Log in
TechniqueBy HypernaturalVersion 4, updated

Restyle a photo as a location plate

hn/nano-banana-restyle-photo-as-plate

Turn a photo of a real place into a location plate in the job's art style with nano banana 2.1 edit, keeping the photo's camera, perspective and architecture exactly, its ads and logos cleared and its real signs spelled out. Use when stills must show a real stadium, building, street or landmark from several angles: one plate per camera direction, then stage people into the plates instead of generating the place.

SKILL.md

Give it a photo of a real place and the job's style frame; get the same view, same camera and same architecture, drawn in the job's style, with the photo's ads and logos cleared. The plate is the place's source of truth for every still that looks that way: people are staged into it, the place is never drawn again. Use it when the place's layout matters and it is seen from more than one angle (a stadium, an arena, a station hall, a street). For an invented place, design the set instead (guide-design-a-set).

Why plates: generating a real stadium from text got a different, more famous one; with photos given as plain references, the roof still came back wrong on 13 of 22 scenes. Restyled photos kept the real geometry on all 6 plates, and 16 scenes staged into them came back with one note.

Inputs

  • One photo per camera direction the shots need: looking in, looking out, from high up, at field or street level, the outside, the entrance with its sign. Ask the person for their own first (their photos from a visit gave 3 of the 6 plates, the views the shots needed most); otherwise free-licensed ones, with their source and licence noted in refs/<place>/sources.md.
  • The job's picked style frame.
  • The place's facts (guide-depict-real-people-and-places): its distinctive structure in words, and the exact text of every sign that should stay.

Run it

Call describe("fal/google/nano-banana-2.1/edit") before the first run in a job. The photo is image 1, the style frame image 2:

run("fal/google/nano-banana-2.1/edit", {
  "prompt": "<the prompt below>",
  "image_urls": ["refs/<place>/inside_behind_home.jpg", "refs/style_frame.png"],
  "aspect_ratio": "16:9", "resolution": "2K"
}, into="refs/plates", unit="plate_in_home")

One take per photo was enough (6 of 6 approved). Name the unit by the direction it looks (plate_out_field, plate_in_upper), and record a table with the job's notes of which plate serves which direction.

Prompt

Redraw image 1 as an illustration in exactly the art style of image 2: <the style in words>.
Keep the composition, camera position, perspective and all architecture of image 1 EXACTLY: <the distinctive structure, named piece by piece: the roof sections and where they meet, the trusses and light rows, the facade, the glass walls, the stands, the field>, and <the people> in the same places.
Remove <clutter in front of the camera: protective netting>.
Replace every advertisement, sponsor logo and piece of text with plain panels in the palette, except: <the real signs, each reading exactly "…">. <Screens> show a simple abstract graphic in the palette. Players and fans are generic, no names or numbers.
Extend the frame to 16:9 without changing the perspective.
  • Restyle in the photo's own light. Change the time of day when staging: "Make it night" in the staging prompt worked on every plate.
  • Name each structure the place is known for; "all architecture" alone is not enough to hold an unusual roof.
  • A photo that isn't 16:9: "Extend the frame to 16:9 without changing the perspective" held; "Crop to 16:9, keeping " when the photo has more than you need.

What you get

A 2K plate per call (2752x1536 at 16:9) in refs/plates/.

Tips and traps

  • Real details the photo carries (retired numbers, a sign) stay only when the prompt says so ("the retired-number discs keep their numbers"); otherwise they get cleared with the ads.
  • Staging into a plate: the plate is image 1 ("the exact view of the "), the picked still with the people is image 2 ("shows the people to add"), then "Keep everything EXACTLY as image 1 draws it" and the structures listed, and the thing the model tends to add named as absent ("do not add any roof structure over the outfield"). A still where the person matters most (a broadcaster at a window): edit that still, and replace the view through the window with the plate as image 2. Worked prompts are in nano-banana-prompt-format (references/examples.md).
  • Edits drift smoother than the style frame's texture. A grit pass fixed it: redraw the staged still "with exactly the same composition, perspective, architecture and people, but in the coarser finish of image 2", the style frame as image 2.
  • A clip's start and end frames: derive the end frame from the start frame by edit, so the place matches; two frames staged separately came back with differently shaped roofs.

Check it

Put the plates to the person with review_approve, each plate an item with its photo in its files. Check the structure the place is known for against the photo and the facts, signs spelled right, no ads or logos left, no skyline or towers the photo doesn't have. The person knows the place: their eye is the check that matters.

FAQs

Turn a photo of a real place into a location plate in the job's art style with nano banana 2.1 edit, keeping the photo's camera, perspective and architecture exactly, its ads and logos cleared and its real signs spelled out. Use when stills must show a real stadium, building, street or landmark from several angles: one plate per camera direction, then stage people into the plates instead of generating the place.

Install Hypernatural Desktop, a free Mac app, and connect it to Claude Code, Codex or Claude Desktop. Then describe the video you want and your agent finds the skill itself, or paste this page's link (https://hypernatural.ai/skill/hn/nano-banana-restyle-photo-as-plate) and it installs hn/nano-banana-restyle-photo-as-plate and follows it.

It runs Nano Banana 2.1. Hypernatural Desktop has every one of them built in, so there are no API keys to set up, though you can bring your own.

The skill is free. You only pay for the AI generations it runs, and your agent shows you the price of every batch before it spends anything. Editing, captions and titles run on your Mac at no cost.

Yes. Ask your agent to adjust it for your job, or to save its own version. You can publish your version under your own handle so anyone can install it.

Go make something cool.

Install the app, open your agent, describe the video. That's the whole onboarding.

Sticker HypernaturalSticker BannerSticker TVBoy