Skip to content
HN On Hacker News ↗

Realistic Shot — a Claude skill for ProductAI

▲ 8 points by withoutshape 2d ago HN discussion ↗

Pangram verdict · v3.3

We believe that this document is fully AI-generated

99 %

AI likelihood · overall

AI
0% human-written 100% AI-generated
SEGMENTS · HUMAN 0 of 2
SEGMENTS · AI 2 of 2
WORD COUNT 477
PEAK AI % 99% · §1
Analyzed
Jul 6
backend: pangram/v3.3
Segments scanned
2 windows
avg 239 words each
Distribution
0 / 100%
human / AI fraction
Verdict
AI
Pangram v3.3

Article text · 477 words · 2 segments analyzed

Human AI-generated
§1 AI · 99%

realistic-shot.skill — for Claude × ProductAI How it works Install

A Claude skill · powered by the ProductAI MCP Kill the plasticAI portrait generations. Every image model ships the same look: glass skin, golden glow, painted blush, studio-perfect speculars. This skill turns any photo or render into an unretouched, straight-out-of-camera file — pores, sensor grain, real light — while your product's label stays pixel-faithful.

AI RENDER REALISTIC SHOT nanobananapro · image-to-image WB imperfect · ISO grain on retouching 0%

Pipeline How it works The skill runs a style called Raw Camera Casting Realism: realism produced by physical evidence, not by the word "photorealistic". 01 / input Drop in any image A render, a model shot, a product photo, even a layered PSD — the skill flattens, classifies it, and reads what needs to change. 02 / prompt Evidence-first prompt One paragraph engineered per image: which artifacts to kill (blush, glow, gloss) and which physical evidence to build in. Smooth renders get transformation-first structure so texture wins over similarity. 03 / generate ProductAI MCP Runs image-to-image on nanobananapro through your ProductAI connector. No other generator, no manual uploads — Claude drives the whole call. 04 / verify Zoom check Results get inspected at 100% — skin texture must survive the crop, and product labels are verified character-for-character against the reference.

What gets engineered in The realism stack visible pores curving with face geometry fine vellus hair catching side light irregular natural flushing, not painted blush under-eye texture, faint creases, discoloration vertical lip lines, dry patches, muted color stray brow hairs breaking the perfect arch

§2 AI · 99%

flyaways, frizz, baby hairs at the hairline sensor grain at moderate ISO, shadow noise slightly imperfect white balance, room cast wool fibers, pilling, loose threads on knits fingerprints, dust, micro-scratches on products soft overcast window light, real shadows

Product fidelity law Imperfections go on the scene. Never on the label. When a product is in frame, its packaging shape, label layout, typography, logo placement, and brand colors are locked to the reference — crisp, undistorted, fully legible, inside the sharp focal plane. Wear and realism land around it: a fingerprint where a hand grips, dust on upward faces, real contact shadows. After generation, the label is verified against the original before anything ships. // prompt suffix: "no changes to the product design whatsoever."

Install Running in two minutes Download the .skill file and open it in Claude — hit Save skill. Connect the ProductAI MCP in Claude's connector settings. Upload any image and say "make this a realistic shot". The skill handles conversion, prompting, generation, and the zoom check. Iterate by clause — "more grain", "less frizz", "cooler light" — the skill edits only that part of the prompt and re-runs. Requirements · Claude (claude.ai, desktop, or Code) · ProductAI account + MCP connector — productai.photo · Generation uses ProductAI credits (1 per variant)