Stats
Actions
Tags
From drawbridge
Use when crafting an image-generation prompt for a specific web UI target (Gemini Imagen 3, ChatGPT DALL-E 3, Grok Aurora, Midjourney). Translates a short user brief into a target-tuned prompt. Invoked by drawbridge commands; can also be used directly when shaping prompts for image generators.
How this skill is triggered — by the user, by Claude, or both
Slash command
/drawbridge:prompt-craftingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Each target has different sweet spots. Pick the section that matches `target`, then expand the brief.
Each target has different sweet spots. Pick the section that matches target, then expand the brief.
Before tuning per-target, decide style from context, not from a fixed preset:
Always include: subject, environment, lighting, composition, mood, technical params (camera/render hints, aspect ratio). Exclude: identifiable real people unless the brief explicitly names one — say "person resembling X" only if the brief insists.
If translate_to_english is enabled in settings, the final prompt MUST be English regardless of brief language.
target=geminiURL: https://gemini.google.com/app
Format: flowing natural-language paragraph. ~480 token soft limit. Single paragraph or two short ones — no bullets, no comma-tag style.
Strengths: photorealism, environments, product shots, lighting nuance. Weak spots: stylized illustration is hit-or-miss; on-image text often broken.
Negatives: inline ("no text in the image", "no extra fingers"). Imagen has no formal negative prompt.
Aspect ratio: mention it in prose ("widescreen 16:9", "vertical portrait 3:4"). Imagen accepts limited ratios via the UI; the prompt itself is the hint.
Template:
A [subject doing X] in [environment]. [Lighting and time of day]. [Composition: framing, angle, depth of field]. [Mood and atmosphere]. Shot on [camera/lens hint] for photorealism, ultra-detailed, [aspect ratio]. No text, no logos.
Example brief: "уставший разработчик ночью у компьютера"
Example prompt:
A weary software developer at his desk late at night, illuminated only by the cold blue glow of three monitors and a half-empty mug of coffee beside the keyboard. Soft volumetric haze in the room, warm desk lamp catching the edge of his face. Medium close-up, 35mm lens, shallow depth of field, slight rim light from behind. Quiet, contemplative atmosphere. Photorealistic, cinematic colour grade, 16:9. No text, no logos.
target=chatgptURL: https://chatgpt.com
Format: clear, structured natural language. GPT-4 rewrites your prompt internally before passing to DALL-E, so over-engineering is wasted — be specific but concise.
Strengths: concepts, in-image text (DALL-E 3 handles short legible text), cohesive scenes. Weak spots: photorealistic faces look slightly plastic.
Negatives: phrase positively where possible. ("clean uncluttered background" beats "no clutter").
Aspect ratio: specify in the prompt — "wide 16:9", "square", "portrait 2:3". DALL-E supports 1024×1024, 1792×1024, 1024×1792.
Template:
[Subject and key action]. [Setting / environment]. [Mood, palette]. [Camera or art-style hint]. [Aspect ratio].
Example prompt: for the same brief:
A weary software developer slumped at his desk past midnight, lit only by three glowing monitors and a single warm desk lamp. Modern apartment workspace, papers scattered, an empty coffee mug. Cinematic, moody, blue-orange palette, shallow depth of field, photo-real. Wide 16:9.
target=grokURL: https://grok.com
Format: dense descriptive prose tolerated; long prompts work fine. Aurora rewards specificity in textures, materials, and lighting language.
Strengths: photorealism with people, looser content rules than Imagen/DALL-E, gritty / cinematic looks. Weak spots: abstract or stylised illustration sometimes drifts toward photoreal.
Negatives: include a short negatives clause at the end ("avoid: text, watermark, distorted hands").
Template:
[Subject + action + emotional state]. [Detailed environment with materials and textures]. [Lighting setup with sources and quality]. [Camera, lens, film stock hint]. [Mood, palette, post-grade]. Avoid: [negatives].
target=midjourneyURL: https://www.midjourney.com/imagine
Format: comma-separated tags + parameter suffixes. Different syntax from text-prose targets.
Tags order: subject → setting → style → mood → technical.
Parameters: --ar W:H aspect ratio, --v 6 model version, --style raw (less stylised), --s N stylise strength (0–1000), --no <thing> for negatives, ::N for token weights.
Template:
[subject], [action], [setting], [style/medium], [lighting], [mood], [camera/lens], [palette] --ar 16:9 --v 6 --style raw
Example prompt: for the same brief:
weary software developer, slumped at desk, late night home office, three glowing monitors, coffee mug, photoreal cinematic, blue-orange grade, soft volumetric haze, 35mm lens shallow depth of field, melancholic atmosphere --ar 16:9 --v 6 --style raw --no text, watermark
When invoked from a drawbridge command, return ONLY the final prompt — no preamble, no labels, no quotes. The command will pipe it into pbcopy directly.
If translation to English is requested, translate first, then craft per-target. Never craft Russian-first then translate — the result loses idiomatic precision.
npx claudepluginhub ivanlutsenko/awac-ai-agent-plugins --plugin drawbridgeProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.