TextEncodeQwenImageEditPlus节点不一样
我看到你给的示例的工作流里面的节点是这样的
我替换了nodes_qwen.py后是这样的
是我哪里没做对吗
The first images are using the standard ComfyUI nodes (which are not that great).
The second image is using the updated nodes I fixed.
The only particular problem I see is, that bottom one I suspect is your negative prompt which should be left blank.
Thank you for your reply. I'm curious why this negative prompt isn't working. For example, I want to change the pose of the original image to the pose of the reference image (by extracting the skeleton diagram), but in the generated result image, there are some extra things like bags and pants that don't exist in the original image.
Because its a well known issue that Lighning doesn't like negative prompts as cfg 1 disables the nagative. Better try: "DO NOT ADD any ...." in positive
i got a workflow that works well with poses. normally no additional prompt for poses is needed, only a very good "Maintain" prompt:
Just grabbed some random images together (simple Workflow included):
Because its a well known issue that Lighning doesn't like negative prompts as cfg 1 disables the nagative. Better try: "DO NOT ADD any ...." in positive
i got a workflow that works well with poses. normally no additional prompt for poses is needed, only a very good "Maintain" prompt:
Just grabbed some random images together (simple Workflow included):
tks,I noticed that you uses the v5 gguf model in your workflow, and I'm using the v5 integrated model. Is there a difference in the quality of the generated images between these two?
i didn't direct compared them but i guess not really, just copy pate in your loader and connect model, vae and clip.
example of pose + added part:
example of pose + added part + changed background its mirrored, more changes needs more steps 4 vs. 8 steps:
As i said, no artwork, just clicked something random as example together...