Large-area and free-standing photonic crystal (PC) polymer films exhibit highly saturated iridescence and robust structural colors, making them promising for applications in the field of display, anti-counterfeiting, and camouflage. However, their practical utilization has been hindered by challenges in achieving both vivid coloration and reusability. Here, we design a sandwich-structured PC film that simultaneously addresses issues of color appearance and reusability by combining colloid evaporative self-assembly on porous substrate and knife coating of polymers. The unique “sandwich” structure, comprising a self-assembled PC intermediate layer and protective polymer encapsulation, demonstrates a great synergistic effect (“1 + 1 > 2”), including unprecedented color fastness stability (affordable for 100 times dye/wet fastness), bright iridescent color, and certain flexibility and reusability. In addition, by replacing the bottom polymer with a double-sided adhesive, a flexible PC sticker can be further obtained, broadening its range of applications to surfaces of different materials. This strategy opens a new avenue for constructing functionalized iridescent PC-polymer films.

The combination of high-precision structural color and luminescence was an effective way to improve encoding capacity. Hence, we synthesized ZnS@QDs (QDs = quantum dots) microparticles exhibiting synchronized size-dependent Mie scattering and tunable luminescence in submicron size. The loading of QDs onto ZnS microspheres not only enables uniform dispersion to prevent quenching but also enhances luminescence through Mie scattering. Macroscopic structural color patterns were fabricated by direct-ink-writing with ZnS microspheres suspension as ink, while the dots formed by ZnS@QDs microspheres were inserted randomly. Meanwhile, the stochastic distribution of microparticles during ink evaporation generated physically unclonable function (PUF) codes with a capacity up to 2²⁵⁰⁰ within areas of 50 μm × 50 μm. ResNet-based few-shot machine learning achieved 100% authentication accuracy with a minimal training set (21 samples). At the same time, open large language models (DeepSeek) enable rapid consistency assessments. In addition, the labels demonstrate long-term stability (> 6 months), customizable colors, and compatibility with flexible substrates, offering potential applications.