All skills
hoodini avatar

/image-master

@83e2083

Master prompt-engineer for photoreal, artifact-free AI still images on ANY tool (Reve, Midjourney, Flux, GPT-image, Imagen, Nano Banana, Stable Diffusion). Builds prompts that hit National-Geographic-grade realism — true skin/fur texture (no plastic), correct anatomy/hands/faces, physically coherent light/shadows/reflections, clean legible text — while engineering deliberate visual IMPACT (color contrast, composition, awe/adrenaline, a striking point of view). Runs a gated process: lock the point-of-view and build the 8-block Capture Stack BEFORE generating, then run a forensic pre-submit inspection against the known artifact list. Use whenever the goal is a single still image that must look REAL and hold up under close inspection — contest entries, hero shots, product/character/wildlife/architecture/concept art, or any time AI images come out plastic, distorted, or fake. Composes with nano-banana-2 (execution) and director (motion). Hebrew triggers: תמונה, תמונות, פוטוריאליזם, ריאליזם, לייצר תמונה, פרומפט לתמונה, בלי עיוותים, עור פלסטיק, ידיים מעוותות, פרצוף מעוות, חדות, נשיונל ג'יאוגרפיק.

Use this Skill: https://skilld.dev/gh/hoodini/ai-agents-skills/image-master

This session only. Nothing lands on disk.

references09-magazine-cover-treatment.md

≈2k tokens on demand. Your agent reads this file only when SKILL.md points to it.

09 — Magazine / Editorial Cover Treatment

When a series should read as an editorial body of work (masthead + short headline captions), add a TYPE LAYER on top of the photography. This showcases legible-text craft (a judge-invited discipline — "graphic design") but courts the double-penalized "typography artifact." So the workflow decides everything.

The cardinal rule: composite the type in post

Generate the clean photographic image first (reserve a text-safe negative-space zone), then set real vector type in a design tool — for Yuval, his instagram-cover-studio (~/instagram-cover-studio, Anton-based) or Figma/Photoshop. Real text = pixel-perfect kerning, zero garbled glyphs, full art direction, and it sidesteps the contest's most-penalized artifact entirely. This is exactly what Yuval's own BRANDING_SYSTEM.md prescribes: "Generate one strong vertical image → use the app to place typography → export." Only let the model render text to deliberately demonstrate its text engine, and then only a SHORT ALL-CAPS headline per 03, inspected at 100%.

Don't infringe (the contest requires original, non-infringing work)

Do NOT mimic a real magazine's trademarked identity — National Geographic's yellow border, TIME's red frame, any real masthead/logo. Borrow the language of covers (masthead, kicker, headline, dateline); invent an ORIGINAL wordmark.

Caption craft — extreme brevity that explains the frame

  • Kicker (small overline): one short line that gives the emotional read — 4–7 words.
  • Headline (the big type): 1–4 words, ALL CAPS, condensed — names the image.
  • Optional dateline/footer: a small consistent line for editorial realism (original, e.g. "ISRAEL — 2026"). Headline names the frame; kicker makes you feel it. Both must read at thumbnail/grid size.

The design-system lock (cohesion across the set)

  • Type: Anton (masthead + headline, condensed all-caps), Inter/Assistant for the small dateline. Same family every cover.
  • Palette / text colour: never default to flat all-white (reads generic). Specify the colours: masthead in clean white/bone, the recurring tagline (or a key word) in alarm RED — it pops like a real magazine AND ties to the "Red Alert" (Tzeva Adom) siren. The single red accent is the grid's owned signature. Do NOT use a high-dopamine neon (the YUV.AI Reels look — tonally wrong here).
  • Placement grid (locked, EVERY page): top — a masthead lockup: the title THROUGH THEIR EYES with the recurring tagline LIFE IN ISRAEL AT WAR beneath it; lower third — the per-image headline on a subtle dark gradient scrim; bottom-right corner — a small YUVAL AVIDANI credit. Fixed every page; only the photo and the headline change.
  • The recurring tagline IS the concept-explainer. Without it a viewer sees a dramatic animal but may miss that this is Israel-at-war seen through the animal's eyes. Keep it SHORT so it never overflows; masthead + tagline + the animal in frame together make the concept legible in one glance. The brief is minimal-text: short, sharp, piercing — never wordy.
  • Signature / anti-theft: the YUVAL AVIDANI credit goes on every page (provenance + theft-deterrent). Keep it ALL-CAPS and short for reliable in-model rendering; it is the single most likely string to need a re-roll.
  • Aspect — treat the set as ONE MAGAZINE FEATURE: each image is a PAGE in one photo-essay, so a consistent 4:5 portrait is correct and premium (cohesive grid, the lockup reads cleanly). Cinematic power comes from light, grade and composition — not from a wide ratio. Option: render the 1–2 most epic establishing frames as 16:9 "opening spreads" if you want a wide cinematic beat, but keep the spine 4:5 for cohesion.
  • Photo is sacred: keep all type in the reserved negative space (clear top band, lower-third scrim, corner) — never occlude the animal. The photograph carries the craft score.

Subject breaks the text plane — the 3D depth cover (premium trick)

The Vanity-Fair / movie-poster move: the subject OVERLAPS the title so part of it is IN FRONT of the letters and part BEHIND — the text stops being a flat sticker and becomes a physical middle layer. Occlusion is the brain's #1 depth cue → instant premium 3D. AI renders text on ONE flat layer by default, so you must STATE THE OCCLUSION EXPLICITLY in the prompt:

The large bold title "THROUGH THEIR EYES" sits across the upper-middle of the frame. The animal's head and ears overlap and partly cover the bottom of the letters (in front of the text), while the rest of the title passes BEHIND the animal's body — the title sits on a middle depth plane between the subject (foreground) and the background. A faint shadow of the animal falls across the letters. Rules: keep the overlap PARTIAL and believable (head / ear / shoulder / a raised paw crossing the letters, not the whole word buried); ONE title, ONE crossover point; match the type's lighting and a faint cast shadow to the scene. Reve (layout-first), Nano Banana and Flux do this; Midjourney can't. If the model flattens it, regenerate with the occlusion stated as its own separate sentence.

Worked captions — the LOCKED text system (the animal-war feature)

Every page carries the SAME lockup, only the photo + headline change:

  • Masthead title (top): THROUGH THEIR EYES
  • Recurring tagline (beneath, smaller — the concept-explainer): LIFE IN ISRAEL AT WAR
  • Headline (lower third, large): the per-image word(s) below
  • Credit (bottom-right, small): YUVAL AVIDANI | # | Image | KICKER | HEADLINE |

|---|---|---|---| | 1 | tigers watching the interception | THE SKY DECIDES WHO LIVES | THE WITNESSES | | 2 | mother over just-asleep cubs | WAKE THEM, OR LET THEM SLEEP | THE WAKING | | 3 | old lion left behind | TOO SLOW TO REACH SHELTER | DON'T WAIT FOR ME | | 4 | animals crammed in a shelter | NINETEEN STRANGERS, ONE ROOM | A TIN OF SARDINES | | 5 | doe + fawns at a bare wall | NO SAFE ROOM, NOWHERE TO RUN | NOWHERE TO GO | | 6 | Hope + cub racing the clock | FIFTEEN SECONDS | TOO LATE | | 7 | Marcus over cubs, struck shelter | THEY DID EVERYTHING RIGHT | THE SHIELD | | 8 | exhausted parent on watch | EVERYONE ASLEEP BUT ME | THE WATCH | | 9 | a hare frozen at the siren | A WORLD FALLING APART | THE SOUND | | 10 | street dog, direct gaze | WE FEEL IT TOO | DO YOU SEE ME? |

In-model recipe (text rendered by the generator — Reve-only, no external editing)

When external tools are off the table, render the type inside the image. Reve runs a dedicated typography pass and is best-in-class at short legible text, so in-model covers are viable IF you respect the limits:

  • The magazine lockup = up to FOUR short strings: masthead title + tagline (one top block), the headline (lower third), the small YUVAL AVIDANI credit (corner). This exceeds the safe minimum, so discipline matters MORE, not less.
  • Quote exact strings, keep them short, prefer ALL-CAPS. Describe the font by style only: bold condensed uppercase sans-serif.
  • Reserve clean zones: uncluttered top band for the masthead lockup; a subtle dark gradient scrim under the lower-third headline; a clean dark corner for the credit.
  • The small credit and the tagline are the highest-risk strings — keep the credit YUVAL AVIDANI in caps. Fallback if text overflows or fights the image: drop the tagline first, then the credit, keeping masthead + headline.
  • No post-fix safety net (Reve-only): render at 4K and inspect EVERY string at 100%. If any glyph is wrong, RE-ROLL (Reve is stochastic on text) or edit that text node. Text is the double-penalized artifact — never ship a garbled page.

Prepping the photo prompt for type

Each series prompt reserves a clear top zone and a darker lower third as negative space, then names the masthead and headline as quoted ALL-CAPS strings inside the prompt.

Source: SKILL.md on GitHub

No alerts3mo3 checks · Risk SAFE
  • Gen Agent Trust Hub3mo

    The skill is a comprehensive instructional framework for AI image prompt engineering. It consists entirely of Markdown documentation and does not contain any executable code, scripts, or mechanisms for data exfiltration. All external references are informative and point to legitimate AI and photography resources.

  • Socket3mo

    No alerts

  • Snyk3mo

    Risk: LOW · No issues

Signed by skilld at 83e2083. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Steadyupdated 4 months ago

README badge

README badge for hoodini/ai-agents-skills/image-master