design-assets: document GPT Image 2 div-by-16 size constraint

[goannajezweb]

Summary

• Adds a new Pixel size constraints subsection under GPT Image 2 Specifics, documenting the empirically-discovered rule that both width and height must be divisible by 16. Aspect ratio is free; the pixel grid isn't.
• Cheat-sheet of useful aspect ratios at div-16-compliant dimensions (1:1, 16:9, 1.91:1 OG, 2:1, 3:1, 2:3, 9:16).
• OG-specific recipe: generate 1792x944 natively (= 1.898:1, OG-canonical), resize to 1200x630 for serving.
• Two new rows in Common Mistakes: "Submitting odd pixel dims" and "Generating at one shape and cropping to another" (the model composes for the canvas; cropping clips intentional wordmarks / accents).

Why

Hit empirically while generating an OG card for a Jezmail newsletter edition. First attempt at native 1.91:1 returned:

HTTP 400: Invalid size '1792x936'. Width and height must both be divisible by 16.

Without this in the skill, the loop became: generate wrong shape → crop → clip the wordmark → regenerate. The doc-only fix is two short sections and saves anyone hitting this later from the same retry cycle.

Test plan

• Skill loads cleanly (no markdown syntax errors)
• Cheat-sheet sizes all satisfy (w % 16 == 0) and (h % 16 == 0)
• Recipe 1792x944 returns a valid image (verified during the OG generation that prompted this PR)

🤖 Generated with Claude Code