How to Use Negative Prompts in Midjourney

Christian Heidorn from Tokenized AI introduces the concept of negative prompting in the context of AI image generation with Mid Journey. He explains that negative prompting is a subtractive process that allows users to remove objects or concepts from their generated images, contrasting with the additive process of standard prompting. He clarifies that Mid Journey does not support in-painting like some other tools do, and that negative prompts are more effective for simpler scenes rather than complex ones. Two methods for negative prompting are discussed: using the 'no' parameter and negative image text weights.

The video demonstrates the use of the 'no' parameter to remove candles from an image of a birthday cake. After activating remix mode, the 'no candles' parameter is added to the prompt, and the AI generates a cake without candles while maintaining the original composition. It is noted that the 'no' parameter struggles with removing core concepts that are intrinsically linked to the main subject, as demonstrated by an unsuccessful attempt to remove eyes from a close-up image of a woman.

Christian illustrates the use of negative prompting with a more abstract concept, 'Nature's Meadow,' and attempts to remove flowers from the scene. He explains that the concept of a meadow can be flexible, allowing for the removal of certain elements like flowers. By using remix mode and specifying 'no flowers,' the AI successfully generates a meadow scene without flowers, maintaining the composition of the chosen image. The concept is further expanded to include animals from the animal kingdom, resulting in a scene with animals but no birds or flowers.

The video explores the use of text weights as an alternative to the 'no' parameter for negative prompting. It is shown that assigning a weight of -0.5 to certain elements, such as 'flowers' or 'birds,' is equivalent to using the 'no' parameter. An example is given where a rocky mountain landscape in winter is generated with an emphasized buffalo herd, but without trees or snow. This demonstrates the potential for crafting detailed scenes through careful use of text weights and negative prompting.

Christian concludes by emphasizing that negative prompting is not an exact science and involves a process of trial and error. He suggests that for simple removals, using the 'no' parameter might be more efficient, but for more complex scenes, text weights offer greater control. He acknowledges that while negative prompting in Mid Journey is not as precise as in-painting features in other tools, it remains a valuable method for refining AI-generated images. The video ends with a call to like and subscribe for more content on AI software navigation.