Nothing to hand? — a short essay on AI agents at work, planned in the Signal Editorial style. The example replays a saved run, free.
Paste the article, pick a style
Each style is a complete visual system — palette, composition grammar, metaphor method and QA rubric — from the open-source zyncli-template library.
The plan chooses the moments
Only passages that genuinely benefit from an image are illustrated — core argument, contrasts, turning points, systems, the conclusion. Filler is skipped, and the plan says why.
Get the images, or take the prompts anywhere
Leave rendering on and every shot is generated right here in the same run on GPT Image. Or copy the standalone prompts (exact palette, full constraints) into any image model, with one-click variants for Midjourney (--ar 16:9 --no …) and for generators that take the negative prompt separately. Labels are honesty-checked against your article. In-app renders are 1024×1024 previews — the prompts themselves target 16:9 for your final generator.
Download your article with the images already placed
Download article + images (.md) gives you your own article back with an image reference at each shot’s planned position, alt text written from the shot’s idea, and filenames that match the PNGs each shot downloads — paste it into Ghost, Substack or an MDX file. The plan itself still exports as Markdown or JSON.
Do I need an image model account?
No. Rendering happens inside the app on your SkillSafe credits with GPT Image — about 25¢ is reserved per image and you pay what the render actually costs. You can also turn rendering off and take the prompts to any generator you already pay for; every prompt is standalone and carries the style’s exact palette hex codes.
Can I run it without signing in?
You can browse all 25 style guides and replay the bundled example plan for free, signed out. The model run itself is metered, and guest sessions carry no balance, so the app says so at the run button and offers to sign you in rather than failing after you press the button.
How does it decide how many images an article needs?
Under “let the plan decide” the model picks between one and nine, illustrating only moments that genuinely benefit — the core argument, a contrast, a turning point, a system relationship, the conclusion — and it lists what it deliberately skipped, with reasons. You can also fix the count exactly.
What stops it inventing text inside the artwork?
Two things. The prompts forbid invented statistics, quotations, dates and interface copy, and every label a shot proposes is checked in your browser against your article: if the words do not appear verbatim, the plan is flagged before anything is rendered, and automatic rendering is held back until you have read the flag.
Are my plans saved?
Yes. Every finished plan is kept in this browser and, once you sign in, on your account, so a plan made on one device opens on another. Reopening a saved plan is free — it never re-runs the model. Rendered images are held in memory for the session only.