ComfyUI Node
DyPE for Qwen Image
A ComfyUI node in model_patches/unet with 13 inputs and 1 output.
DyPE for Qwen Image
- model
- model
◄width1024►
◄height1024►
◄auto_detecttrue►
◄base_width1024►
◄base_height1024►
◄methodyarn►
◄enable_dypetrue►
◄dype_exponent2.0►
◄base_shift1.15►
◄max_shift1.35►
◄editing_strength1.00►
◄editing_modeadaptive►
Categorymodel_patches/unet
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | Qwen Image model to patch with DyPE. | |
| width | INT | 102416–16384 | Target output width in pixels. |
| height | INT | 102416–16384 | Target output height in pixels. |
| auto_detect | BOOLEAN | true | Automatically derive patch size and base resolution from the model when possible. |
| base_width | INT | 102416–16384 | Training width used by the base Qwen model when auto detection fails. |
| base_height | INT | 102416–16384 | Training height used by the base Qwen model when auto detection fails. |
| method | COMBO | yarn | Spatial RoPE extrapolation strategy. |
| enable_dype | BOOLEAN | true | Enable Dynamic Position Extrapolation scaling. |
| dype_exponent | FLOAT | 2.00–4 | Controls how strongly DyPE ramps across sampling timesteps. |
| base_shift | FLOAT | 1.150–10 | Baseline shift applied to the flow-matching noise schedule. |
| max_shift | FLOAT | 1.350–10 | Maximum shift applied when operating at the target resolution. |
| editing_strength | FLOAT | 1.000–1 | Scale DyPE while editing images (1.0 = full strength, 0.0 = disable DyPE scaling in edits). |
| editing_mode | COMBO | adaptive | Strategy for tapering DyPE during edits. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |