Brutally Honest #209

Matej Lancaric's April 20, 2026 newsletter issue, aimed at mobile user acquisition and creative professionals, centers on a practical assessment of Google's Nano Banana 2 (Gemini 3.1 Flash Image) for localizing ad creatives. Its thesis is that the model makes in-image translation production-viable for the top 10–15 languages, but does not eliminate the need for localization QA, since errors are now subtler and surface only when users notice them.

The tip section separates what ships from what fails. Latin-script languages (Spanish, French, German, Portuguese, Italian) with short taglines need near-zero retouching. Simplified Chinese and Japanese headlines work for short text, with English accuracy cited at 94–97%. Korean works for short Hangul. Traditional Chinese drops to roughly 78% accuracy, Arabic, Hebrew, Thai and Hindi remain rough, and custom brand fonts are not reproduced. The recommended workflow is to master in English, batch markets by script family, feed in pre-translated copy (said to boost accuracy by 10–15%), generate 4–6 variants per locale, have a native speaker check the output, and composite text separately for high-stakes assets like App Store banners. A 10-market rollout with 5 variants at 2K costs about 300 credits.

Other sections cover Township's creative evolution from pull-the-pin and hyper-casual stacking mechanics to a 90% playable-ad output with five-to-ten-minute sessions, and a 2026 "Kingshot" style format. Creative volume reportedly exceeded 7,000 in one six-month window. A podcast teaser on 120 AI-generated creatives concludes that AI accelerates creative strategy rather than replacing it, shifting teams from ideas to reusable creative systems, with one example reaching 320% D7 ROAS. The issue also announces a relaunched 2026 UA survey, PlayableMaker template updates, sponsored FastSpring D2C content, and a brief item on Anthropic's White House meeting.