- - - - - - - - - - - - - - - - - - - - - - - - - -
š½ GET THE FREE WORKFLOWS & 1000 BONUS COINS š½
āØ Register on RunningHub through my link to get 1000 FREE RH Coins!
š° Plus, claim 100 extra coins every single day just by logging in.
1ļøā£Qwen2512 T2I Automatic expansion(N version):
https://www.runninghub.ai/ai-detail/2008075941536342017/?inviteCode=rh-v1291
š½Free download models/Loras š½
1ļøā£Download link for Qwen Image 2512 (huggingface):
https://huggingface.co/Qwen/Qwen-Image-2512
2ļøā£Download link for Qwen Image 2512 (Online storage):
https://pan.quark.cn/s/75e6e346a8e3
3ļøā£Download link for Qwen3VL NSFW Version:
https://pan.quark.cn/s/71503112e6e8
- - - - - - - - - - - - - - - - - - - - - - - - - -
Description
FAQ
Comments (24)
Where do you download and place Huihui-Qwen3-VL-8B-Instruct-abliterated for comfyui?
@BiuBoomFlowĀ Thanks, but where in my comfyui folder do i clone this?
@delta45424155Ā Directly overwrite the original model folder in the local qwen3vl directory.
@BiuBoomFlowĀ I tried overwriting the original model folder in the local qwen3vl directory as suggested, but it doesnāt work. ComfyUI still throws this error:
Prompt outputs failed validation: Qwen3_VQA: Value not in list: model 'Huihui-Qwen3-VL-8B-Instruct-abliterated' not in [...]
@FakcupĀ After overwriting the files, you'll need to restart ComfyUI, then reselect the path. If you can see the HUIHUI model as expected, it should be ready to use.
@BiuBoomFlowĀ I restarted ComfyUI after overwriting the files, but the HUIHUI model simply does not appear in the list of available models. I can't select it at all.
@BiuBoomFlowĀ just to confirm, the model files are supposed to be placed in the prompt_generator directory, right?
@FakcupĀ Place the entire model folder inside the comfyui/models directory.
@BiuBoomFlowĀ dont work
@FakcupĀ That's right, here. It's better to clone in prompt_generator. You need to go inside the folder and run the command through Powershell: git clone https://huggingface.co/huihui-ai/Huihui-Qwen3-VL-8B-Instruct-abliterated/tree/main
@bionovafood863Ā I tried that, but it still doesnāt work. Maybe Iām missing something? the model still doesnāt show up in the list
@FakcupĀ I'm not an expert at this myself. Try downloading this file. mmproj-F16.gguf. Put them in the same folder....\ComfyUI\models\prompt_generator\mmproj-F16.gguf"
....\ComfyUI\models\prompt_generator\Huihui-Qwen3-VL-8B-Instruct-abliterated" - (32.6 GB). That's it. There shouldn't be any problems. I struggled with this myself. For a long time. In the end, it worked.
@bionovafood863Ā Iāll try that when I get some time, thanks!
@bionovafood863Ā Yeah I tried that, but it didnāt change anything š
The model still doesnāt show up, so I donāt think the issue comes from that file.
@FakcupĀ I suggest moving over to ZiT or Chroma. Chroma, when properly used, has some insane prompt adherence for nsfw, porn images.
@FakcupI usually upload the log from the AI āāconsole; it reads it quickly and finds errors that I fix. Try it. It will find the error. But I wouldn't bother with this process too much. It's too long. I deleted it.
@bionovafood863Ā I was actually thinking the same thing. I installed the module to try other image models, but it takes a really long time and the results arenāt optimal
@FakcupĀ if you really want ai to write your prompts. try koboldcpp with a gguf type llm that is uncensored or qwen2.5-7b-instruct.gguf. you'll want to write a decent system prompt to help direct it to what you want.
@delta45424155Ā I'm not sure I fully understand what you're referring to. Is that a node to install, a local application, or a web tool?
@FakcupĀ
https://github.com/LostRuins/koboldcpp/releases
Then download the gguf version llm you want to use. google how to add system prompt in koboldcpp. You can run it with your cpu so your gpu is free to do comfyui
Hello, everything worked for me. But with your settings, I get a black Malevich square. I'd like to see a normal image. And 320 seconds on a 5080-16GB device is really long for a single image, almost like a video in length.
There are two settings that can improve speed. First, qwen3vlās back-training process takes quite a while; you can try using 4b or my latest qwen3.5 method, which can back-train the prompt in about 20 seconds. Second, changing the "Load Diffusion Model" mode to "fp8_e4m3fn" can significantly speed up image generation.
For "black Malevich square," you'll first need to update the ComfyUI core to the latest version. If that doesn't work, try changing the cross-attention mode to "xformers."