"Want to figure out what a good prompt might be to create new images like an existing one? The CLIP Interrogator is here to get you answers!\n",
"This experimental version of CLIP Interrogator supports finding good \"negative\" prompts for Stable Diffusion 2. Note this is very *WIP* and more work needs to be done building out the dataset to support this (and perhaps a reverse BLIP) so for many images it may struggle to find a well aligned negative prompt. Alignments are displayed to help see how well it did.\n",
"\n",
"<br>\n",
"\n",
"For Stable Diffusion 1.X choose the **ViT-L** model and for Stable Diffusion 2.0+ choose the **ViT-H** CLIP Model.\n",
"\n",
"This version is specialized for producing nice prompts for use with Stable Diffusion and achieves higher alignment between generated text prompt and source image. You can try out the old [version 1](https://colab.research.google.com/github/pharmapsychotic/clip-interrogator/blob/v1/clip_interrogator.ipynb) to see how different CLIP models ranks terms. \n",
"\n",
"You can also run this on HuggingFace and Replicate<br>\n",
"If this notebook is helpful to you please consider buying me a coffee via [ko-fi](https://ko-fi.com/pharmapsychotic) or following me on [twitter](https://twitter.com/pharmapsychotic) for more cool Ai stuff. 🙂\n",
"\n",
"And if you're looking for more Ai art tools check out my [Ai generative art tools list](https://pharmapsychotic.com/tools.html).\n"
"For Stable Diffusion 1.X choose the **ViT-L** model and for Stable Diffusion 2.0+ choose the **ViT-H** CLIP Model.\n"