VERSION 2.1
use with my v2.0 qwen image edit 2511 female nudity lora
Try adding "make 8k ultra high resolution photorealistic image"
resize image node to 2.0 megapixel
stacked loras for nudity
updated prompts
VERSION 2.0
use with v1.5 of my qwen image edit 2511 female nudity lora (doesnt need the unblur lora as its baked into it)
Description
version 2.1 for nudity v2.0 lora
FAQ
Comments (3)
Good workflow, but i prefer to upscale to 1.0x and then after generating.
intersting moment- in some variants of sampler/scheduler it`s deleting text watermarks
Does changing the seed do anything? If so is there a way to randomly generate a new seed?
Oh mate, an update you might like to use is :
To resolve pixel-shift issues and prevent unintended alterations to faces or fine details, modify your node connections as follows:
Disconnect the VAE and Image inputs directly from the TextEncodeQwenImageEditPlus (both Positive and Negative) nodes.
Insert a standard VAE Encode node. Pass your source image through this to generate the initial Latent.
Add a ReferenceLatent node for both the Positive and Negative chains.
Route the Latent output from your VAE Encode into these nodes.
Connect the Conditioning from your TextEncodes into these Reference nodes.
Feed the resulting Conditioning outputs from those Reference nodes into the Edit Model Reference Method input.
This should sort out the issues with the unintended editing or the shifting pixels.