media-content-image-prompting

name: media-content/image-prompting description: > Craft optimized image generation prompts for photorealistic, editorial, cinematic, and commercial photography. Use when generating images, photos, portraits, product shots, mood boards, or any visual asset. Supports physical format specifications (banners, cans, posters, apparel) and context-aware camera/lens selection. Works with nano-banana-pro (Gemini 3 Pro) for execution.

Image Prompting

Core Workflow

1. Analyze intent — what, who, where, why, mood, format
2. Select prompt formula — photorealistic, character-focused, cinematic, enhanced lighting, etc.
3. Choose equipment — camera + lens matched to scenario (read references/visual-assets.md for full tables)
4. Layer detail progressively — subject → environment → lighting → technical
5. Apply physical format if user specifies dimensions (read references/dimensions-formats.md)
6. Execute via nano-banana-pro skill
7. Offer 2-3 alternative directions with different angles/styles

Prompt Structure

Order matters — models weight elements at the beginning more heavily.

[Subject + Action], [Environment/Setting], [Medium (camera/lens/film stock)],
[Depth of Field], [Composition], [Lighting/Time of Day], [Color/Mood], [Textures]

Element Priority

Strategy: Place whatever needs emphasis FIRST.

Prompt Length

Use strong keywords, not full sentences. Separate with commas. One subject per prompt.

Power words: luminous > bright, jubilant > happy, enormous > big, ethereal > dreamy

Prompt Formulas

Photorealistic (General)

[Shot Type], [Photo Style], [Subject + Action], [Environment],
[Camera/Lens/Film Stock], [Depth of Field], [Color Scheme],
[Textures], [Composition], [Lighting]

Character-Focused

[Shot Type], [Subject + Action + Environment], [Expression],
[Wardrobe], [Body Position], [Unique Features], [Props],
[Camera/Film Stock]

Cinematic Still

Cinematic still, [Subject + Action], [Environment],
[Camera/Film Stock], [Cinematic Lighting], [Details/Modifiers]

Enhanced Lighting

[Lighting Conditions], [Subject + Action], [Environment],
[Camera/Film Stock], [Composition], [Details]

Enhanced Motion

[Motion Type], [Subject + Action], [Environment],
[Camera/Film Stock], [Composition], [Lighting], [Details]

Super Control (Maximum Detail)

[Shot Type], [Photo Type], [Subject + Action], [Subject Position],
[Wardrobe], [Props], [Subject Focus], [Environment],
[Environment Focus], [Environment Details], [Colors & Tone],
[Lighting], [Time of Day], [Textures], [Camera/Technical]

Progressive Detail Layering

Start simple, add incrementally. Each layer gives more control:

1. Subject: satisfied Brazilian woman sitting, looking at camera
2. + Framing: center of image, medium shot, editorial
3. + Focus: sharp focus, mid-ground, slightly blurred background
4. + Wardrobe: wearing muted blue oversized t-shirt, headband, red lipstick
5. + Environment: cozy favela bedroom, house plants, rundown and liberated
6. + Lighting: low light, light from window, dramatic shadows, early morning
7. + Aesthetic: smoke blending with light rays, muted 1990s color grading
8. + Technical: Bergger Pancro 400 film

Context-Aware Equipment

Match camera/lens to scenario:

For full camera, lens, film stock, lighting, and composition tables: read references/visual-assets.md.

Film Stock Quick Reference

Physical Dimensions

Gemini 3 Pro understands real-world format specs. State dimensions early:

Design a product label for a 2.75 inch aluminum energy drink can...
Create a horizontal vinyl banner, 6 feet wide by 2 feet tall...
Poster at 24x36 inches for a music festival...

Include material (vinyl, aluminum, canvas) and viewing distance if relevant.

For full format table and examples: read references/dimensions-formats.md.

Digital Asset Dimensions (Web / Blog / Social)

Always match dimensions to the target surface. Mismatched aspect ratios (e.g. square images for blog posts) waste generation cycles and require manual re-generation.

Include pixel dimensions in every prompt. Do not default to square unless the target is explicitly a social feed post.

Photography Elements That Create Realism

Simply adding a camera and lens transforms AI output into photorealistic results:

• ❌ African wild dogs, wilderness, hunt → looks artificial
• ✅ African wild dogs, wilderness, intense hunt, Sony A9 II, 85mm f/1.8 GM lens → photorealistic

Execution

Generate images via nano-banana-pro:

uv run {baseDir}/scripts/generate_image.py --prompt "your prompt" --filename "output.png" --resolution 1K

• Resolutions: 1K (default), 2K, 4K
• Timestamps in filenames: yyyy-mm-dd-hh-mm-ss-name.png
• Multi-image composition: add multiple -i flags
• Do not read the image back; report the saved path only

Image-to-Video Seeding

Generated images can seed video generation for maximum first-frame control:

# 1. Generate still image
uv run .../nano-banana-pro/scripts/generate_image.py --prompt "..." --filename "frame.png" --resolution 2K
# 2. Animate with Veo 3.1
uv run .../veo3-gen/scripts/generate_video.py --prompt "motion description" --filename "out.mp4" -i frame.png
# 3. Or animate with Sora 2
uv run .../sora2-gen/scripts/generate_video.py --prompt "motion description" --filename "out.mp4" -i frame.png

See media-content/video-prompting/SKILL.md § Image-to-Video Workflow for full details.

Key Rules

1. Specificity over generality — always specify camera, lighting, palette
2. One subject per prompt — multiple subjects confuse the model
3. Place emphasis first — whatever matters most goes at the beginning
4. Iterate, don't rewrite — change one element at a time
5. Always offer alternatives — provide 2-3 directions per request