Skip to content

veryCoolTimo/imagegen-skills / skills/image-prompt

image-prompt

Flagged

Use when the user wants to turn a short idea into a rich, production-grade image-generation prompt — posters, landing-page or UI mockups, ads, editorial layouts, photoreal scenes, game screenshots, logos, or illustrations. Builds ONE structured, copy-paste-ready prompt optimized for OpenAI GPT Image 2.6 by default, following proven layered-composition patterns (named zones, hex palette, real font references, mood/finish, IP-safety). Triggers on requests like "сделай крутой промт для постера", "нужна картинка/постер/лендинг для…", "generate a prompt for a landing hero", "prompt for an ad / logo / game screenshot".

references/edit-remix.md

View on GitHub
# Edit / remix workflows (GPT Image 2.6)

Use when the user provides (or refers to) an existing image and wants to transform, composite,
restyle, localize, or place someone/something into it. This is an **edit** (image[+text]→image,
`client.images.edit`), not a generate. Still prompt-only output — the user attaches the image(s)
in their tool and pastes the prompt.

The params here (`markup`, `background`, `complexity`) are **OpenAI GPT Image's**. The change-vs-preserve
discipline is universal, but for edits targeting another model use that model's own editing
section for its params: Gemini/Nano Banana Pro, FLUX.2 (edit = T2I + reference images), Recraft,
and Reve each cover editing in their adapters.

## Core discipline (from the OpenAI image-prompting guide)

- **Change vs preserve.** State exactly what to change, then "keep everything else the same".
  List the invariants explicitly (identity, geometry, layout, palette, camera, background,
  logos) and **repeat the preserve list on every iteration** — this is what stops drift.
- **Surgical edits.** If the change is small, also say: do not alter saturation, contrast,
  layout, labels, arrows, camera angle, or surrounding objects.
- **Multiple inputs.** Reference each by index + role ("Image 1: the person's photo; Image 2:
  the product"). Be explicit about which element moves where.
- **In-image text.** Literal copy in quotes, verbatim, rendered once; specify font + placement.
- **Params.** Omit `input_fidelity` (2.x always uses high input fidelity). `complexity=low` for
  a targeted single-detail edit; `background=transparent` + PNG/WebP for cutouts;
  `quality=high` for small text / portraits; pick `size` to match the source aspect or the
  intended output. Use `gpt-image-2.6-sunburst` for precise edits.
- **Markup.** To point at a region, the user can send a transparent PNG overlay (`markup`,
  same pixel size as Image 1) with a circle/arrow drawn on it. The prompt then says "the X
  inside the red circle … Remove the red circle from the finished image". Markup applies only
  to the first input; other images are references. Don't send `mask` together with `markup`.
- **Pixel-identical regions.** Repeated edits can still drift; if something must stay exactly
  the same, tell the user to composite the approved edit back into the original.
- **Iterate** with single-change follow-ups, re-specifying critical invariants each time.

## Edit output template

```
[One line: what this edit does + intended use. Note: this is an IMAGE EDIT.]

INPUTS:
- Image 1: <what it is / its role>
- (Image 2: <…>)

CHANGE:
<only what should change>

PRESERVE (do not alter):
<identity / geometry / layout / palette / camera / background / logos — the invariants>

[Add STYLE / SCENE / TEXT sections as needed, using the normal 9-block discipline.]

PARAMS: model=gpt-image-2.6-sunburst · size=<WxH> · quality=<low|medium|high> [· complexity=low] [· background=transparent] [· markup=overlay.png]
CONSTRAINTS: no watermark, no added logos or trademarks; render each text string once, verbatim.
```

## Workflow templates

| Workflow | Change → Preserve | Key params |
|---|---|---|
| Identity-preserving hero / character-in-poster | build a poster around the person in Image 1; add costume/scene → **preserve exact face, features, skin tone, likeness** | quality=high |
| Insert person into a scene | place the Image 1 person into a described scene → preserve their likeness; grounded photo look | quality=medium/high |
| Multi-image composite | move the element from Image 2 into Image 1, matching lighting/perspective/scale/shadows → preserve the rest of the scene | quality=medium/high |
| Style transfer | apply Image 1's style (palette, texture, brushwork, grain) to new content → keep style cues, change subject; no extra elements | quality per detail |
| Outfit swap (virtual try-on) | replace only the clothing with the garments from Images 2..N, fitted to the existing pose → preserve face, features, skin tone, body shape, pose, hairstyle, background, camera, framing; match lighting | quality=medium/high |
| Background / scene swap | replace the background → **preserve the subject exactly**; match new lighting + contact shadows | quality=medium/high |
| Lighting / weather transform | change only environmental conditions (time of day, weather, mood) → preserve identity, geometry, camera, placement | short follow-up on the previous output |
| Transparent product cutout | isolate the product on a fully transparent background; centered, crisp silhouette, no halos/fringing; light polish only; no backdrop, checkerboard, scenery, or shadow → **preserve geometry + label legibility**, don't restyle | background=transparent, png |
| Product on clean white | extract product to a plain opaque background; crisp silhouette, no halos; light polish + subtle contact shadow → **preserve geometry + label legibility** | background=opaque |
| Marked-region edit | change only the object inside the red circle (markup overlay) → keep its shape + the rest of the image; remove the circle | markup, complexity=low |
| Marketing text / billboard mockup | place the product into a scene + add the exact tagline (in quotes, once, legible) → preserve product + label | quality=high |
| Object add / remove | remove or add only X → keep everything else the same ("Do not change anything else.") | complexity=low |
| Interior / object swap | replace only object X (e.g. the chairs) → preserve camera, room lighting, floor shadows, surrounding objects | quality=medium/high |
| Translate / localize | translate the in-image text verbatim to <language> → preserve typography, placement, spacing, hierarchy; don't touch logos, icons, imagery, layout | quality=high |
| Sketch / wireframe → render | turn the drawing into a photoreal image or shipped UI → preserve layout, proportions, perspective; add plausible materials/lighting; add no new elements or text | quality=high |

## Notes

- Detect edit intent from cues: "add me to…", "change the background", "in this photo",
  "turn this sketch into…", "same style as this", "remove the…", "put X into Y", "localize /
  translate this", "extract the product", "make it night / snowy".
- If the idea is a poster/ad built around a supplied person or product, that is the
  identity-preserving hero workflow (build the poster/ad around the supplied person, preserving
  their exact face and likeness; add only wardrobe/scene).
- Keep IP-safety: original design, no added logos/trademarks/watermark; for likenesses, only
  the person the user supplied — do not invent recognizable real people.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall