Answer the wizard → get a brand-locked prompt that makes Codex write per-slide image prompts.
1 · Brand CI
Everything downstream is locked to this. Upload the logo you actually ship with — it stays in your browser, and it is composited as a real file, never drawn by the image model.
Logo file *PNG / SVG with transparency preferred
DROP LOGO
Drop a logo, or click to browse
Used in the live preview and referenced by filename in the generated prompt.
Colour tokens *hex is authoritative — the prompt quotes these exactly
2 · Format
Canvas, deck length, and the safe area every generated image must respect.
3 · Topic & objective
One carousel, one job. If you can't name the single thing a reader should walk away with, the deck won't land.
Objective *pick one
4 · Audience
Awareness level decides slide 1. A problem-unaware reader needs a symptom; a solution-aware reader needs a differentiator.
5 · Voice
Tone chips shape the copy Codex writes for the overlays; language governs the actual slide text.
Tonepick up to 3
6 · Substance
The points become the middle slides, in this order. Proof is what stops it reading as opinion.
Key pointsone per body slide
7 · Visual direction
This is what actually goes into each image prompt. Pick the model too — the syntax differs.
Moodpick up to 3
8 · Your prompt
Paste this whole block into Codex. It asks for one image prompt per slide, the overlay copy, and a compositing spec that drops your real logo file on top.
Fill the wizard, then hit Generate.
Deck skeleton
1080 × 1350
Preview only. It shows CI, order and logo placement — not the final art.
CI check
—
How the logo survives. Image models can't draw a logo correctly — they hallucinate it. So the generated prompt splits the job: the model produces clean background art with reserved negative space, then Codex writes an HTML/CSS compositing layer that places your actual logo file and type on top. That's the only way the CI stays exact.
Everything stays in this browser. No upload, no server.