Qwen Image Edit All-In-One full checkpoint: U-Net + CLIP + VAE. Merged with lightning 4-step LoRA, and some magic added too.
Simple sample workflow is on "Workflow" page, marked as "Training data". Use Euler sampler and simple or normal scheduler, 4 steps, CFG 1.
32Gb RAM is enough, but don't forget to add more virtual memory in your system (enlarge swap file to at least 64Gb).
Feel free to use Qwen Image Edit LoRAs. Please share your creations here.
Enjoy.
Description about NSFW for those, who are still here and reading this.
Not everyone uses models to generate NSFW content. I don't see the point in integrating NSFW LoRA into the model, otherwise someone might end up with boobs in the generated materials for a children's party. That's hardly what we want to do in real life, right? So this is a basic model with a standard level of censorship. Everything beyond the censorship limits is achieved through LoRA.
Description
FAQ
Comments (13)
How to use qwen Lora??
As any other LoRAs I think 🤩
I've added link to Qwen Image Edit LoRAs to model description.
Qwen Image Edit LoRAs on Civitai
Nice thing, thank you
How much vram I'll need to run this bro!
If model doesn't fit in your VRAM, then VRAM size doesn't matter, only usual RAM does.
32 Gb RAM will be enough. Don't forget to set swap file large enough.
@mistporyvaev gonna release any gguf!
@deditz111802 fp8 is not as good as fp16 but acceptable. GGUF always gives lower quality than fp8.
Which kind of bad quality you prefer: 4 bits or 2 bits quantization? 😃
@mistporyvaev 4 bits
@deditz111802 take one of GGUFs here, then apply one of LoRAs there, take corresponding CLIP, then VAE, and your super-duper generation is almost ready!
I don't like to cook it all file-by-file, that's why I prefer to compile all in one checkpoint.
@mistporyvaev bro I'm running severe gguf already. I asked for this one gguf bcz of nsfw. None of the gguf work well for nsfw and o tried multiple nsfw lora but again none of them are good
A great gift for the new year! Thanks!
enjoy it 🤩