Introducing Recipes: Fill in the Blanks, Get the Video

We just shipped Recipes in the Video Studio. 17 fill-in-the-blanks forms that turn a couple of photos and a sentence or two into a finished video, without you writing a prompt at any point.

Here's why I care about this one. Every video model we shipped this year got better, and every time the same thing happened: people who knew how to prompt got amazing results out of it, and everyone else typed "a cool ad for my product" and got mush. At some point I stopped believing the models were the problem. The prompt is the problem. So we went after the prompt.

What a recipe is

You pick a card, you fill in a short form, you hit Generate. That's it.

Want a UGC-style product ad? The recipe asks for a photo of a person, a photo of your product, a script, and a setting. Want your logo as a 3D animated sting? Upload the logo and pick a motion style from a dropdown. Drone flyover? Type a location, pick a time of day. Nothing in any of these forms requires skill to fill in. It's a photo you have, a name, a sentence.

Under each form there's a prompt scaffold we built and tested for that one format. The camera language, the lighting, the audio direction, the aspect ratio and length, the reference-image handling that keeps your product's logo readable and a person's face the same person, and the "avoid" instructions for whatever that format tends to get wrong. Your words get slotted into that scaffold and the whole thing runs as one generation.

We already have Templates, and this is not that. A template gives you a good starter prompt and you edit it, so you're still prompting, just with a head start. A recipe never shows you a prompt at all.

17 recipes at launch

The gallery groups them into four sections:

• Ads: UGC Product Ad, Cinematic Product Spot, Giant Product Scene, Luxury Jewelry Spot, Fashion Try-On, Ingredient Burst. Photo in, commercial out. Person plus product becomes a vertical talking ad, a garment photo becomes a try-on video, your bottle gets the slow-orbit hero treatment.
• Brand: Logo Animation. Soft float, liquid metal, glass shatter, or neon glow.
• Cinematic: One-Shot Scene, Drone Flyover, Freeze-Time Effect, Action Fight Scene, Music Video. You describe the scene, the recipe brings the film school.
• Social: Award Ceremony Winner, Viral Talking Character, Animal Vlogger, Action Figure Reveal, 3D Cartoon Dance. The stuff you've watched go viral, except it's your face and your joke in it.

A few favorites so you get a sense of the range. Award Ceremony Winner takes two photos, the winner and the host, and gives you a broadcast-style ceremony: envelope, your name called out loud, spotlight sweep, trophy handoff, name up on the LED wall. We've been making these for each other in the office all week, it does not get old. Freeze-Time Effect takes one photo and has that person snap their fingers, stop the world mid-confetti, and stroll through the frozen crowd. Animal Vlogger doesn't even need a photo. Describe an animal, tell it where it's vlogging from, give it a monologue, and you get a selfie-stick walking tour with lip-synced commentary.

The prompting knowledge is baked in

If you've put real hours into video models you know the tricks by now. Dialogue needs an explicit audio directive or you get random music over silence. Multi-person scenes need identity locks or faces drift between shots. Reference images need role grammar so the model knows which photo is the person and which is the product. Action only reads as tension when the impacts land on close-ups. And "single continuous take" is quietly doing enormous work in almost every good prompt you've ever copied from someone.

None of that should be required knowledge for making a birthday video where your friend wins an Oscar. Each recipe's scaffold carries those rules for its format. And because a recipe targets exactly one deliverable, the scaffold gets to be opinionated in a way a general-purpose prompt enhancer can't be.

Two honest notes on the internals. First, recipes skip our prompt enhancement step on purpose. The scaffold already is the engineered prompt, and a rewrite pass would just water it down. Second, each recipe pins the model, format, and clip length its scaffold was tested against, and the form shows you which model and duration you're getting before you spend anything. Yes, that means fewer knobs. That's kind of the point. If you want the knobs, Text to Video and Image to Video are right there with every model and every setting.

Your words stay yours

Small detail I like: when the video lands in your library, the caption is what you typed, something like "UGC Product Ad - your script · your setting". Not the several hundred words of scaffold we wrapped around it. The engineering stays behind the curtain where it belongs.

Every recipe also has its own URL, so you can send a teammate straight into the Fashion Try-On form and skip the explanation.

The usual reminder about privacy

Recipes run through the same anonymity layer as everything else here. Requests are encrypted on your device, identity gets split from content by Oblivious HTTP relays, and decryption only happens inside attested secure enclaves. No single party holds both who you are and what you generated. The full architecture is here.

Worth saying out loud for this feature in particular, because recipes are built around personal photos. Your face, your kids' birthday, a product you haven't announced. That's exactly the material nobody should have to hand over under their own name just to get a good result.

Try one

Live now, free to start, no card. Costs show in credits before you generate, same as everywhere in the studio (1,000 credits = $1).

Browse the recipe gallery→

Make a UGC Product Ad→

Put someone on the winner's stage→

We've watched too many people bounce off a blank prompt box that the pros were pulling magic out of. This is our attempt at fixing that. And if the recipe you keep building by hand isn't in the gallery yet, tell us. It's built to grow.