Moonshine Streaming Tiny on the GPU. English, up to 40 s. Upload WAV/MP3/FLAC or record from your microphone; audio is converted to 16 kHz mono in the browser.
Pocket TTS on the GPU, 24 kHz. Keep sentences under ~30 words or use punctuation; very long run-on sentences are rejected.
Text-to-image with the first served image model (e.g. Krea 2 Turbo: 512×512 ~10–15 s, 1024×1024 ~40 s on Strix Halo).
Embeddings from the first served embedding model (e.g. Qwen3-Embedding-0.6B on the XDNA NPU). Enter one sentence per line (each ≤ ~100 tokens, at most 16 lines). Shows pairwise cosine similarity.