I Tested ChatGPT's New Image 2.0 and Accidentally Stumbled Upon an Awesome Workflow

What's New with GPT Image 2.0?

Introduction to GPT Image 2.0

• The speaker introduces the topic of GPT Image 2.0, emphasizing that they will share new insights not widely discussed yet.

• They mention their long-term use of Nana Banana and express excitement about potentially finding a superior tool in GPT Image 2.0.

Discovering a New Gem

• The speaker reveals they have created an updated version of a gem for image prompting while testing GPT Image 2.0.

• They explain the advantages of using JSON code for image prompts over traditional text inputs, citing better structure and results.

Example Prompt and Results

• A unique prompt is introduced: NASA astronauts on the moon playing kickball without gravity, leading to humorous imagery.

• The process involves submitting this prompt to generate JSON code automatically, which includes meta tokens for enhanced cinematic effects.

Generating Images with Aspect Ratios

• After generating the initial image, the speaker demonstrates how to adjust aspect ratios easily within the tool.

• They showcase different aspect ratios (16:9 and 9:16), allowing users to create images suitable for various platforms like YouTube or TikTok.

Editing Features in Action

• The speaker illustrates editing capabilities by changing the color of an object in the generated image (the ball).

• This feature allows for specific alterations while maintaining overall image integrity, showcasing flexibility in creative processes.

Nature Photo Creation with AI

Prompting for Nature Images

• The speaker discusses creating a nature photo prompt featuring a bald eagle flying over a crystal-clear lake surrounded by evergreens and tall oaks during summer in the late afternoon.

• If the AI attempts to create an image instead of providing a prompt, users are advised to edit the request rather than re-prompting to save tokens.

• The speaker demonstrates pasting the prompt back into the chat, indicating that it should work without issues if kept within the same conversation.

Image Editing Techniques

• After generating an initial image, the speaker decides to add a fish in the eagle's talons using an editing tool that allows selection and painting over specific areas.

• The tool used for dictation is called Whisper Flow, which significantly saves time when prompting; it's not sponsored but highly recommended by the speaker.

Observations on AI Performance

• The generated image includes a fish in the eagle's talons, although some aspects appear strange. The speaker notes that early versions of models may have quirks as they launch.

• Acknowledging this is day one of launching this model, there are expectations for improvements as development continues.

Creating Unique Prompts

Experimenting with Unusual Concepts

• The next prompt involves a chimpanzee and a miniature donkey-giraffe creature playing football in front of a suburban house in northern New Jersey.

• This unusual combination aims to test whether the AI can accurately depict both elements while maintaining their unique characteristics.

Results from Creative Prompts

• Upon submission, the AI successfully generates an image featuring both creatures playing football while dressed appropriately, showcasing its ability to handle complex prompts.

Advanced Techniques for Image Generation

• The speaker introduces a two-part hack available only for paid ChatGPT users: putting ChatGPT into "thinking mode" before requesting multiple variations of an image based on prompts.

• Users can request up to eight variations; however, four are demonstrated here. Each variation shows different interpretations of the original prompt.

Final Thoughts on Image Variations

• After generating several images based on one prompt, new chats can be opened to explore further variations or adjustments using similar prompts.

Creating a Storyboard with AI-Generated Images

Generating Initial Images

• Jason initiates the process by submitting a prompt to create an image, noting that each version will differ slightly due to the AI's nature.

• He decides to generate five images that will serve as a storyboard for a narrative involving a chimpanzee and a miniature donkey giraffe.

Developing the Storyboard Concept

• The goal is to use the generated images as references to tell an overall story, emphasizing their role in visual storytelling.

• Excitement arises from the potential of integrating these images into cinematic tools like Higgsfield and Seedance, allowing for movie creation based on the storyboard.

Image Generation Process

• During testing, Jason observes that sometimes the initial image may not be first in sequence but still contributes meaningfully to the story layout.

• The AI focuses on creating images around a consistent theme—specifically, a football chase scene—demonstrating its ability to maintain narrative coherence.

Analyzing Image Consistency

• As he reviews the generated images, Jason notes their order reflects logical progression: starting with play in the street and moving to other scenes.

• He highlights how consistency is maintained across images regarding character appearance and setting details.

Practical Applications of Generated Images

• Users can download these images for various purposes such as creating movies or enhancing articles with visual storytelling elements.

• While up to eight images can be created, Jason recommends limiting this number to six for better quality and coherence in storytelling.

Quality Considerations in Image Editing

• Repeated editing of original images may lead to degradation in quality due to artifacts introduced during modifications.

• A suggested workaround involves copying original images into new chats for fresh edits without losing quality over time.

Image Editing Workflow

Removing Elements from Images

• The process allows users to remove unwanted elements from images, such as a "keep off the grass" sign. This is done by prompting the system with the desired changes.

• When a second image is created based on modifications, it does not revert to the original image but instead uses the latest version for further edits. This highlights how iterative editing works in this context.

Resources and Documentation

• Users will receive both a gem and source files, enabling them to create their own versions of the workflow if they choose to do so. This provides flexibility for those interested in customization.

• A Notion document will accompany these resources, summarizing key information about the workflow and offering guidance on its implementation. This aims to enhance user understanding and usability of the tools provided.

User Engagement and Feedback

• The speaker encourages users to provide feedback on their experiences with the tool, emphasizing community engagement and shared learning as vital components of improvement. Users are invited to share insights or challenges they encounter while using it.

• The speaker expresses hope that users will find value in the gem and looks forward to seeing creative applications stemming from it, fostering an environment of innovation among users.