Use dtype= instead of deprecated torch_dtype

Simplify controls: use model generation defaults, keep reasoning strength only

Blog alignment: reasoning_strength template kwarg, video input support

Show reasoning in collapsible block; simplify header nav

Annotate chat return type so api outputs are inferred; filter reasoning preamble

Serve homepage as HTMLResponse

Convert to gradio.Server backend + custom HTML/JS frontend

Switch to ZeroGPU xlarge (96GB) with 240s duration

Stream output tokens via TextIteratorStreamer

Fix examples format for additional_inputs (list of lists)

Switch to multimodal gr.ChatInterface with ZeroGPU large

Add Gradio app: HF-style logo maker + ZeroGPU multimodal chat