ChatGPT Prompt系列教學9:保持ChatGPT連續生圖“一致性”

26 Oct 202310:35

TLDRThe video discusses the evolution of AI image generation tools, highlighting their transition from promising productivity enhancers to more recreational 'toys'. It addresses the initial optimism for creating valuable illustrated books and comics with AI, which has yet to be fully realized due to consistency issues in continuous image generation. However, recent improvements through the combination of ChatGPT and DALL-E 3 have shown significant progress. The video then delves into techniques for creating consistent AI-generated images using structured prompts, demonstrating the process with examples of different styles and actions. It concludes by emphasizing AI's potential as a powerful assistant in various tasks, despite not yet reaching the ideal of one-click comic or film creation.


  • 🙂 AI image generation tools were initially met with high expectations for significantly boosting productivity.
  • 💬 In practice, these tools have often been used more as entertainment 'toys' than professional utilities.
  • 🔍 The main challenge for creating sequential, consistent AI-generated images lies in the difficulty of maintaining style and character traits across images.
  • 🧠 Recent advancements in integrating ChatGPT with DALL-E 3 have notably improved the consistency of generated images.
  • 📝 By employing specific prompt strategies in ChatGPT, users can achieve more consistent and coherent images in various styles.
  • 📈 The tutorial demonstrates techniques for maintaining consistency, including setting a structured prompt format that combines style, basic information, additional details, action, and an identifier.
  • 💁‍🎨 Illustrative examples show how to generate images of a character named Anna in different scenarios, highlighting the effectiveness of these techniques.
  • 🛠 The illustrative style proved to be more stable than the realistic style, suggesting a strategy for choosing styles based on the desired consistency.
  • 👨‍🔧 For complex projects like comics or picture books, it's recommended to generate key elements separately and then composite them for better control and consistency.
  • 📱 The guide underscores the importance of using AI tools creatively and methodically to overcome limitations and enhance productivity.
  • 📺 The video also introduces the FutureList series, aiming to showcase tech companies and products that could shape the future.

Q & A

  • What was the initial expectation for AI image generation tools?

    -The initial expectation for AI image generation tools was that they would significantly improve productivity by being used to create outstanding picture books or comics.

  • Why have AI-generated picture books not been successful as initially hoped?

    -AI-generated picture books have not been successful because AI's consistency in continuous image generation is poor, making it difficult for new images to延续 the style or character traits of previous ones.

  • How has the issue of inconsistency in AI image generation been addressed recently?

    -The issue of inconsistency has been significantly improved by combining ChatGPT with DALL-E 3, which has enhanced the consistency of the generated images.

  • What is the unique approach ChatGPT takes in generating images compared to traditional AI tools?

    -ChatGPT's unique approach is that it first understands the description provided and then recreates a more suitable prompt based on that understanding to generate the image.

  • What is the structure of the prompt that the speaker used to achieve consistency in image generation?

    -The structure of the prompt used for consistent image generation includes 'Style', 'Basic Information', 'Additional Information', 'Action', and an identifier followed by a number (e.g., -0001).

  • Why is using a fixed character name like 'Anna' beneficial in maintaining image consistency?

    -Using a fixed character name like 'Anna' helps improve the consistency of the generated images by ensuring that the character's identity remains the same across different prompts and generations.

  • How does the 'identifier+1' part of the prompt contribute to maintaining image consistency?

    -The 'identifier+1' part allows ChatGPT to track each generated image with a unique number, helping to maintain consistency by tracking changes and ensuring that subsequent images follow the same pattern.

  • What is the speaker's suggestion for creating complex works like picture books or comics with AI?

    -The speaker suggests splitting complex works into key parts, generating them individually, and then using tools like Canva for overall composition, which results in a more stable outcome.

  • What are the challenges in maintaining image consistency when using realistic styles compared to cartoon styles?

    -Realistic styles present more challenges in maintaining image consistency because they require more detailed representation of features, making subtle differences more noticeable and thus harder to control.

  • How does the speaker address the issue of subtle differences in hairstyle and facial features across generated images?

    -The speaker suggests generating multiple images and selecting the ones that are closest in appearance to ensure consistency, especially when dealing with realistic styles.

  • What is the speaker's overall assessment of AI's capability as a tool for work and life?

    -The speaker acknowledges that while AI is not yet at the level of一键出漫畫 or 一键出影片, it has become a valuable assistant in work and life for those with strong动手和思考的能力 (ability to动手操作 and think critically).



🎨 AI Art Tools: Promises and Limitations

This paragraph discusses the initial high expectations for AI art tools as productivity enhancers, but notes their current use more as entertainment. It highlights the challenge of maintaining consistency in AI-generated images, especially in style and character traits, and how this issue has been significantly improved with the combination of ChatGPT and DALL-E 3. The speaker shares their experience using specific prompt techniques in ChatGPT to generate consistent images of characters in different scenarios and styles, emphasizing the simplicity and effectiveness of this method compared to traditional AI art tools.


📝 Understanding ChatGPT's Image Generation Process

The paragraph explains the unique process of image generation with ChatGPT, which involves understanding the user's description and creating a more suitable prompt for image generation. It contrasts this with traditional AI tools that directly generate images from the provided prompt. The speaker describes their approach to prompt construction, including defining a structure with 'style,' 'basic information,' 'additional information,' 'action,' and an identifier to ensure consistency across multiple images. The paragraph also touches on the importance of fixed character names and the role of additional information and actions in refining the generated images.


🚀 Demonstrating ChatGPT's Image Consistency

This section presents a practical demonstration of ChatGPT's image generation capabilities, following the structured prompts previously discussed. The speaker guides the audience through the process of generating images of a character named Anna in various scenarios and styles, noting the consistency in the generated images. It addresses minor differences in hairstyle and provides solutions for maintaining consistency. The paragraph concludes with a summary of the challenges faced, the higher stability of cartoon styles over realistic ones, and suggestions for creating complex works like illustrated books or comics by generating and compositing separate elements.

🌟 Closing Remarks and Future展望

The speaker concludes the video by thanking the viewers and introducing additional content on their channel, including the FutureList series, which will feature companies and products that could impact the future. The speaker, 大悅, encourages viewers to look forward to these new segments and signs off with a farewell, promising more content in the next episode.



💡AI-generated images

AI-generated images refer to the visual content created by artificial intelligence algorithms without human intervention. In the context of the video, it highlights the initial promise of AI tools to enhance productivity but also their current use as entertainment 'toys'. The video discusses the challenges in maintaining consistency in AI-generated images, such as style and character traits, and how recent advancements with ChatGPT and DALL-E 3 have improved this aspect.


Consistency in AI-generated images refers to the ability of the AI to produce images that maintain a uniform style, character features, and overall aesthetic across multiple generations. The video emphasizes the importance of consistency in creating cohesive visual content, such as illustrated books or comics, and how it has been a challenge for AI tools until recent improvements.


ChatGPT is an AI language model developed by OpenAI that is capable of generating human-like text based on the prompts it receives. In the video, ChatGPT is highlighted for its role in improving the consistency of AI-generated images by understanding the description and recreating a more suitable prompt for image generation, which was previously a challenge for traditional AI image-generating tools.


A prompt in the context of AI refers to the input or instruction given to the AI system to generate a specific output. The video discusses the importance of crafting effective prompts for ChatGPT to generate consistent AI images, including detailed descriptions and structural elements that guide the AI in creating the desired visual content.


DALL-E 3 is an AI model developed by OpenAI that is capable of generating images from textual descriptions. It is mentioned in the video as one of the recent advancements that have significantly improved the consistency of AI-generated images, allowing for better continuity in style and character features.

💡Illustration style

Illustration style refers to a specific visual art technique used in creating images, often characterized by simplified forms, exaggerated features, and a focus on storytelling or conveying a particular mood. In the video, the illustration style is one of the styles used to generate images of the character Anna, with examples shown of her in different scenarios like drinking coffee and riding a bike.

💡Realistic style

Realistic style in art and image generation aims to depict subjects as they would appear in real life, with attention to detail, accurate proportions, and lifelike textures. The video contrasts the illustration style with the realistic style, noting that while the latter provides richer detail, it also presents greater challenges in maintaining consistency due to the subtle differences in facial features and expressions.

💡Character consistency

Character consistency refers to the maintenance of a character's visual identity and traits across multiple images or episodes within a narrative. In the context of AI-generated images, it is crucial for creating cohesive stories in mediums like illustrated books or comics, where the same character must appear consistently.

💡Structured prompt

A structured prompt is a formatted input provided to an AI system that includes specific elements in a particular order to guide the output. In the video, a structured prompt is used to instruct ChatGPT on how to generate images with a consistent style and character features, by including elements like style, basic information, additional information, actions, and an identifier for tracking.


An identifier in the context of AI-generated images is a unique label or number assigned to each image to track and ensure continuity in a series of generated content. The video discusses the use of identifiers to maintain consistency in image series, where each subsequent image is numbered sequentially, allowing for easier comparison and selection of images that match the desired consistency.

💡Creative process

The creative process refers to the steps taken to produce original work, such as art or literature. In the video, the creative process involves using AI tools like ChatGPT and DALL-E 3 to generate images, carefully crafting prompts, and selecting the most consistent images to create a cohesive final product, such as an illustrated book or comic.

💡FutureList series

The FutureList series is a new program mentioned in the video where the host, 大悅 (Da Yue), plans to introduce a thousand companies or products that could influence the future. This series is presented as an additional content offering alongside the Prompt teaching series, indicating the host's focus on exploring and discussing emerging technologies and their potential impact.




ChatGPT和DALL-E 3的結合改善了AI生圖的一致性問題。