Turn Source Text into a Hand-Drawn Sketchnote: Copy, Logic and Review
Reduce approved material to one theme and three to seven nodes. Prepare exact labels, review a 4:3 raster image and understand what the local validator can actually prove.
Fastest way
Ask your AI to do it
Paste this into the AI agent you already use:
Use Hand-Drawn Visual Notes to plan one sketchnote from my supplied source. Ask for audience and output language, preserve conditions, freeze one title and three to seven exact labels, and choose a clear relation pattern with drawn metaphors. Prepare the prompt without paid API calls. Inspect an actual image before marking review flags true; the validator does not perform OCR or fact checking. Do not generate or purchase image services without my approval.
What does Hand-Drawn Visual Notes produce?
The v1.0.0 workflow guides your agent from supplied text to one 4:3 landscape sketchnote raster image. It plans the meaning, exact wording, illustration metaphors and visual review. It does not deliver an editable diagram editor, perform automatic source verification or generate an image locally without a model. The complete free package requires a verified account; any image service has separate access and costs.
Read one complete source and preserve its conditions
Provide the actual text, document or accessible page, its audience and one question the image should answer. Mark statements as source facts, interpretation or unknown. If a page cannot be read, supply its text rather than asking the agent to guess. Do not turn a conditional step into a guaranteed outcome.
For practice, consider this fictional source: requests enter an inbox, a reviewer triages them, and an owner must approve a release before publishing. Freeze the title and the four labels below. Preserve the approval condition on the final relationship. No response-time target or success metric is supplied.
| Exact English copy | Role in the picture | Meaning to preserve |
|---|---|---|
| Request review · title | One large theme | A request-review process |
| Inbox · node | Tray illustration | Requests arrive here |
| Triage · node | Branching line illustration | A reviewer classifies the request |
| Approve · node | Approval mark illustration | Owner approval is required |
| Publish · node | Outgoing page illustration | Only after approval |
Freeze the copy before requesting the image
Use one theme and three to seven short labels; English is the package default. Request another output language explicitly if needed. Choose a process, comparison, hierarchy, causal chain or feedback loop and assign one drawn metaphor per node. Read references/prompt-framework.md and style-grammar.md. Keep a pure white background, thin rounded black lines, restrained flat accents and clear space around connectors.
Use an interactive system map when nodes need live exploration
Separate preparation from a billable generation
Install the complete package and review SKILL.md. These help commands run locally and do not submit an image task. The provider runner submits a new task only with --confirm-paid; do not add that flag until the user approves the actual service and cost. A host image tool can have its own charges or quota too. This guide does not require a paid generation to practice source planning or inspect an existing image.
- When a generation is authorized, keep the exact title and labels unchanged in the prompt, request one 4:3 image and record its task path. Preserve an existing task record and original inputs when resuming; a missing request ID or uncertain submission needs investigation before a new task.
node scripts/validate-output.mjs --help
node scripts/fal-image.mjs --helpReview the image before confirming checklist fields
Open the actual image at full size. Compare every visible word with the approved copy, trace every connector and verify each relationship against the source. Check illustration hierarchy, line weight, spacing, opaque white background and the absence of extra claims or decorative glyphs. Read references/quality-checklist.md and only mark a field true after doing that inspection.
After reviewing visual-note.png and writing the real checklist to visual-note-review.json in your installed package directory, run the following. The command checks supported image dimensions, three to seven declared labels and required true fields. It does not read words from pixels or determine whether your true flags are honest.
node scripts/validate-output.mjs --image visual-note.png --review visual-note-review.jsonDoes a passing validator prove the picture is correct?
No. An offline negative control passed a blank 400×300 PNG when given invented labels and all review flags set to true. A 400×225 image was correctly rejected, and an unconfirmed verifiedText flag was rejected. This demonstrates the boundary: the tool checks dimensions and declared review completeness, not OCR, source truth or visual quality. Never use --skip-review as final approval; it checks media dimensions only.
Deliver the actual image with its source and review
Keep the raster image, approved source, exact copy, prompt and completed review together. Add task records if a provider was used. Summarize the source meaning in one sentence and name unresolved wording or visual issues. The checklist JSON does not make a raster illustration editable. A correction through an external image service is another task, not a free automatic retry.