Skip to content

Global-AI-Group-Inc/content-generation-skills / skills/mira-image-prompting

mira-image-prompting

Verified

How to write stills for Mira AI's image models: GPT Image 2, 2.5 and 2.6, Nano Banana Pro and 2.1, Seedream 4, 4K and 5 Pro, Muse Image, Ideogram 4.5, Flux. Which model holds text, which keeps a product identical across a series, how the craft banks for light, optics, composition, grade and materials are used, and the words that make an image worse. Use it before any generate_image call on Mira, or when the user says "image prompt", "промпт для картинки", "нарисуй", "сгенерируй картинку", poster, packshot, banner, cover. NOT for video (mira-video-prompting).

FileKindSize
SKILL.mdskill md4.2 KB
references/composition.mdmarkdown2.3 KB
references/craft.mdmarkdown3.0 KB
references/grade.mdmarkdown2.3 KB
references/light.mdmarkdown3.7 KB
references/materiality.mdmarkdown8.0 KB
references/models.mdmarkdown7.0 KB
references/optics.mdmarkdown2.4 KB

references/craft.md

View on GitHub
<!-- generated from the registry by comfy_study/comfy_video/build_skills.py; edit the registry, not this file -->
# Building the image prompt

HOW TO BUILD THE PROMPT. Write ONE paragraph of flowing English prose, 90-160 words, in this order: subject with material detail -> composition and framing built for the exact aspect ratio -> background and depth -> lighting setup (key with its modifier and direction, then fill, rim, reflections) -> palette -> optics and shot size -> surface and texture detail -> grade -> where the calm empty area of the frame sits and which scenery fills it (open sky, a bare wall, an unbroken tabletop), written as part of the scene. CHOOSE, DO NOT COLLECT: exactly ONE lighting scheme, ONE optics entry, ONE grade and at most TWO materials from the banks below. Taking one item from every list produces a scene that cannot physically exist and is the single most common way these prompts go wrong. NEVER PRINT THE BANK KEYS. The keys (three_point, portra_400, macro_100 ...) are your selection interface; the prompt must contain the described light, optics and grade in plain words, and no key, no bank name and no bracketed label. Say what is physically true of the frame rather than how good it is: a reader who has not seen the image should be able to rebuild it from your sentences alone.

## Banned language

BANNED LANGUAGE. These models degrade on vague praise, so the prompt must not contain: stunning, incredible, epic, gorgeous, breathtaking, masterpiece, award-winning, ultra-detailed, hyper-realistic, photorealistic as a lone claim, 4k, 8k, HDR, best quality, professional as a lone claim, trending on artstation. They occupy space and steer nothing. NO TAG SOUP: write sentences, not comma-separated keyword strings. 'minimalist brutalist luxury photoreal' says nothing; 'cream background, one hero object, heavy black sans-serif set hard left, generous empty space at the right' says everything. NO BRAND SHORTHAND: never write 'like an Apple ad' or name a photographer or studio as a style. State the concrete visual facts that shorthand stands for. Every adjective must be checkable against the finished image. If it cannot be, delete it.

## When the user attached a real photo

WORKING ON TOP OF A REAL PHOTOGRAPH. You can see the attached reference images. Use that: name what is actually in them - the real object, the real garment, the words really printed on the label - instead of inventing a generic stand-in. BUT DESCRIBE THE CHANGE, NOT THE TEXTURE. When a subject is present in a reference photo, its texture is already real and better than anything a model can generate: do not re-describe that face's skin, that fabric's weave or that surface's finish. Say what must become different - the new setting, the new light, the new framing, the new grade - and let the reference carry the rest. Re-describing a real client's face is the fastest way to turn it into plastic, and it is the specific failure this rule exists to prevent. Reserve the material and texture language for subjects that exist only in your prompt.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 323kUpdated Claude CodeCodexVerifiedInstall