20251104 update QwenEditUtils 2.2
add Mask option to modify the specific area
https://www.runninghub.ai/post/1985595549766365186?inviteCode=rh-v1279
20251104 update v5.1
add QwenEditAdaptiveLongestEdge to easier adjust longest edge for different image
RH workflow:
QwenEditUtils 2.0 Any Resolution Reference
Experience link:
https://www.runninghub.ai/post/1985595549766365186?inviteCode=rh-v1279
Custom_node:
https://github.com/lrzjason/Comfyui-QwenEditUtils
20251104 update v5.0
20250927 update v4.0
Use with lora https://civarchive.com/models/1939453 v2 for QwenEditPlus Consistence Editing
20250924 update v3.0
Updated plugin for QwenEditPlus
Removed old nodes. The new node should capable old function when disable vl resize.
20250915 update v2.0
Completely solve pixel shift issue.
Use with Consistance Edit Lora, strength: 1 https://civarchive.com/models/1939453/qwenedit-consistance-edit-lora
Use plugin in https://github.com/lrzjason/Comfyui-QwenEditUtils
RH workflow:
QwenEdit Consistance Edit workflow v2.0
Experience link: https://www.runninghub.ai/post/1965033238311915522/?inviteCode=rh-v1279
Kontext like workflow.
Avoid to encode ref image to qwenvl.
Use with Consistance Edit Lora, strength: 0.48 https://civarchive.com/models/1939453/qwenedit-consistance-edit-lora
RH workflow:
It's extremely fun and I highly recommend it to you!
Workflow: QwenEdit Consistance Edit workflow
Experience link: https://www.runninghub.ai/post/1965033238311915522/?inviteCode=rh-v1279
Description
FAQ
Comments (14)
Can you add the workflow with the same example images and the same settings to get the same high res outputs? Bc i only get 1024 outputs without details, would help, thanks.
You could just set the ref longest edge to any size to get high res result. But it depends on your hardware and take times to run. 2K result usually take 5 minutes using 8 steps lora
@xiaozhijason Yes but what are the optimal settings to get the max amount of details from second refeference (tshirt in this case) should i max the second image longest edge? increase the vl target size? A tutorial explaining this would be great. Dou you mind share the settings that you used in the example photos?
@alucardx651 I have one but it is in Chinese. The vl size always keep 384 is ok. Just increase the ref longest edge is ok.
@xiaozhijason But cant you do a screenshoot of the workflow that you used "with the same settings" in the image to get this results and upload to imgur for example to share here? Or upload the complete workflow with example images into a zip to some file host and link it here.
in the new version there are 2 workflows. what is the different between the workflows?
full included crop with pad info for crop method pad. simple doesn't
Even with increasing the scale to longest side, its still low quality :(.
That the model limitations. With finetuned lora on specific task, like tryon. It could perform better
@xiaozhijason What lora do you reccomend I havent been able to find a good one.
@xiaozhijason Like the inpaint crop and stitch works well, but when it stitches it makes the final image really wonky. But it produces high quality results but sucks for the final output.
Keep getting huge heads on tiny bodies...
You might adjust parameters to fit your images and usage.
If you have to mask at all the purpose of the model isn't being fulfilled. Go back to kontext inpaint.