Every trained network carries a signature in its weights. This playground reads a .safetensors file and paints it as art that deterministically encodes the actual weight values — a visual fingerprint of the model. Pick from five art forms — mandala, strata, mosaic, flow, bloom — and five palettes. Same weights always give the same image; even a tiny change shifts it.
How the fingerprint is built. Each tensor becomes a 2-D matrix (1-D
biases are padded into squares; conv kernels are flattened), sorted largest-first and laid out as
concentric rings from the center. Within a ring the angle indexes columns and the radius indexes rows,
with a per-ring spiral twist; the sampled weight sets the color — hue from sign, saturation and
brightness from magnitude (tanh-compressed). The central core encodes the top singular
values of the biggest layer. The palette seed is derived from the raw weight bytes, so the image is a
reproducible function of the model.
safetensors is HuggingFace's format and the de-facto standard for sharing
weights: a small JSON header (tensor names, shapes, dtypes) followed by raw bytes. Unlike PyTorch
.pt/.pth pickles, it can't execute code on load — which is exactly why it's
safe to let a webpage open one. Supports F32 · F16 · BF16 · F64 · int types.