Phr00t's picture
Update fixed-textencode-node/README.md
5a2a90c verified

Replace your nodes_qwen.py file with one of these, which will improve your "native" TextEncodeQwenEditPlus node. You should be able to find the nodes_qwen.py file in your ComfyUI/comfy_extras/ folder.

  • It increases the number of image inputs from 3 to 4.
  • It slightly tweaks the system input to be a bit more versatile.
  • It fixes an issue where the image would get cut off and mirror itself sometimes by cropping better.
  • Addresses unexpected zoom / cropping problems

The "original" allows you to input a "target_size" which will scale the input images to better match your output size (which can fix zooming issues). Usually setting it to 90% of your largest image size works.

v2 takes a latent as an input, which should contain your target image size. The node will automatically scale input images to best match that target latent size. If you do not provide a latent input, it will not do any scaling.