Qwen3-TTS freezes on RTX 3060

This looks more like a runtime or memory bottleneck than a pure app bug: the heavier Qwen3-TTS paths may be too much for a consumer 12GB card, or the frontend may be waiting on a backend job that never finishes starting.

• –Qwen3-TTS includes separate variants, and the 1.7B VoiceDesign path is the most likely place for a load stall on midrange GPUs.
• –No terminal output usually means the problem is upstream of the UI, such as model download, CUDA initialization, or a backend process that never fully binds.
• –Pinokio adds another abstraction layer, so dependency mismatches can surface as a “freeze” instead of a clean error.
• –For users on RTX 3060-class hardware, the smaller CustomVoice/Base variants are the first sanity check before assuming the whole stack is broken.