Communitygithub.com

Mr-funny/hbg-epic-giant-worlds

Three HBG Agent Skills for keyword colossal worlds, Eastern colossal adaptation, and prompt-bank roaming.

Qu'est-ce que hbg-epic-giant-worlds ?

hbg-epic-giant-worlds is a Gemini CLI agent skill that three HBG Agent Skills for keyword colossal worlds, Eastern colossal adaptation, and prompt-bank roaming.

Compatible avec~Claude Code~Codex CLI~CursorGemini CLI
npx skills add Mr-funny/hbg-epic-giant-worlds

Demander à votre IA préférée

Ouvre une nouvelle conversation avec cette compétence d'agent déjà préchargée.

Documentation

HBG Keyword Giant World

Turn one short keyword into an impossible inhabited giant construct. Everything can become architecture, terrain, weather, a world-body, or a new physical law. Keep the rendering believable while making the premise impossible.

Read first

  • Read references/world-expansion-system.md before choosing motifs or composition.
  • Read references/colossal-transmutation.md before accepting any giant concept.
  • Read references/prompt-contract.md before writing image or video prompts.
  • For image generation, use the environment's image-generation skill and obey its visible-delivery rules.
  • For video creation, read the environment's mandatory video skill first. Keep Gemini/Veo when the user explicitly chooses it.

Input contract

Accept one keyword as enough input: a country, region, religion, civilization, city, myth, person, action, object, emotion, or abstract idea. Infer a respectful default. Ask only when an ambiguity would materially change the culture or subject.

If the user asks only for prompts or a proposal, stop before paid generation. “生成”“直接做”“先来一版” authorizes one first draft.

Build the concept card

Create this compact card before prompting:

Keyword:
Type:
Visual thesis: <keyword> × <representative environment> × <core spirit> × <one impossible-scale wonder>
Cultural anchors:
Reality baseline to reject:
Giant transmutation: <what the keyword itself becomes>
Category fusion: <at least two normally separate domains>
Impossible law and visible consequence:
Foreground family:
Midground human stage:
Tiny figures and action:
Atmospheric expanse:
Primary world-body or giant construct:
How people inhabit, cross, excavate, worship at, or live inside it:
Three independent scale proofs:
Cosmic anchor, if any:
Camera relationship and lens:
Materials and palette:
Light and mood:
Sensitive or forbidden elements:
Absurdity score / 10:

Choose one thesis, not a collage of every associated symbol.

“Everything becomes a giant construct” rules

  1. Make the premise impossible before making the rendering realistic. Realism describes light, material, gravity effects, mist, water, and wear; it must not reduce the idea to a plausible landmark.
  2. Transform the keyword itself. Do not merely place a normal temple, village, volcano, gate, statue, or moon beside it.
  3. Fuse at least two normally separate categories: anatomy + city, object + continent, animal + sky, craft + geology, action + weather, shadow + architecture, or interior + horizon.
  4. State one impossible physical law and show a second-order consequence: rivers follow palm lines, bridges enter an eye socket, a creature's belly becomes the sky ceiling, cloth folds hold cities, or columns carry a cloud ocean that spills as waterfalls.
  5. Make people use the giant form. Add roads, stairs, dwellings, bridges, ports, terraces, rituals, or maintenance paths attached to or inside it.
  6. Provide at least three independent scale proofs: ant-sized people, human architecture attached to the giant, weather or water interacting with it, atmospheric occlusion, horizon curvature, or long traversal infrastructure.
  7. Keep figures below about 1–2% of frame height. Use one dominant giant construct controlling roughly half or more of the visual field and at most one cosmic anchor.
  8. Reject any concept that could be photographed as a real travel destination with only a larger building, dramatic weather, or a normal natural event.
  9. Do not default to a giant tree. Rotate body, animal, object, craft, terrain, architecture, negative space, water, and weather transformations.
  10. A colossal form may be fully visible or extend beyond the frame. Scale comes from wide-angle proximity, inhabitation, physical interaction, leading lines, atmosphere, and cognitive contradiction—not cropping alone.
  11. Respect living religions and cultures. Use culturally valid sacred sculpture or symbolic space carefully; do not fabricate scripture, secret rites, sacred-person depictions, or incompatible traditions.

Reality-rejection gate

Reject before prompting if any answer is yes:

  • Is this essentially a real temple beside a volcano, a normal sea gate, a large village roof, or a familiar landmark with dramatic clouds?
  • If the people disappeared, would the central form read only as an ordinary mountain, building, statue, animal, or object?
  • Could ordinary engineering plausibly build it at a smaller scale without changing its fundamental idea?
  • Is the keyword merely decoration while a generic palace or mountain remains the true subject?

Require an absurdity score of at least 8/10 using references/colossal-transmutation.md. If it fails, generate three stronger transmutations and keep the least reality-bound one.

Camera variation

Choose camera separately from subject:

  • ground-skimming low angle, 14–18 mm;
  • eye-level wide shot, 18–24 mm;
  • 25–55° oblique high angle, 18–28 mm;
  • high aerial overlook retaining a horizon, 20–35 mm equivalent;
  • selective 70–90° top-down for processions, mandalas, terraces, water geometry, or impossible engineering.

For three or more images, use at least three camera relationships, three foreground families, and three leading-line patterns. Do not repeat tree frame + lower-right platform + distant palace as a house template.

Still-image workflow

  1. Write the concept card and exact 16:9 mother-image prompt.
  2. Run the reality-rejection and absurdity gates. Do not generate a merely scenic concept.
  3. If generation is requested, generate an ultra-wide cinematic still.
  4. Inspect whether the giant form dominates, whether humans visibly inhabit it, and whether three scale proofs survive the image.
  5. Inspect cultural coherence, architecture, perspective, anatomy, duplicated motifs, false writing, logos, and watermarks.
  6. Save the prompt and selected still in a versioned project folder.
  7. A generated image is delivered only when it is visibly rendered in chat and its absolute path is provided.

Motion workflow

Never paste the still prompt into Veo. Inspect the actual mother image, then rewrite motion from visible anchors.

Lock:

  • foreground object and its parallax behavior;
  • midground stage and tiny figures;
  • atmosphere and existing light direction;
  • megastructure geometry, materials, and silhouette;
  • existing palette and composition.

Select one principal camera path and one environmental action family. Prefer one continuous 8-second shot, slow stabilized movement, restrained people, and layered parallax. Keep distant architecture nearly fixed. Ban cuts, snap zooms, fast orbits, morphing, structure growth, new landmarks, enlarged people, false text, and arbitrary creatures.

Multi-shot mode

For four scenes, design four distinct realms:

  1. Encounter — extreme-proximity low angle revealing only part of the world-body and human infrastructure.
  2. Passage — eye-level travel through, beneath, or across the giant construct.
  3. Revelation — oblique high view explaining the impossible category fusion and inhabitation.
  4. Sublime — aerial or selective top-down view showing how the giant construct rewrites terrain, weather, or horizon.

Use four different transmutation families, not four enlarged buildings. Lock a shared material family, restrained palette, and cultural thesis; vary camera, foreground, leading line, weather, and motion. Generate and verify each clip independently before stitching.

Required outputs

For prompt-only work: concept card, image prompt, motion prompt, and negative constraints.

For completed media: visible mother image, saved prompt paths, playable video, verified duration/codec/dimensions, and a representative contact sheet.

Skills associés