GPT Image Teaser

Overview

This skill turns paper content into prompts tailored for GPT Image style academic figure generation. It is best when the user wants a compact, paste-ready OpenAI image prompt plus a practical revision loop for teaser figures and method diagrams.

When To Use

Use this skill when the user asks for any of the following:

• A GPT Image prompt for a paper teaser
• An OpenAI image-model workflow for academic diagrams
• A one-shot or two-pass prompt for method figures
• A revision prompt to fix labels, arrows, layout, or caption leakage in a generated academic figure

Prefer paperbanana-teaser instead when the user explicitly wants the full planner-stylist-visualizer-critic decomposition. Use this skill when the user wants the GPT Image-facing layer distilled into a simpler workflow.

Inputs

Collect as many of these as possible:

• Paper topic
• Core problem
• Prior-work limitation
• Proposed method or system
• Main technical components
• Main takeaway
• Methodology section
• Figure caption
• Target conference style
• Optional reference image for style only

If the input is incomplete, preserve what is known and simplify the visual narrative instead of inventing unsupported details.

Default Workflow

Use this lightweight sequence:

1. Build a Figure Brief

Summarize the problem, method, and outcome into a faithful teaser narrative.

1. Draft the GPT Image Prompt

Convert the brief into a model-facing prompt with explicit text, layout, and academic-style constraints.

1. Run a Revision Pass

If the first image has bad text, wrong arrows, clutter, or title leakage, issue a focused correction prompt instead of rewriting everything from scratch.

Working Rules

• Preserve scientific fidelity. No invented modules, claims, or evidence.
• Do not place paper titles, figure captions, figure numbers, watermarks, or UI chrome inside the image.
• Keep labels short, exact, and readable.
• Prefer white or very light backgrounds and a restrained academic color palette.
• Use natural-language visual descriptions, not CSS, hex colors, or pixel specs.
• If the figure has many components, group them into panels or bounded regions to maintain readability.

Output Modes

• brief-only: return the structured figure brief
• gpt-image-prompt: return a paste-ready prompt for GPT Image
• prompt-plus-revise: return an initial prompt plus a follow-up repair prompt template
• caption-safe: emphasize title and caption exclusion for models that often leak text

Reference File

Read the full templates here when prompt text is needed:

• references/gpt-image-prompts.md

Use that file for the figure brief schema, the main GPT Image prompt, and the revision prompt templates.