Prompting HACKS no-one talks about

24 May 202311:27

TLDRThis video delves into advanced techniques for refining AI-generated images, focusing on the concept of 'bleeding' where certain words or concepts inadvertently alter the output. The host explains how aspect ratio can influence the quality of results, using examples like selfies and running images. The video also explores 'concept bleeding,' where AI's understanding of images is shaped by the frequency of associated elements rather than their logical connection. To combat this, the host demonstrates prompt editing strategies, such as removing or altering problematic words at specific stages of the image generation process. Additionally, the video covers the use of seed variations to create diverse outputs from a single prompt and the strategic use of 'break' and 'AND' in prompts to improve color accuracy and ensure all elements are considered. The host also shares creative applications of these techniques, like generating surreal designs and modifying backgrounds. The video concludes with a tip on using metadata from images to enhance prompts.


  • 🚨 Bleeding occurs when elements in a prompt affect the AI's output in unintended ways, such as aspect ratio impacting the focus of an image.
  • 🧠 Concept bleeding is when a word or concept indirectly alters the AI's output, sometimes leading to unexpected changes in the generated image.
  • πŸ™οΈ Removing certain words from a prompt can help avoid unwanted concept bleeding, such as taking 'city' out to generate interior images instead of exterior cityscapes.
  • βœ‚οΈ Prompt editing is a powerful technique to control the AI's output, allowing for the creation of complex images by defining the composition first and then altering the content.
  • πŸ”„ Using seed variations can create multiple versions of an image with slight differences, offering a range of options to choose from.
  • πŸ› οΈ The 'BREAK' keyword in all caps can help reset the AI's focus, improving the positioning of elements in the generated image.
  • 🎨 Adding 'BREAK' before complex parts of the prompt can help the AI better understand and execute the desired output, especially when the prompt becomes too intricate.
  • πŸ”΅ Using the '&' symbol in all caps can create a mix of different generated images into one, offering a unique way to blend concepts.
  • 🌐 The 'AND' keyword can be used to prioritize certain elements of the prompt, providing more control over the final composition.
  • πŸ“ˆ Mid-prompt word strength adjustments can influence the AI's focus, though practical applications are still being explored.
  • 🧩 Prompt editing can be used creatively to transform one concept into another within the same image, such as turning a flower pot into fireworks.
  • βž• Adding metadata to prompts by dragging an image can automatically include relevant information, simplifying the process of generating specific images.

Q & A

  • What is 'bleeding' in the context of AI image generation?

    -Bleeding refers to when a certain element or concept unintentionally and unexpectedly influences the AI's generated image in a way that deviates from the intended prompt. It's often related to how the AI model was trained and the associations it has learned between words and visual elements.

  • How does changing the aspect ratio affect the AI's interpretation of a prompt?

    -Changing the aspect ratio can significantly affect the AI's interpretation by making it follow the prompt more or less effectively. For instance, a vertical aspect ratio might be better for selfies, while a horizontal aspect ratio could be more suitable for full-body shots.

  • What is concept bleeding and how does it manifest in AI-generated images?

    -Concept bleeding is when a word or concept indirectly alters the AI's generated results. It can cause the entire image to change in ways that may or may not make logical sense to humans, depending on the AI's learned associations from its training data.

  • How can one identify and combat concept bleeding in AI image generation?

    -To identify concept bleeding, one can look for words that strongly influence the generated image. To combat it, one can edit the prompt by removing or altering the problematic words, or by adding new words after the composition has been decided to steer the AI's interpretation in the desired direction.

  • What is the purpose of using the 'break' keyword in a prompt?

    -The 'break' keyword can be used to signal the AI to pause and restart reading the prompt from a certain point. It can help with better color positioning and can make the AI better understand complex prompts by separating different elements.

  • How can prompt editing be used to create images that are hard to describe?

    -Prompt editing allows users to create two separate prompts: one that defines the overall shape and composition, and another that transforms this composition into the desired final image. This technique can be powerful for creating complex images that the AI might otherwise struggle to understand from a single prompt.

  • What is the effect of using the 'AND' keyword in all caps in a prompt?

    -Using 'AND' in all caps can cause the AI to generate multiple interpretations of the prompt and then attempt to mix them into a single image. It can be used to create unique and unexpected designs, but it should be used with caution as the results can be unpredictable.

  • How can seed variations be used to generate different versions of an image?

    -Seed variations allow the AI to create a new noise and mix it with the original, resulting in different versions of the same image. By adjusting the variation strength and seed size, users can control the extent of the variations.

  • What is the benefit of adding metadata to a prompt using an image?

    -Adding an image to the positive prompt allows the AI to automatically incorporate the metadata from that image into the prompt, which can help guide the generation process towards specific visual elements or styles.

  • How can one use the vertical bar method for prompt editing?

    -The vertical bar method involves adding two dots and a weight to a prompt to indicate which part of the prompt should be prioritized. This gives users more control over the final image, allowing them to guide the AI more effectively.

  • What is the significance of restarting the web UI file when generating new images?

    -Restarting the web UI file can help to prevent old prompt components from bleeding into new ones, ensuring that each new generation is less influenced by previous ones and maintaining the uniqueness of each image.



πŸ–ΌοΈ Concept Bleeding in AI Image Generation

The first paragraph discusses the phenomenon of 'concept bleeding' in AI-generated images, where a word or concept indirectly alters the output. The speaker explains how certain prompts can lead to unexpected results due to the AI's training data. For example, using 'running' in a prompt might lead to full-body shots because the AI associates running with a full-body context based on its training. The speaker also provides strategies to identify and combat concept bleeding, such as editing prompts and using the right steps in the generation process to influence the outcome.


πŸ”„ Advanced Prompt Editing Techniques

The second paragraph delves into advanced prompt editing techniques to create complex images. It introduces the idea of using two distinct prompts: one to define the overall shape and composition, and another to transform that composition into a completely different subject. The speaker also discusses the use of 'BREAK' in all caps to help the AI better understand and prioritize different parts of the prompt, leading to more consistent results. Additionally, the paragraph touches on the use of 'AND' in all caps to blend multiple concepts into a single image, which can be unpredictable but also creative.


🌱 Creative Uses of AND and BREAK in Prompts

The third paragraph explores the creative applications of the 'AND' and 'BREAK' commands in prompts. It explains how 'AND' can be used to blend different concepts, such as combining 'donuts' and 'grounds' to create unique designs. The speaker also highlights the utility of 'BREAK' for improving color positioning and getting the AI to pay attention to all parts of the prompt. Furthermore, the paragraph mentions an extra tip about adding metadata to prompts by dragging an image into the positive prompt field.




In the context of the video, 'bleeding' refers to the phenomenon where certain elements or concepts in a prompt unintentionally influence the AI's image generation process, leading to unexpected or incorrect results. The term is used to describe how aspects like the aspect ratio or specific words can cause the AI to generate images that are not exactly what the user intended. For example, changing the aspect ratio for selfies or running images can lead to better or worse adherence to the prompt depending on the orientation chosen.

πŸ’‘Concept Bleeding

This term describes a specific type of bleeding where a word or concept indirectly alters the outcome of an image generation in a subtle or unintended way. It is a significant issue in AI-generated content where the AI might associate certain concepts with particular visual elements based on its training data, even if those associations do not logically align with human understanding. The video provides examples such as 'blue house' and 'watermelon', which when added to the prompt, drastically change the generated image beyond the expected outcome.

πŸ’‘Aspect Ratio

The aspect ratio in the video script refers to the proportional relationship between the width and the height of an image, which can significantly impact how an AI interprets and generates an image based on a given prompt. The speaker notes that for certain types of images, like selfies or full-body shots, specific aspect ratios can lead to better or worse results, highlighting the importance of this parameter in the image generation process.

πŸ’‘Prompt Editing

Prompt editing is a technique discussed in the video where the user modifies the initial prompt during the image generation process to steer the AI towards a desired outcome. This can involve adding or removing certain words or concepts at specific steps of the generation to either correct unwanted bleeding effects or to intentionally manipulate the AI's output. The video demonstrates how prompt editing can be used to create complex images that are difficult to describe or that the AI might not understand from a single prompt.

πŸ’‘Stable Diffusion

Stable Diffusion is the AI model being referenced in the video for generating images from textual prompts. The term is used throughout the script to describe the AI's behavior and its tendency to 'bleed' concepts or to generate images based on statistical patterns in its training data rather than a logical understanding of the concepts. The video aims to provide insights on how to work with Stable Diffusion to achieve better results.

πŸ’‘Seed Variations

Seed variations are a feature in the AI image generation process that allows for the creation of multiple variations of an image by introducing different 'noise' into the system. The script explains that this can be used to generate a range of images from a single prompt, offering a way to explore different outcomes and find the most appealing result. The strength of the variation and the point at which it is introduced can affect the final output.


In the context of the video, 'break' is used as a command within the prompt to signal the AI to pause and then restart the interpretation of the prompt from that point. This can help in managing complex prompts and ensuring that the AI focuses on specific parts of the prompt that might otherwise be overlooked. The video illustrates how using 'break' can lead to better color positioning and a clearer understanding of the user's request by the AI.


The use of 'AND' in all caps within the prompt serves as a tool to create a composite image from multiple concepts. The video suggests that the AI treats each instance of 'AND' as a separate generation attempt, which it then tries to blend into a single image. This can result in unique and complex designs, although the method can be unpredictable and requires careful experimentation to control the outcome.


Metadata refers to the data that provides information about other data. In the video, it is mentioned in the context of dragging an image into a prompt, which automatically adds metadata to the prompt. This can be useful for including specific details or characteristics in the generated image that are present in the image file being used.

πŸ’‘Batch Size

Batch size in the video refers to the number of images generated simultaneously by the AI using a given prompt or set of parameters. The script discusses how adjusting the batch size can affect the outcome when using certain prompt editing techniques, such as with 'AND' in all caps, where a higher batch size can lead to a more diverse mix of generated images.

πŸ’‘Environment Setting

Environment setting is a concept used in the video to describe the background or context of an image generated by the AI. The speaker talks about how adding certain words to the prompt, like 'wild forest' or 'deep ocean', can change the setting of the image to match the described environment, even if the main subject of the image remains the same.


Bleeding occurs when a factor affects the result in an implicit, unexpected, or incorrect way.

Aspect ratio changes can influence how well a model follows a prompt.

Concept bleeding is when a word or concept indirectly alters the outcome.

AI does not understand the meaning of things but learns from patterns in the data.

Implicit bleeding can make sense, like associating running with full body shots.

Unexpected bleeding can lead to unpredictable changes in image generation.

Combating concept bleeding involves identifying and manipulating key concepts in the prompt.

Editing prompts can help control the composition and style of generated images.

Prompt editing can be used to create hard-to-describe images by first defining the shape and composition.

Using 'break' in prompts can help with color positioning and understanding complex prompts.

The '&' symbol can create unique designs by mixing multiple concepts into one image.

Seed variations allow for the creation of different versions of an image by altering the noise and seed parameters.

Prompt editing can be used to change specific elements of an image, like turning a flower pot into fireworks.

Using 'break' can help ensure that all parts of a complex prompt are considered by the AI.

The '&' symbol can be used for environment setting, changing backgrounds without affecting the main subject.

Adding metadata to prompts by dragging an image can automatically include relevant information.