Stable Diffusion Negative Prompt for Human

Negative prompts are an important technique for controlling the image generation process in Stable Diffusion. By specifying attributes and elements that you want the AI to avoid including in the generated image, negative prompts give you more direction over the final output. Mastering negative prompts takes experimentation and practice, but can greatly improve your Stable Diffusion results.

What Are Negative Prompts?

A negative prompt tells Stable Diffusion what not to include in a generated image. For example, adding the negative prompt “ugly, deformed” signals to the AI to avoid creating images with ugly or deformed elements.

Negative prompts act as a high-dimensional anchor that the image generation process moves away from. So they influence the output by specifying attributes and elements that you want to avoid, without needing to provide additional positive instructions.

Why Use Negative Prompts?

There are several key reasons to use negative prompts with Stable Diffusion:

They give you more control over the image generation process
Help avoid strange artifacts or anatomical issues
Improve image quality and aesthetic
Override parts of the positive prompt

Negative prompts are especially useful for avoiding common AI flaws like incorrect anatomy or limbs, duplicate elements, and other strange distortions.

Examples of Negative Prompts

Here are some common negative prompts that many Stable Diffusion users rely on:

ugly, boring, bad anatomy, blurry, pixelated
trees, green, obscure, unnatural colors
poor lighting, dull, unclear
bad hands, missing fingers
extra limbs, mutated body
duplicate, watermark, signature

These prompts target attributes that the AI often struggles with, like hands and lighting. Specifying them as negatives guides the model away from those failure types.

Using Negative Prompts Effectively

Here are some tips for using negative prompts successfully:

Start simple – Add only 1-2 negative prompts initially and build up slowly
Avoid overprompting – Too many negatives can restrict the AI’s creativity
Check search results – Browse for new negative ideas from other users
Fine-tune over time – Monitor image quality and tweak prompts as needed

It also helps to generate a batch of images without any negative prompts first. Then examine the results for flaws and use negatives to address those specific issues.

Over time, you can build up an optimized set of negative prompts tailored for your style and prompt types.

Advanced Negative Prompt Techniques

In addition to manually adding negative prompts, advanced techniques like Textual Inversion can incorporate negative embeddings to further influence image generation:

Train a Textual Inversion model on a dataset of images showing attributes to avoid
Generate negative embeddings from that model
Add those vectors to your prompt with a weight/strength hyperparameter

This creates a more nuanced negative signal compared to manually specifying prompt terms. However, training embeddings takes more effort.

Example Negative Prompt Workflow

Here is an example workflow for iteratively improving a prompt with negative prompts:

Start with a base positive prompt
Generate a batch of images
Identify the most common failure modes and flaws
Add 1-2 negative prompts targeting those issues
Repeat steps 2-4, adding/tuning negatives until image quality improves

This allows efficiently honing in negatives that address the specific flaws you see for a prompt.

Over time, you can build intuition for which negatives work best for different positive prompt types and subjects.

Key Takeaways

Negative prompts specify attributes for Stable Diffusion to avoid when generating images
They give you more control over the final output
Help avoid common AI flaws like bad anatomy and distortions
Experiment with negatives to address specific prompt issues
Build up an optimized negative prompt over time for each style

Mastering negative prompts alongside carefully crafted positive prompts will greatly improve your Stable Diffusion results.