Menu refresh, 30 dishes: a dish card and a 5-second clip each on Sume

A 30-dish seasonal menu with one Ideogram 4.5 dish card and one 5-second Wan 3.0 clip per dish costs $11.85 at 480p and $21.15 at 720p on Sume, with a medium-quality card at $0.075 and a clip at $0.32 or $0.63. One music bed for the whole set adds $0.125. Four seasonal menus a year is $47.40 at 480p, or $84.60 at 720p, before retakes.

The menu is a good fit for a small pipeline because every dish is a repeatable unit: a name, a description, a photo or a prompt. The card is a still with the dish name quoted in it, and the clip animates the card. Wan 3.0 supports image-to-video, so the card goes in as the first frame. Prices are list times 1.25 (read 2026-10-06); the Ideogram list is $0.03, $0.06 and $0.22 for low, medium and high.

Where the money goes

The clip is 81 percent of the 480p dish and 89 percent of the 720p dish with a medium card. Upgrading the card from medium to high costs $0.20 a dish ($6.00 for the menu) and buys sharper type, which matters when the dish name is in the image. Upgrading the clip from 480p to 720p costs $0.31 a dish ($9.30 for the menu), and it matters when the clip plays full screen on a table display.

Two rules keep the bill down. Make the clip from the approved card, so you do not pay for a clip of a card that you would throw away. And render the cheap tier first for the whole menu, then re-run only the dishes that read badly.

• Quote the dish name and price in the card prompt only if you will check the spelling; otherwise add the text later in a caption or overlay.
• A 5-second clip at 480p is $0.32 and at 720p $0.63. Both round each job up to the cent.
• One music bed is $0.125 for a Music 1.0 generation, and it does not change with prompt length.
• Re-use the same bed for every dish clip in a menu, so the set sounds like one menu.

Price a menu from a CSV of dishes

The script reads the dish count and the chosen tiers, adds the single music bed, and prints the menu total. Dishes that you re-run are added as a retake percentage.

CARD = {"low": 0.0375, "medium": 0.075, "high": 0.275}
CLIP_5S = {"480p": 0.32, "720p": 0.63}
MUSIC = 0.125


def menu_cost(dishes: int, card: str, res: str, retakes: float = 0.0) -> float:
    per_dish = CARD[card] + CLIP_5S[res] * (1 + retakes)
    return round(dishes * per_dish + MUSIC, 2)


print(menu_cost(30, "medium", "480p"))
print(menu_cost(30, "medium", "720p"))
print(menu_cost(30, "high", "720p", retakes=0.2))

A retake allowance and a seasonal rhythm

Budget a retake allowance from the first menu, not from a guess. If one dish in five needs a second clip at 720p, the allowance is 6 clips at $0.63, $3.78, and the menu costs $25.06 with the music bed. At 480p the same allowance is 6 clips at $0.32, $1.92, and the menu costs $13.90.

The rhythm of a restaurant helps. A menu changes four times a year, so the same prompts, with new dish names, run again each season. Keep the card prompt and the clip prompt in the repo, and a new menu is a one-line change per dish. The cost per menu then stays flat from one season to the next, which makes it easy to put in an annual plan.

A caveat on the card

Even a model built to draw text can misspell it, and a menu card with a misspelled dish is a bad card at any price. Check every card by eye before it becomes a clip, and keep the check in the workflow rather than in someone's head. The images docs list the quality tiers and sizes, and the Video Router catalog lists the live Wan 3.0 limits.

The music bed is a separate job in the Music docs. The image tiers are in the images docs.