Would this be a good model for prompt generation for Image-Image and Image-Video workflows?

#4
by Tomcat2048 - opened

I've been using Qwen3-VL abliterated model for prompt engineering for Image-Image and Image-Video, is this model better at that sort of thing?

I ended up taking a look, this model is like an abliterated one but better - it's just fancier automated abliteration that retains more quality and doesn't damage it much. You can think of this as a drop-in replacement!

By the way there is qwen_3_4b-hereticV2-zimage (Z-Image-Turbo uses Qwen3 4B for reading your prompt) but I didn't notice a difference.

I ended up taking a look, this model is like an abliterated one but better - it's just fancier automated abliteration that retains more quality and doesn't damage it much. You can think of this as a drop-in replacement!

By the way there is qwen_3_4b-hereticV2-zimage (Z-Image-Turbo uses Qwen3 4B for reading your prompt) but I didn't notice a difference.

It looks like this model doesn't support Vision (at least in LM Studio). Text only...

Hi. For LM Studio, the mmproj file is needed for Vision mode. Can you recommend a suitable option? Without this file, the model only works in Text generation mode.

Update
Who wants Vision mode: find and download mmproj-Qwen3-VL-32b-Instruct-Heretic-F16.gguf Put this file into folder near model and restart LM Studio. It works for me.

Sign up or log in to comment