Using the qwen3VL model to reverse engineer NSFW image prompts
It's very simple to use; you just need to upload a image.
Online workflow experience address:https://www.runninghub.ai/ai-detail/1984095416230805506/?inviteCode=rh-v1058
It's recommended to try the workflow online before downloading it, completely free, and ready to use as soon as you open it. It includes detailed instructions on how to use this workflow! New users receive 1000 coins upon registration, and 100 coins daily for logging in. The standard RTX4090 brings faster performance, with a vast number of high-quality workflows already published on the platform that can be run with one click. Fast, no overheating, and reduces the hassle of local deployment!
Description
FAQ
Comments (7)
Instructions for using the abliterated model would be helpful; I downloaded it, but the “QwenVL (Advanced)” node does not recognize it. The API on runninghub only generates random images.
You need to add the model to the hf_nodes.json inside the custom_nodes/Comfyui-Qwen3VL/
Add:
"huihui-aiHuihui-Qwen3-VL-8B-Instruct-abliterated": {
"repo_id": "huihui-ai/huihui-aiHuihui-Qwen3-VL-8B-Instruct-abliterated",
"default": true,
"quantized": false,
"vram_requirement": {
"full": 12.0,
"8bit": 7.0,
"4bit": 4.5
}
},
to the hf_models.json (i put it just onder the line "hf_vl_models": { )
Restart comfyUI and you can select/load the abbliterated model.
Now to figure out how to get english instead of chinese as output :)
*Edit: Behind the QwenVL node is a text prompt for custom prompt description, just add the line : text must be in english, and it outputs an english prompt
doesnt work for me
it does not recognize the abliterated..
@TheLastRemain Thanks, it worked, just a small correction: the exact link to the repo is
"repo_id": "huihui-ai/Huihui-Qwen3-VL-8B-Instruct-abliterated"
because letter case matters :)
The runninghub link is broken.