for each episode.
Can Seedance 2.0 use a storyboard as reference? Yes — attach pencil-sketch storyboards and scope them: use the storyboard "only for composition, camera angles, hand movements, and continuity," and add "do not retain the black-and-white pencil sketch look" so the output renders in full color.
How long should an anime cooking prompt video be? 10–15 seconds works best. Ten seconds fits a ten-step per-second montage; fifteen seconds fits about twelve distinct scenes. Longer sequences drift on continuity — for full recipes, generate multiple clips with the same character reference and style block.
What keywords make AI food look appetizing? Texture and motion words: glossy sauce, lacquer-like shine, rising steam, trembling yolk, oil bubbles, crispy cross-section, soft fluffy interior. Pair them with warm lighting ("warm kitchen lighting," "golden hour") and a shallow depth of field.
Browse the anime prompt gallery for more anime examples, or see food & cooking prompts for live-action food styles and animation style prompts for other animation looks. Read how to write Seedance 2 prompts for the complete prompting guide.
