Qwen3-VL conditioner
An embedding service: sends back the conditioning tensors (prompt_embeds, token tags and the resolved
plan) that a downstream video generation Space consumes. Call it over the gradio API; the UI below is
for inspection only.
rewrite_prompt is optional and off by default. Turned on, the conditioner first writes the structured
prompt format the video model was trained on — task instruction, shot-by-shot timeline, soundscape — and
encodes that instead; it comes back under the plan's refined_prompt.
Ready · Qwen3-VL conditioner, bfloat16, unquantized, 64 decoder layers, read at layer 50 · loaded in 180s · repo MiniMaxAI/MiniMax-H3
Canvas