Instructions to use dx8152/Qwen-Image-Edit-2509-White_to_Scene with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use dx8152/Qwen-Image-Edit-2509-White_to_Scene with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import load_image # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("dx8152/Qwen-Image-Edit-2509-White_to_Scene", dtype=torch.bfloat16, device_map="cuda") prompt = "Turn this cat into a dog" input_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png") image = pipe(image=input_image, prompt=prompt).images[0] - Notebooks
- Google Colab
- Kaggle
can you please make it for FLUX klein !
my dear genius friend can you make this amazing lora for Flux klein ?
my dear genius friend can you make this amazing lora for Flux klein ?
Okay, I'll do it when I have time.
Flux2 Klein 9B can sort of do this by default and not just with white color backgrounds either - you can tell it what colors to use for replacement in hex or rgb values - did quite a bit of tests with all kinds of pure value (like R 255 or G 255 which are most useful cause they are not encountered in real photos or generations all that often compared to white) color backgrounds and it works rather well - it can also remove backgrounds (very promptable for what to keep and what to remove too) and put stuff on the specified color backgrounds pretty well too . Ofc it might benefit from some additional training in both of those functionalities.
Flux2 Klein 9B can sort of do this by default and not just with white color backgrounds either - you can tell it what colors to use for replacement in hex or rgb values - did quite a bit of tests with all kinds of pure value (like R 255 or G 255 which are most useful cause they are not encountered in real photos or generations all that often compared to white) color backgrounds and it works rather well - it can also remove backgrounds (very promptable for what to keep and what to remove too) and put stuff on the specified color backgrounds pretty well too . Ofc it might benefit from some additional training in both of those functionalities.
there is big difference in this lora and only write a prompt .. the prompt lots of time missing up with scene object ..like offset or const of object .. and also changing sometime the whole object even if you write strict prompt .. but with this lora if you try it out on qwen , its actually never miss !! and always keep the main obj lighting same .. which is somthing you can't do accuratly in klein .. i mean it's posible but you will need to keep playing with seeds and hit run multi times .. i realy hope he do this lora for klein !!!!!!!!!!!!!! please
This lora is awesome in Qwen, not arguing that at all ) and maybe same kind of training can benefit flux2 Klein 9B as well, for you are right about prompting part - flux2 is quite finicky in that regard by default, but still, it does work quite well out of the box more often than not from what i seen, if prompted correctly that is.
What i would really wish ported to Flux2 Klein 9b is the awesome Fusion lora - that's what is not that easy to do in klein by default - doable but a lot less predictable than using Qwen + Fusion lora.
Klein also needs such high-quality Lora.
any news on this please ? :(
any news on this please ? :(
Klein itself has this capability; it only needs to be paired with Consistency LoRa to perform the task well. (https://huggingface.co/dx8152/Flux2-Klein-9B-Consistency)
any news on this please ? :(
Klein itself has this capability; it only needs to be paired with Consistency LoRa to perform the task well. (https://huggingface.co/dx8152/Flux2-Klein-9B-Consistency)
the thing is kelin doesn't keep same light or small/micro details like fabrics, in your image edit lora it keep the model typicaly as the reference ...
in klein i used your amazing consistency lora but it doesn't lock the reference model in final result, when i try to keep all details same from prompt it still not working as it ment to be in image edit.
i mean in qwen image white scene lora, my model was always come with it's light and details like the lora only changed the outside part and kept the reference untouch .. i hope you can understand me. :(
any news on this please ? :(
Klein itself has this capability; it only needs to be paired with Consistency LoRa to perform the task well. (https://huggingface.co/dx8152/Flux2-Klein-9B-Consistency)
the thing is kelin doesn't keep same light or small/micro details like fabrics, in your image edit lora it keep the model typicaly as the reference ...
in klein i used your amazing consistency lora but it doesn't lock the reference model in final result, when i try to keep all details same from prompt it still not working as it ment to be in image edit.i mean in qwen image white scene lora, my model was always come with it's light and details like the lora only changed the outside part and kept the reference untouch .. i hope you can understand me. :(
I'm not sure if I can implement this feature well; I've been quite busy lately and I'm unsure if I'll have the time to do it.