What problem does it solve? AI image models cannot proofread the text they draw, so generated posters, banners, thumbnails, and logos often contain misspelled words, broken diacritics, or garbled line breaks. This Skill eliminates text corruption by choosing between a deterministic overlay approach and a tightly constrained bake-in approach. ## Core Features & Use Cases - Reserve → Overlay mode: Prompts the model to render a flat, text-free region, then overlays real vector text with fonts like Be Vietnam Pro using the Text Overlay tool or the text_overlay workflow node. - Safe bake-in mode: Constrains prompts to short, uppercase, accent-free strings (≤3 words, one line) when overlay is not possible, with multi-variant generation and selection. - Verification workflow: Compares rendered text against the target string using image-qa / TextIntegrity checks instead of trusting visual inspection. - Use Case: You need a Vietnamese sale banner with accented text. Use Mode A to generate the banner background with a reserved blank area, overlay the exact copy, and verify it — changing the text later only requires re-rendering the overlay, not regenerating the image. ## Quick Start Ask the assistant to create a poster with correctly spelled Vietnamese text using the reserve-and-overlay method and verify the result against the target string.
