Free, no signup
The Porsche hero, start to finish
The two-step prompt behind the hero on that 3D Porsche page. Nothing held back. Around 2 credits in Higgsfield.
Step 1: the still
Model nano_banana_pro, aspect 21:9, resolution 1k. The 1k is deliberate. This is a background, not a poster, and 4k costs four times as much for something nobody looks at closely.
Ultra-wide 21:9 studio photograph of a red Porsche 911 GT3 RS, three-quarter front view, dead centre of the frame on a wet reflective black floor. Deep charcoal seamless background. Hard rim light down both flanks, soft key from above, crimson bounce on the floor. Crisp automotive commercial photography, shallow depth of field, high detail. IMPORTANT: absolutely no text, no words, no lettering, no logos with letters, no badges with writing, no typography anywhere in the image. Leave generous EMPTY dark background on the far left and far right thirds of the frame, so the car sits compactly in the middle with plenty of clean side room.
Step 2: the move
Feed that still into image-to-video, still 21:9.
Slow cinematic push in toward the car, drifting very slightly to the left as it moves. Camera moves forward only, no orbit, no shake, no cuts. Light sweeps along the bodywork as the camera travels. Faint haze in the air. End on a clean, steady, text-free frame.
The two lines that actually matter
Everything else in that prompt is ordinary. These two are the difference between a hero that works and one that falls apart the moment it hits a real browser.
“absolutely no text, no words, no lettering, no typography anywhere”
Without this the model bakes fake, misspelled words into the pixels, usually across the car. Once text is part of the image it can never reflow, so it gets sliced on every screen size except the one you previewed on.
“leave generous EMPTY background on the far left and far right thirds”
Without this the car fills the frame, so a phone crops the sides and cuts the nose and tail off. With the empty room, cover only ever eats empty background, and the car is never cropped on any screen.
Then put every word in HTML
The image is only the background. The big PORSCHE wordmark, the nav, the spec row and the buttons are all real HTML text sized with clamp() sitting on top. That is the whole trick. The picture never crops the car, and the copy reflows instead of clipping.
It is also why the wordmark can sit behind the car. Two layers, not one flat image.
Skip that and you get the three bad options everyone ends up choosing between: cover crops it, contain gives you black bars, and blurred side-bars look cheap.
Want the rest of it
The prompt gets you the hero. The page is the other 80 percent: the layout, the scroll behaviour, the motion, the responsive rules.
Every template in the gallery comes with the full recreation prompt. The stack, the CSS blocks, and a section by section breakdown. Paste one into Claude, Cursor, v0 or Bolt and you get the page.
Would rather have it built for you? That is Stroupe Digital. Reply to the DM and tell me what you are building.