ComfyUI Node
RT Qwen Image 2.1 Prompt Enhancer
A ComfyUI node in RareTutor/Prompt with 24 inputs and 5 outputs.
RT Qwen Image 2.1 Prompt Enhancer
- image_1
- image_2
- image_3
- image_4
- image_5
- image_6
- image_7
- image_8
- image_9
- image_10
- image
- enhanced_prompt
- json_output
- wh_ratio
- ratio_follow
- diagnostics
◄llm_modelNo supported models (.gguf or .safetensors) found in clip/text_encoders/llm►
◄vision_modelNo supported models (.gguf or .safetensors) found in clip/text_encoders/llm►
◄user_input►
◄mode01. Edit Prompt Enhancer (Image-to-Image / Multi-Image)►
◄thinking_modeDisabled (Fast / Direct JSON - Recommended)►
◄image_resolution1.0 Megapixel (1008x1008 - Ultra Detail, Recommended)►
◄max_tokens4096►
◄creativity0.8 - Balanced (Recommended)►
◄seed-1►
◄n_ctx8192►
◄n_gpu_layers-1►
◄keep_model_loadedtrue►
◄debug_consoletrue►
CategoryRareTutor/Prompt
Inputs (24)
| Name | Type | Default | Description |
|---|---|---|---|
| llm_model | COMBO | No supported models (.gguf or .safetensors) found in clip/text_encoders/llm | Select a GGUF text model or multimodal base model. |
| vision_model | COMBO | No supported models (.gguf or .safetensors) found in clip/text_encoders/llm | Select a GGUF vision projector / mmproj model (required when input images are used). |
| user_input | STRING | — | |
| mode | COMBO | 01. Edit Prompt Enhancer (Image-to-Image / Multi-Image) | 01: Rewrites image editing instructions with strict attribute disentanglement. 02: Expands short text into rich T2I prompts. 03: Generates a style transfer instruction from image1 (content) + image2 (style reference) - no text input needed. |
| thinking_mode | COMBO | Disabled (Fast / Direct JSON - Recommended) | Disable internal <think> reasoning for instant generation (3-5 seconds) and to avoid token budget truncation. Enable only if you want full chain-of-thought analysis. |
| image_resolution | COMBO | 1.0 Megapixel (1008x1008 - Ultra Detail, Recommended) | Visual resolution for reference images. 1.0 MP (1008x1008) provides ~1296 tokens per image, preserving small text, facial features, and subtle textures. 28-pixel patch alignment is strictly enforced. |
| max_tokens | COMBO | 4096 | Maximum output tokens to generate. Default 4096 provides ample headroom for reasoning models. |
| creativity | COMBO | 0.8 - Balanced (Recommended) | Sampling temperature. Lower is more literal; higher is more elaborative. |
| seed | INT | -1-1–18446744073709550000 | Random seed for reproducible generation (-1 for random). |
| n_ctx | INT | 81922048–32768 | Context window size in tokens. |
| n_gpu_layers | INT | -1-1–128 | Number of model layers to offload to GPU VRAM (-1 = all layers to GPU, 0 = CPU only). |
| keep_model_loaded | BOOLEAN | true | Keep model loaded in VRAM across workflow runs to avoid reloading latency. |
| debug_console | BOOLEAN | true | Print detailed progress and diagnostic metrics to the terminal console. |
| image_1opt | IMAGE | Reference image 1 (<image1>). Primary canvas, target scene, or main subject. | |
| image_2opt | IMAGE | Reference image 2 (<image2>). Secondary reference, style, face, clothing, or background. | |
| image_3opt | IMAGE | Reference image 3 (<image3>). | |
| image_4opt | IMAGE | Reference image 4 (<image4>). | |
| image_5opt | IMAGE | Reference image 5 (<image5>). | |
| image_6opt | IMAGE | Reference image 6 (<image6>). | |
| image_7opt | IMAGE | Reference image 7 (<image7>). | |
| image_8opt | IMAGE | Reference image 8 (<image8>). | |
| image_9opt | IMAGE | Reference image 9 (<image9>). | |
| image_10opt | IMAGE | Reference image 10 (<image10>). | |
| imageopt | IMAGE | Alternative single image or batched image input. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| enhanced_prompt | STRING | — |
| json_output | STRING | — |
| wh_ratio | STRING | — |
| ratio_follow | STRING | — |
| diagnostics | STRING | — |