| name | gpt-image-prompt-builder |
| description | Assembles OpenAI gpt-image prompts with subject, style, composition, palette, and typography for quote cards and carousels |
| version | 0.1.0 |
GPT-Image Prompt Builder
Specialized prompt assembler for OpenAI gpt-image. Strong on typographic precision, quote cards, carousel slides, and inpaint or local edits. Emits the parameter object the OpenAI image API expects.
When to invoke
- Quote cards and typographic posts where the text rendering must be sharp and on-brand.
- Carousel slide decks with consistent template across slides.
- Before-and-after consulting frames that require precise edit on a portion of the image.
- Transparent background assets for compositing into other layouts.
- Any image task routed by
.specs/architecture/PROVIDERS.md to gpt-image.
Inputs
subject: string. What is in the image.style: string. Visual style descriptor (editorial, minimal, studio product shot, flat illustration).composition: string. Layout cues (centered, rule of thirds, full bleed, top half text bottom half image).palette: array of strings. Color cues (muted earth tones, #1a1a1a background, gold accent).typography: object, optional. { text, font_style, weight, color, position }. Required for quote cards.aspect: string. 1:1, 4:5, 9:16, 16:9.transparent_background: boolean, optional. When true, request a PNG with alpha.negative: array of strings, optional. Things to avoid.seed: integer, optional.
Process
- Validate inputs. Reject if
subject, style, composition, aspect are missing.
- Compose the prompt in the order gpt-image responds best to: subject, style, composition, palette, typography (if present), then technical hints.
- For typography, embed the exact text in quotes and pin font style, weight, color, and position.
- Append negative prompt items as a separate string if present.
- Resolve the OpenAI parameter object: prompt, size (mapped from aspect), background (transparent if requested), n (default 1), seed if provided.
- Run a token sanity check on the prompt length.
- Return the assembled prompt, the parameter object, and the model id (
gpt-image-1 default).
Outputs
prompt: string.negative_prompt: string.params: object. Ready for the OpenAI Images API call.model: string. Default gpt-image-1.
Examples
Example 1: quote card for a launch post
Input: { subject: "minimal beige card", style: "editorial minimal", composition: "centered text", palette: ["#f5efe6 background", "#1a1a1a text"], typography: { text: "Sua cor te trai. Descubra qual e.", font_style: "serif display", weight: "regular", color: "#1a1a1a", position: "centered" }, aspect: "1:1" }
Output: { prompt: "Minimal beige card, editorial minimal style, centered text composition, palette beige #f5efe6 background and dark #1a1a1a text, typography 'Sua cor te trai. Descubra qual e.' in serif display regular black centered, 1:1 aspect", params: { size: "1024x1024", background: "opaque", n: 1 }, model: "gpt-image-1" }
Example 2: transparent product cutout
Input: { subject: "amber perfume bottle", style: "studio product shot", composition: "centered", palette: ["transparent"], aspect: "1:1", transparent_background: true }
Output: parameter object with background: "transparent" and PNG output.
Failure modes
- Typography text exceeds the area implied by composition: warn and suggest shortening or splitting across slides.
- transparent_background true but subject implies a background scene: warn and proceed; the model may render a default background.
- Aspect ratio not in the OpenAI supported set: map to the closest supported size and surface the change.
- Negative prompt conflicts with subject: drop the conflicting term and warn.
Related skills
wavespeed-batch: cheaper batch alternative when the typographic precision is not the priority.qa-tech-specs: validates the generated image against platform specs.caption-multi-platform: pairs caption variants with the generated quote card.compliance-generic: audits any text rendered inside the image.
