Greetings fellow latent space explorers! I present to you a revolutionary negative embedding straight from my laboratory. With this free tool, you will be able to imbue your images with improved cohesion, detail, and photorealism.
This embedding has been rigorously trained on upscaled "768x768" versions of the seed images used in OpenAI's renowned COCO CLIP R-Precision evaluations, included with the release of OpenAI's "Point E" Img-2-3D script.
I have personally found that the DIMM sampler yields the most impressive results with this negative embedding. Do not hesitate, try it out for yourself and witness the transformative power of cutting-edge science!
Examples are of images with and without the negative embedding. The gird image displays results when using as a positive prompt embedding.
Description
Do not miss out on this opportunity to enhance your image processing capabilities with the base version of our powerful negative embedding. While it was originally designed to enhance the performance of OpenAI's Point-E, it has proven to be much more versatile in its abilities. Trust me, you will not be disappointed.
FAQ
Comments (10)
amazes me that the prompt for the flowery painting is just "flower"
I remember the original days when such a prompt would result in horror images or something a baby would paint :)
Any chance of making a 1.5 version? This looks really interesting but most of my preferred models are from 1.5.
Textual inversion embeddings skipped(1): DrD_PNTE768
How fix this?
Make sure you have SD v2.1 768 model loaded.
Yeap! It is =))
Not sure, wish I could assist but without more information not sure I can.
Hi. Where can I find SD v2.1 768 ?
@BouncAI Here: https://huggingface.co/stabilityai/stable-diffusion-2-1
What does this do exaclty? Sorry did not get it
This of it like a coherency and quality embedding that for the most part will help with 2.1 results.
Use in the negative prompt field or with a negative strength for the effect.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.