Log in
TechniqueBy HypernaturalVersion 4, updated

Re-render small print in an enlarged crop

hn/nano-banana-rerender-small-print

Get small print or a logo on a prop exactly right in a 1080p still: crop the prop, enlarge it about 6x so it fills the image, re-render only its print with nano banana Pro edit against the product photo, and paste back only the pixels that changed, so grain and faces stay. With a match reference cut at the same scale from the shot's other pinned frame, the print keeps its size and place between first and last frame. Use for garbled lettering, wrong logos, labels and packaging text.

SKILL.md

You give it a still, a square box around the prop, the product photo and a sentence on what the print must look like; you get four patched stills where only the print's pixels changed. A whole-frame edit at 1080p garbles lettering, and a second whole-frame pass (or a 4K render scaled down) smooths the grain and redraws background faces. Here the prop fills the model's image, so the lettering comes out right, and nothing else in the frame is touched. This leaf runs on python-crop-and-paste-back; load it for the script.

Inputs

  • The still, conformed to 1920×1080 (the script does it).
  • A square box around the prop at 1080p (find it on a gridded crop), e.g. 790,320,990,520.
  • The product photo, for the design. Turn it to face the way the prop faces.
  • design: what the print must look like, against the product photo: "near the top the white logo reads ACME over CO., in the same letter weights and spacing as image 2; centred on the side facing the camera, following its tilt".
  • For a shot with pinned first and last frames: the other pinned frame and the prop's centre in it, for the match reference.

Run it

  1. Crop and enlarge 6×: exec python scripts/crop_paste.py crop --image <still> --box 790,320,990,520 --scale 6 --out <job>/refs/shot09_last_crop.png.

  2. Optional match reference, on a pinned pair: exec python scripts/crop_paste.py match-cut --image <still> --box 790,320,990,520 --other <the other pinned frame> --center 885,410 --out <job>/refs/shot09_last_match.png. It cuts the prop at the same scale as your crop (the box scaled by the zoom between the frames, measured).

  3. Re-render, four takes, one call each, seeds 9100 to 9103 (call describe first in a job):

    run(model="fal/fal-ai/nano-banana-pro/edit",
        input={"prompt": <the region prompt>,
               "image_urls": ["<job>/refs/shot09_last_crop.png", "<job>/references/product.jpg",
                              "<job>/refs/shot09_last_match.png"],
               "aspect_ratio": "1:1", "resolution": "2K", "num_images": 1, "seed": 9100,
               "output_format": "png", "safety_tolerance": "5"},
        into="<job>/refs", unit="regionedit")
    
  4. Paste back each take's changed pixels: exec python scripts/crop_paste.py paste --image <still> --box 790,320,990,520 --edit <job>/refs/regionedit_a.png --out <job>/stills/shot09_last_q.png.

Prompt

The grammar is in nano-banana-prompt-format: load it before you change this prompt.

Image 1 is an enlarged close crop of a frame from a live-action commercial.
Image 2 is a product photo of {prop}, for its design only.
[Image 3 is the same prop in the same shot a moment earlier, cropped at the same scale as image 1, so the print must come out the same size in pixels. The print must sit in exactly the same place on the prop as in image 3: the same distance from {print_landmarks}, the same size, width and letter proportions. Only the prop's angle may differ; the print must not slide around it.]
Re-render ONLY {prop} in image 1 so its printed design matches image 2 exactly: {design}

Keep its position, angle, size, silhouette, color and lighting exactly as in image 1, including any backlight and rim light. Keep hands, fingers, faces, hair and everything else in image 1 pixel-identical. Match image 1's softness and fine film grain on the print: never studio-sharp. Ignore the studio lighting of image 2. Same framing and crop as image 1.

print_landmarks: the edges the print's placement is measured from, "the prop's left edge and from its lid" (default "the prop's left edge and from its top"). Include the bracketed line only with a match reference.

Pinned first and last frames

Fix the print on the last frame after it is cropped to the source's push-in (python-match-push-in), with the match reference cut from the locked first frame. A match reference at a different scale makes the logo come out a different width, and it slides across the prop mid-clip. If the two logos still differ, run it the other way: re-render the first frame's print to match the better last one. Either direction works.

Tips and traps

  • Keep the box square: the edit is asked for at 1:1.
  • Never warp the product photo onto the prop yourself; let the model render it in the crop.
  • Condition the references before they go in (colour, size, orientation): guide-keep-small-details.

Check it

paste prints changed_px: it should cover the print, not the whole box. Put the patched takes to the person with review_choose, the original still as original; zoom on the print, then judge it at playback size. On a pinned pair, put first and last in the same option: the print keeps its width and place on the prop.

FAQs

Get small print or a logo on a prop exactly right in a 1080p still: crop the prop, enlarge it about 6x so it fills the image, re-render only its print with nano banana Pro edit against the product photo, and paste back only the pixels that changed, so grain and faces stay. With a match reference cut at the same scale from the shot's other pinned frame, the print keeps its size and place between first and last frame. Use for garbled lettering, wrong logos, labels and packaging text.

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-rerender-small-print) and it installs hn/nano-banana-rerender-small-print and follows it.

It runs Nano Banana Pro. 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