nanorealism

Original：🇺🇸 English

Translated

Create realistic AI images with Nano Banana Pro (and similar high-adherence models). Use when generating photoreal people/scenes and you want the camera-first, block-structured prompt method (camera → lighting → subject → outfit → pose → location) plus photographic parameters (lens, f-stop, ISO) and a post-process imperfection pass (grain/film burn). Pairs with the `Realism` skill (skin/eyes/hair imperfection stacks) — use both together for max realism. Source: Curious Refuge "How to Create Realistic AI Images".

16installs

Sourcemariusmariusmariusmariusmarius/nanorealism

Added on2026-08-30

NPX Install

npx skill4agent add mariusmariusmariusmariusmarius/nanorealism nanorealism

SKILL.md Content

View Translation Comparison →

Nano Realism — realistic AI images (Nano Banana Pro method)

Camera-first, block-structured prompting for photoreal images. Built from the Curious Refuge guide. Use alongside the
Realism
skill — that one supplies the skin/eyes/hair imperfection language; this one supplies the camera/lighting structure, photographic settings, and the post pass.

Tools

Nano Banana Pro (Google) — best prompt adherence; access via Gemini or Freepik. Beats Midjourney/Flux for following detailed prompts.
Prefer an aggregator / batch workflow — you almost never nail it first try; generate several, pick the best.
Optional: Leonardo Blueprints (templates with auto film grain / physics).
Reference images: upload a real photo to Gemini and ask it to write the prompt ("blueprint") — then refine.

The block prompt structure

Write the prompt in clear hierarchical sections, in this order:

CAMERA — device + settings: e.g.
```
shot on iPhone 16/17 Pro, f/2.0, RAW
```
(or
```
disposable camera
```
,
```
35mm film
```
). Add lens (
```
35mm
```
,
```
50mm
```
), f-stop (depth of field), ISO (grain).
LIGHTING — source + quality + direction: e.g.
```
natural window light from the left, soft shadows, realistic highlights
```
. One main source reads more real than multi-source studio.
SUBJECT — physical description incl. age + ethnicity (and, from the
```
Realism
```
skill, the skin/eyes/hair imperfection stack).
OUTFIT — specific clothing items.
POSE — exact body positioning / a caught moment.
LOCATION — detailed, specific background with real props.

Be technically specific (not generic)

Generic = AI look. Add real photographic parameters:

Lens: 35mm / 50mm
Aperture / f-stop: e.g. f/2.0 for shallow DOF, f/8 for everything sharp
ISO: higher ISO → more natural grain
Light direction & type: natural vs strobe, side vs overhead

Post-generation refinement (kills the "too perfect" look)

Add imperfections in post: film grain, subtle film burn, slight vintage grade, mild chromatic aberration / vignette.
Fix elements with masking: Photoshop mask + targeted re-prompt to repair a specific wrong object/hand/text.
Iterate: batch-generate, cherry-pick, then post.

Quick ffmpeg grain/imperfection pass (single image)

bash

ffmpeg -y -i in.jpg -vf "noise=alls=8:allf=t+u,curves=preset=lighter,vignette=PI/5" -q:v 3 out_grain.jpg

(adjust

alls

4–12 for grain strength; drop

curves

if too bright.)

Skeleton prompt (combine with Realism skill)

CAMERA: shot on iPhone 17 Pro, 35mm equivalent, f/2.0, natural lens distortion, slight grain.
LIGHTING: soft natural window light from the left, warm highlights, soft real shadows — single source.
SUBJECT: [age + ethnicity] + [Realism-skill skin/eyes/hair imperfection stack].
OUTFIT: [specific clothing].
POSE: [caught moment, asymmetric, natural].
LOCATION: [specific room/place with 2+ lived-in imperfection anchors].
Do not smooth skin. Do not blur background into bokeh. No beauty filter. Preserve identity.

Relationship to other skills

Realism
= the imperfection/anti-AI-look language (skin zones, pigmentation-not-"freckles", banned words, eye/hair/expression stacks). The detail engine.
nanorealism
(this) = the camera-first block structure + photographic settings + Nano Banana Pro tooling + post grain pass. The framing/structure engine.
Best results: camera/lighting/structure from here + subject imperfection stack from
Realism
.

Source: Curious Refuge — "How to Create Realistic AI Images."