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.

references07-pre-submit-inspection.md

≈1.3k tokens on demand. Your agent reads this file only when SKILL.md points to it.

07 — The Forensic Pre-Submit Inspection

Run twice: PRE-FLIGHT on the prompt (predict the likely artifact, pre-empt it in words) and POST-FLIGHT on the output (the judges will zoom in — so do you). The contest explicitly penalizes the 13 artifacts below; this checklist is built to catch each.

PRE-FLIGHT (on the prompt, before generating)

  • Blacklist swept? No 8k / ultra / hyperreal / masterpiece / octane / flawless / smooth / beautiful / glossy.
  • One light only? Direction + quality + resulting shadow direction named.
  • Anti-plastic present? Pores / vellus / subsurface / texture-revealing light / film grain for any skin or fur.
  • Hands handled? Hidden, posed simply, cropped, or on one solid object — never a complex free hand unless it's the hero (then plan to inpaint).
  • Faces sized? Hero faces large in frame; gaze + catchlights specified; slight asymmetry; ≤2 faces or expect inpaint.
  • Text safe? Quoted + short + ALL-CAPS for the hero string; background text left vague; Hebrew/exact text routed to post.
  • Negatives converted? On Reve/GPT-image/Nano Banana, every "no X" rewritten as a positive.
  • Camera real? Body + lens + aperture + shutter/ISO named.
  • Depth staged? Fore/mid/background + atmospheric haze.
  • One impact lever locked? A color recipe AND a composition/drama device.

POST-FLIGHT (on the generated image — zoom to 100%)

Inspect in this order; the first failure usually hides the others.

# Penalized artifact Where to look Pass test
1 Hands rendering every visible hand/paw exactly 5 fingers, natural joints, no fusion/extra; thumbs correct
2 Object & hand hand–object contact points fingers wrap correctly; object geometry consistent, not melting into the hand
3 Anatomical error limbs, joints, count, proportion no extra/missing/floating limbs; natural posture; symmetric pairs match
4 Facial rendering eyes, teeth, ears, symmetry eyes aligned + focused, both catchlights match the key; no melted ear/teeth; slight (not dead) symmetry
5 Artificial / plastic skin cheeks, forehead, nose visible pores, real tonal variation; NOT waxy/airbrushed/uniform
6 Unnatural material fabric, metal, fur, water, foliage weave/grain/strands/wear visible; light behaves correctly on the surface
7 Rendering issue (light/reflection) shadows, mirrors, glass, wet ground, eyes all shadows agree on ONE light direction; reflections contain the actual scene
8 Typography every glyph, foreground and background hero text spelled right, clean kerning; no invented characters in background signage
9 Image quality / artifacts edges, gradients, repeats, smears no oversharpen halos, banding, duplicated patterns, smudge zones, warped backgrounds
10 Composition sanity horizon, perspective, scale horizon level; consistent vanishing point; subject scale plausible vs surroundings
11 Species consistency every animal, esp. cubs/young/groups each animal is unmistakably its correct species with the right markings (lion cubs ≠ cheetahs); no generic spotted "big-cat cub"
12 Reflection accuracy eyes, wet ground, glass, water every reflection contains the ACTUAL scene/light in the correct position and direction; catchlights match the key light

The PRECISION LOCK (bake into every prompt's tail)

Append this to any prompt, setting the species line per image:

Every animal anatomically correct and true to its EXACT species and markings — [species note, e.g. "lion cubs are tawny with rounded faces and only faint juvenile rosettes, never cheetah spots or tear-stripes"]. Correct limb count, each paw/foot planted and clearly attached, five toes with natural pads and claws, no fused, extra or floating limbs. A single dominant light source with all shadows falling consistently away from it — no contradictory shadow directions. Accurate reflections: light sources and the scene mirror correctly in eyes, wet surfaces and glass, matching their real position. True materials, correct perspective and scale. Photoreal down to whiskers, individual fur/hair direction, pores and catchlights. No anatomical, material, typography or perspective errors.

Triage when something fails

  1. One local element (a hand, a sign, an eye) → on Reve, edit that node / inpaint that mask only (denoise 0.35–0.5) — do NOT re-roll the whole image and lose a good composition.
  2. Systemic (plastic skin everywhere, wrong light all over) → it's a prompt problem; fix the words (add texture/one-light), regenerate.
  3. Text wrong → fix BEFORE upscaling; if exact/brand/Hebrew, blank it and composite in post.
  4. Still failing after 2 inpaints → change the pose/crop to hide the failure zone (back-turned, hands away, face larger) rather than fighting the model.

The submit gate

An image ships only when: it passes all 10 post-flight rows at 100% zoom, it pulls at least one deliberate impact lever, and — for a SET — it adds range the other images don't already cover. If it only "looks fine at thumbnail size," it is not done; the judges zoom in.

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