image

name	image
description	Use whenever Kosta invokes /image or asks Hermes to generate, draw, create, mock up, render, or edit an image using the default Codex/GPT Image 2 path. This is the preferred default image-generation skill; use ComfyUI only when the user explicitly asks for local models, LoRAs, SDXL/Flux workflows, or advanced node/control workflows. Turns loose requests into strong GPT Image 2 prompts and calls Hermes image_generate.
version	0.3.0
author	Hermes Agent
license	MIT
targets	["hermes-default","hermes-gpt","claude-hermes"]
metadata	{"hermes":{"command_priority":500,"tags":["image-generation","gpt-image-2","codex","image","prompt"]}}

Image — Codex GPT Image 2

Use this as Kosta's default image command. /image <request> means: understand the desired artifact, write a good GPT Image 2 prompt, then call Hermes' image_generate tool.

Do not route ordinary /image requests to ComfyUI. ComfyUI is for explicit local/LoRA/workflow requests. The default path is Codex/OpenAI GPT Image 2 through Hermes image_generate.

First move

Generate immediately. Do not ask clarifying questions unless the user's intent is genuinely ambiguous (e.g., "image" with no other content, or an unresolvable conflict like "landscape portrait of a building in square format"). All of these are clear enough — proceed without asking:

• Any subject described in ≥4 words
• Any /image <noun/scene/concept> with obvious visual interpretation
• Requests that name style, mood, format, brand, or use case
• Edit requests with before/after described
• UI/mockup/screenshot requests (even if only the category is named)

Never ask about: style, aspect ratio, level of detail, color palette, realism vs. illustration, number of variants, model, or background — unless the user explicitly raised it. Infer defaults from context.

Aspect ratio inference

Pick the ratio before prompting. Use the first matching rule:

Trigger words / context	Ratio
wide, banner, hero, desktop, cinematic, landscape, panorama, scene, wallpaper (horizontal)	landscape
phone, mobile, portrait, poster, tall, story (IG/social), wallpaper (vertical)	portrait
everything else (product shot, logo mark, avatar, icon, thumbnail, default)	square

When the request contains conflicting signals (e.g., "portrait of a landscape"), pick portrait for human subjects and landscape for scenes.

Prompt structure

Rewrite the user request into a concise production prompt. GPT Image 2 responds best to structure, not keyword sludge.

Use this order when it helps:

Create/draw/render [artifact type and intended use].

Scene / background:
[where it exists, environment, mood]

Subject:
[main subject, pose/action, scale, expression, important relationships]

Visual details:
[materials, textures, era, palette, realism/style, visible props]

Composition:
[framing, viewpoint, placement, negative space, orientation]

Lighting:
[soft window light, golden hour, studio, high contrast, etc.]

Exact text, if any:
"TEXT TO RENDER" — [font style, color, placement]. No other text.

Constraints:
[no watermark, no logos/trademarks, no extra text, preserve X, avoid Y]

Short requests do not need the whole template. A crisp paragraph is fine when it carries the same information.

GPT Image 2 rules that matter

• Use action words: create, draw, render, or edit. For edits, say "edit the image by changing X" rather than vague "combine/merge."
• For photorealism, include photorealistic or "real photograph / professional photography / iPhone photo." Do not rely on "8K, ultra detailed, masterpiece" boilerplate.
• Be concrete about visual facts: materials, shape, texture, placement, lighting, camera angle, and mood.
• For text in the image, put exact copy in quotes, specify typography and placement, and say verbatim, no extra text. Spell unusual words letter-by-letter if accuracy matters.
• For edits or reference-based work, separate Change from Preserve. Example: "Change only the background to a rainy Tokyo street. Preserve the person's face, pose, clothing, camera angle, lighting direction, and color grade."
• Iterate with small changes. If the first result is close, ask for one targeted edit rather than rewriting everything.
• For dense charts/spreadsheets/tiny labels, warn that image generation is the wrong final-format tool; use real document/design tooling if precision matters.
• GPT Image 2 does not support transparent backgrounds in the current API path. If transparency is requested, generate on a plain high-contrast background or use post-processing unless the configured provider changes.

Calling the tool

image_generate(prompt=<final prompt>, aspect_ratio="landscape" | "square" | "portrait")

aspect_ratio must be exactly one of "landscape", "square", or "portrait". No other values are accepted.

After generation, return the image directly. Add only a short note if useful, e.g. "I kept it square and optimized for product-shot realism."

Failure recovery

If image_generate returns an error or empty result:

1. Content policy rejection — rewrite the prompt to remove potentially flagged phrasing (e.g., real brand names, explicit depictions, celebrity likenesses). Try once with cleaned prompt before reporting.
2. Tool unavailable / timeout — report: "image_generate failed: [error]. Try again or check Hermes tool status."
3. Result misses a key explicit constraint — after receiving the result, check it against the request's primary constraints: required text present and legible, correct orientation for the use case, correct subject. If a constraint is clearly violated (text missing or garbled, wrong orientation, wrong subject count), identify the broken constraint, rewrite that section of the prompt only, and retry once automatically — do not ask permission. Deliver the final result with a one-line note on what was fixed. If the second attempt also fails the constraint, report both attempts and the specific mismatch.

When to use high-effort prompting

Use the full structured prompt for:

• UI mockups, screenshots, app screens, product concepts
• Posters, ads, packaging, logos/marks, thumbnails with text
• Infographics and diagrams
• Photorealistic product/people/scene images
• Edits where identity, geometry, layout, or brand feel must be preserved

For UI/UX images, also follow the gpt-image-2-uiux-prompting skill if available.

Examples

User: /image a slick product shot of a black ceramic mug for a coffee brand

Tool prompt:

Create a photorealistic studio product photograph for a premium coffee brand.

Subject: a matte black ceramic mug with a subtle curved handle, empty, centered on a warm stone surface.
Visual details: soft ceramic texture, tiny rim highlights, faint coffee-bean shadows nearby, no visible logo.
Composition: square crop, mug centered with tasteful negative space, eye-level product photography.
Lighting: soft warm key light from upper left, gentle fill, natural contact shadow.
Constraints: no text, no watermark, no fake brand logo, no extra objects except subtle coffee beans/shadows.

User: /image mobile onboarding screen for a meditation app, calm but not generic

Tool prompt:

Generate one realistic iPhone portrait screenshot of a native iOS meditation app onboarding screen.

Product intent: help a new user choose a calming daily practice without feeling like a wellness cliché.
Screen state: first onboarding screen with a single hero illustration, short headline, one primary CTA, and small secondary sign-in link.
Layout: native iOS safe areas, SF Pro-like typography, generous spacing, bottom CTA above the home indicator.
Visual style: calm editorial UI, warm off-white background, muted sage and clay accents, subtle grain, no stock-photo look.
Exact visible text: headline "Find your quiet minute", CTA "Start", secondary link "Sign in". No other text.
Constraints: straight-on screenshot, normal iPhone proportions, no device mockup frame, no poster layout, no duplicate buttons, no fake logos, no watermark.

References

• references/gpt-image-2-prompting.md — current GPT Image 2 prompt rules and constraints.