Workflow for 2511 version (v18)?
Same workflow as always.
The workflow you mention is this one https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/blob/main/Qwen-Rapid-AIO.json? (I'm new)
Thanks, I just noticed the official 2511 model workflow had a new node; Edit Model Reference Method
Yeah, I've seen it too... not sure what it does. I haven't tried it, but so far the model seems to work fine without it, so I've been skipping it for the meantime. I did briefly try the AuraSampling whatever and that didn't seem to help, so I haven't been adventurous.
@Phr00t It should help with character consistency. A node between positive and negative prompt and ksampler. I do use AuraSampling, I'm getting great results, can't complain with my current workflow.
@Phr00t It should help with character consistency. A node between positive and negative prompt and ksampler. I do use AuraSampling, I'm getting great results, can't complain with my current workflow.
Original image:
Prompt:
Her hair is platinum blonde and she's at a rooftop of a penthouse in the evening, keep exact same hairstyle, face, clothes.
bro can you please share your workflow in .json format or embedded in an image? I would like to try it.
@Phr00t It should help with character consistency. A node between positive and negative prompt and ksampler. I do use AuraSampling, I'm getting great results, can't complain with my current workflow.
Wow, stunning result! If you don't mind, would you be kind enough to share the workflow? I'd really appreciate it. Thanks!
In my very brief testing, using ModelSamplingAuraFlow of 3.1 + CFGNorm either made insignificant variations or worse anatomy. The "edit model reference" nodes (which are actually renamed "FluxKontextMultiReferenceLatentMethod" nodes, making them hard to find) did absolutely nothing -- same output. For now, I'm sticking with the default workflow without these extra nodes. It might make a difference when using the "euler/simple" sampler configuration, but I find "euler_a/beta" better.
v18 NSFW seems nice so far. I tried those extra nodes (FluxKontextMultiReferenceLatentMethod) with same seed etc and they seemed to make zero difference so I'm not going to use them either. Also using euler_a/beta.