| name | report-to-flowchart-prompt |
| description | Turn a report, proposal, thesis chapter, research plan, or other long-form academic text into a highly detailed image-generation prompt for a flat research-overview flowchart. Use when Codex needs to read the source text, extract the internal logical structure between background, objectives, methods, modules, stages, inputs, outputs, and expected results, then produce a Nano Banana compatible prompt that tightly controls layout, wording, aspect ratio, palette, typography, and diagram style. Especially useful for Chinese academic materials such as doctoral proposals, NSFC-style research overview figures, technical roadmaps, and research framework diagrams. |
Report To Flowchart Prompt
Overview
Convert report text into a flowchart-image prompt that preserves the source logic rather than paraphrasing it loosely. Produce prompts for clean, blue-white, flat, research-overview diagrams with solid-color backgrounds, explicit text selection, and no 3D or decorative noise.
Read references/prompt-spec.md when writing the final prompt. Read references/relation-patterns.md when the source text is dense, nonlinear, or spans multiple research modules.
Workflow
1. Read the source as a logic system
Identify the report's real structure before drafting the image prompt:
- Research background or motivation
- Core problem or scientific question
- Overall objective
- Research modules or work packages
- Methods, data, models, or experiments attached to each module
- Sequence, dependency, branching, convergence, and feedback relations
- Expected outputs, metrics, or application targets
Do not flatten the report into a generic top-down diagram if the source implies parallel modules, staged iteration, or layered abstraction.
2. Extract only text that can appear inside the figure
Select diagram text directly from the report whenever possible. Compress only for fit, clarity, or visual balance.
- Keep technical terminology faithful to the source.
- Preserve distinctions between objectives, methods, and results.
- Do not invent claims, modules, or evaluation targets not grounded in the report.
- Prefer short noun phrases or verb-object phrases suitable for boxes and labels.
- When shortening, keep the same meaning and technical scope.
If the source uses Chinese, keep the figure text in Chinese unless the user asks for English.
3. Build the logic map before writing the image prompt
Create an internal representation of:
- Nodes: what boxes or labeled regions exist
- Edges: what relationship each arrow expresses
- Groups: what nodes belong to the same stage or module
- Reading order: how a reviewer should understand the diagram in 5 to 10 seconds
Use the relation patterns in references/relation-patterns.md to classify the structure. If several patterns coexist, choose one dominant reading path and subordinate the rest visually.
4. Choose the visual composition deliberately
Default to a landscape research-overview layout, because this matches proposal-style diagrams better than a poster or infographic layout.
- Prefer
4:3 landscape when the structure has 3 to 5 major stages.
- Prefer
3:2 landscape when the diagram has more horizontal breadth.
- Prefer
16:9 only when the user explicitly asks for a presentation-like canvas.
The figure must look like a polished grant-application overview chart:
- Minimal
- Flat
- Clean
- Structured
- Review-friendly
- Design-aware but not decorative
5. Write a high-control image prompt
The final answer should prioritize a single detailed prompt that an image model can follow with minimal ambiguity. The prompt must explicitly control:
- Aspect ratio and orientation
- Background color
- Overall visual style
- Palette, contrast, and accent usage
- Grid or spatial composition
- Exact box hierarchy
- Arrow directions and relation semantics
- Figure text content and language
- Typography style
- Restrictions against 3D, gradients that imply depth, shadows, mockup elements, and scene-like rendering
Use the output structure in references/prompt-spec.md.
Output Rules
- Output one final production-ready prompt by default.
- Include a short
Logic skeleton section before the prompt only when the source material is complex enough that the user benefits from seeing the extracted structure.
- Make the prompt concrete enough that another model could reconstruct the whole diagram without seeing the source report.
- Treat text fidelity as a hard constraint, not an optional suggestion.
- When box text is long, instruct the image model to keep the labels concise, aligned, and legible rather than reducing the font to an unreadable size.
Style Guardrails
- Prefer white or very light solid backgrounds.
- Use blue-white as the dominant palette, with restrained cyan or slate-blue accents only if needed.
- Keep the diagram flat and editorial.
- Avoid 3D, bevels, glassmorphism, dramatic shadows, photographic elements, and textured paper.
- Avoid dark backgrounds unless the user explicitly requests one.
- Avoid poster, game UI, sci-fi HUD, and isometric aesthetics.
- Avoid cartoon icons unless the source explicitly calls for them.
Failure Modes To Avoid
- Copying the report into the image prompt without restructuring it
- Producing a generic "technology roadmap" that loses the report's actual causal logic
- Mixing objectives, methods, and outputs in the same visual level
- Using vague wording like "some arrows connect the modules"
- Leaving text choice unspecified
- Asking for a beautiful diagram without constraining layout, spacing, and hierarchy
- Introducing colorful infographic styling that weakens academic credibility
