Please use NEGATIVE PROMPTS with Stable Diffusion v2.0

The negative prompt can be used to describe what you don't want to see in the image. The AI sometimes ignores parts of the negative prompt, especially if the prompt and negative prompt contain very similar terms. In other cases, the negative prompt has big impact on the composition of the image. For example, if you put "shoes" into the negative prompt, you might get a barefoot person - or you get an image where the feet are out of frame.

Emad Mostaque - the model process deduped and flattened the latent space so negative prompts and weighting has a huge impact

References:

❤️ If you want to support what we are doing ❤️
Support here:

[erics]

This information just blew my mind. I've always ignored negative prompts.

[victormultanen]

first in my AI journey I hear about negative prompts from this channel

[rdm]

I just installed it tonight and noticed the rapid difference with using negative prompts after watching this. Thanks!

[tycho]

as a general rule of thumb, I'll generate an image using only positive prompts, look at the result, and type in the negative prompts everything that is wrong with the image or that I want to change about it and generate it again. But it's helpful to start with some that you know you want as well.

[scrpin]

Deduplicate and flatten the latent space is a mumbo jumbo term for data compression to save on storage space and bandwidth so that they could pack more stuff into a 4GB model. By packing more stuff, the model is pruned to be non-specific, so by using a negative prompt, you're filtering out non-specific portion of the latent space.

[iamzeid]

This is really useful! It solves a lot of 'AI Art' problems. Thank you!

[sadekf]

Thank you for the wonderful work. Please continue.

[CrhystyannDavidBernalChaves]

I have a question:
I use DreamStudio, do you know if in this page you can use negative prompts?

[davidwilliams]

I always use negative prompts in any search, but still with SD2.0 images are jank regardless of sampler. Doing backgrounds and stuff works great, but once you start getting specific with a subject and posing it really falls apart especially in the anatomy sense. Hoping overtime it gets better, really seems something related to the CLIP change causing it to be bad overall consistency wise.

[vladimir-ioanvararu]

Awesome video. Thank you! But could you help us a little? There are many web or mobile platforms that use SD but there is no special text field for negative prompts. Is there any special command to use them? For example in MidJourney we use "--no ugly faces". Thanks again 💕🙏

[ZeroCool]

Can you do a video how to update the "Dreambooth ShivamShrirao" Repo to train on SD2 using "WSL2 + Ubuntu" on windows?

[swannschilling]

Nice one!! Seems like its also a way to safe on tokens?

[Drakmour]

Do SD really recognize brackets, double brackets, square brackets and so on?

[hypersonicmonkeybrains]

If it really can do these negative prompts it means it should have the power to remove blur from input images or remove piexlation or fog or anything else.. Which would be an incredibly powerful tool.

[alonsorojas]

Very good video, its even more important now. Thanks

[BuildNewSkillsZ]

I don't know what's going on in the mobile version, but with negative prompts, all done as you said and I cannot get realistic ones

[hypersonicmonkeybrains]

I tried dreamStudio and i dont have a box that says negative prompt.

[xolaider]

cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)), ((poorly drawn)), ((extra limbs)), ((close up)), ((b&w)), weird colors, blurry

[SpaceCattttt]

Judging what's ugly and beautiful sounds way more negative to me than any prompt...

[jennyml]

maybe you can help me? how can i prompt empty eyes i want only see the wait from the eyes.