Qwen AIO giving fully black generations need help
ok this might sound simple for a lot of you but i have an issue i just decided to start using Qwen image edit first downloaded 14.1 aio version all the generations came out completely black so i decided to test it with the official 2509 version same results
just also downloaded 5.3 which is praised as the best AIO version same results i asked gemini and chatgpt no working solution came out of them so im out of ideas does anyone knows what the actual solution is?
I've seen this pop up a few times, and it might be related to your PyTorch version or using Sage Attention. Make sure that is all updated to the latest (2.9.1 PyTorch, cu130, Python 3.13, latest ComfyUI etc.). Looks like you did replace your nodes_qwen.py with mine (which is a good thing). Also try different samplers, I haven't been recommending sa_solver for awhile (although I don't think this will fix the black image). Also make sure you have the latest NVIDIA studio drivers.
Remove --fast or --force-fp16 flags and dont use sage attention.
I've seen this pop up a few times, and it might be related to your PyTorch version or using Sage Attention. Make sure that is all updated to the latest (2.9.1 PyTorch, cu130, Python 3.13, latest ComfyUI etc.). Looks like you did replace your nodes_qwen.py with mine (which is a good thing). Also try different samplers, I haven't been recommending sa_solver for awhile (although I don't think this will fix the black image). Also make sure you have the latest NVIDIA studio drivers.
Would like to add to this , Python 3.13 can still even to this day conflict with somethings in comfyui (it's rare) in current version but the official word / response from contributor devs has been said to fall back to "Python 3.12" venv to seperate from your systemwide python 3.13 and test , speaking from personal exp - I had issues with Qwen and Wan2.2 with python 3.13 - only python 3.12 fixed - your mileage may vary but absolutely try isolation Venv environment python 3.12 for comfyui , not only will that seperate from the systemwide python and you'll have less conflict - it's also an increase in security by being in venv environment :).
source: https://docs.comfy.org/installation/system_requirements
Python Version Recommended Python 3.12
Supports Python 3.13 (some custom nodes may not be compatible)
source: https://github.com/comfyanonymous/ComfyUI
Python 3.14 works but you may encounter issues with the torch compile node. The free threaded variant is still missing some dependencies.
Python 3.13 is very well supported. If you have trouble with some custom node dependencies on 3.13 you can try 3.12
I've seen this pop up a few times, and it might be related to your PyTorch version or using Sage Attention. Make sure that is all updated to the latest (2.9.1 PyTorch, cu130, Python 3.13, latest ComfyUI etc.). Looks like you did replace your nodes_qwen.py with mine (which is a good thing). Also try different samplers, I haven't been recommending sa_solver for awhile (although I don't think this will fix the black image). Also make sure you have the latest NVIDIA studio drivers.
Would like to add to this , Python 3.13 can still even to this day conflict with somethings in comfyui (it's rare) in current version but the official word / response from contributor devs has been said to fall back to "Python 3.12" venv to seperate from your systemwide python 3.13 and test , speaking from personal exp - I had issues with Qwen and Wan2.2 with python 3.13 - only python 3.12 fixed - your mileage may vary but absolutely try isolation Venv environment python 3.12 for comfyui , not only will that seperate from the systemwide python and you'll have less conflict - it's also an increase in security by being in venv environment :).
I had the same experience. The solution was falling back to Python 3.12 then I could use Sage Attention with KJ's patch sage attention node. No --use-sage-attention for the launch. And lots of trial and error.