Y7 Prompt Enhancer (Flux2)

Prompt Enhancer (Flux2)<div style="font-size: 0.8em"><div id="Y7_shortdesc" style="margin-bottom: 15px;">Takes a basic prompt and enhances (upsamples) it for Flux.2</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">ComfyUI's Flux.2 workflows do not implement any prompt upsampling that is seen and used in some of Black Forest Lab demos, so what you write is what you get. This uses the Qwen3-8B model to enhance prompts with support for thinking mode.</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Inputs:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">prompt</code>: Your basic text prompt to enhance</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">llm_name</code>: LLM model to use (currently only Qwen3-8B, more to be added)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">quantization</code>: Model precision. Lower precision saves memory and speeds up inference at the cost of quality:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 40px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">none</code> = ~16GB, <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">8bit</code> = ~8GB, <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">4bit</code> = ~4GB</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">enable_thinking</code>: Enable Qwen3's thinking mode to show the model's reasoning process</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">max_new_tokens</code>: Maximum tokens to generate (256-40960, default 1024). Model context is 40960 total including input</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">temperature</code>: Controls randomness (0.1-2.0, default 0.7). Higher values produce more diverse outputs</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">top_p</code>: Nucleus sampling parameter (0.8-1.0, default 0.9). Limits tokens to the most probable ones</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">top_k</code>: Limits token selection (22-100, default 50). Restricts to the k most likely next tokens</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">seed</code>: Random seed for reproducible results</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Outputs:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">thinking_output</code>: The model's reasoning process (only populated if <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">enable_thinking</code> is True)</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 20px; font-size: 12px;">- <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">enhanced_prompt</code>: The enhanced prompt output</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">LLM Model:</div></div><div style="font-size: 0.8em"><div style="margin-bottom: 8px; margin-left: 0px; font-size: 12px;">Note: First-time use will download the model (~16GB) if it does not exist. The model uses <code style="border: 1px solid #666; border-radius: 3px; padding: 0px 1px; font-family: monospace; display: inline-block;">trust_remote_code=True</code> for custom model code.</div></div>