The Free & Uncensored Version of MidJourney! (FLUX.1)

Introduction to Flux One

Overview of the New AI Image Generating Tool

• A new AI image generating tool called Flux One has emerged, claiming to rival and even outperform Mid Journey in certain aspects.

• Developed by a team from Black Forest Labs, many members contributed to the creation of Stable Diffusion and other significant innovations in AI image generation.

Models Available in Flux One

• Flux offers three models:

• Flux One Schnell: Fastest model for local development; open-source under Apache 2.0 license.

• Flux One Dev: Middle-tier model that is more efficient than Schnell but not for commercial use.

• Flux One Pro: Top-tier model designed for enterprise solutions with state-of-the-art performance.

Using Flux Models

Accessing the Models

• Users can access the models through various websites, including Hugging Face where both Schnell and Dev models are available for free.

• The interface allows users to input prompts and adjust settings like random seed, height, and inference steps.

Building Workflows with Glyph

• Glyph is a platform that enables users to create custom workflows using Flux models without cost.

• Users can enhance their prompts using LLM tools like ChatGPT before generating images with either Flux Pro or Schnell.

Performance Insights from User Experience

Feedback on Model Capabilities

• Insights were gathered from an experienced user named Miguel (Angry Penguin PNG), highlighting strengths and weaknesses of Flux.

Illustrating an Angry Penguin: A Comparison of AI Art Models

Exploring Different Artistic Styles

• The speaker attempts to create a hand-drawn illustration of an angry penguin, noting that while the image looks good, it lacks the true hand-drawn quality.

• An oil painting version is generated; although visually appealing, it does not convincingly represent an oil painting style.

• A watercolor painting of the same subject is produced, which captures some watercolor elements but still doesn't fully embody the medium's characteristics.

• In comparison with MidJourney's output for similar prompts, MidJourney's results appear more aligned with traditional artistic styles like watercolor and oil painting.

Evaluating Flux Model Capabilities

• The speaker discusses Flux’s strengths in realism and aesthetic training, suggesting it produces high-quality outputs comparable to MidJourney when using appropriate prompts.

• A photo-realistic image prompt generates a realistic depiction of a man eating ice cream on a city sidewalk, though some details may be misrepresented (e.g., sniffing instead of eating).

• Another prompt for a woman taking a selfie on a tropical island yields decent results but lacks ultra-realism due to potential prompt optimization issues.

Simplifying Image Generation Workflows

• The speaker demonstrates creating a basic image generator workflow using Flux Pro without optimizing prompts.

• A portrait prompt generates an image that appears realistic but falls short compared to MidJourney in terms of realism and context (e.g., "about to give a meeting" isn't visually represented).

Strength in Text-Based Creations

• Flux excels at generating images involving text elements such as logos or memes; examples include humorous depictions like George Washington crossing the Delaware with modern twists.

• A polar bear holding a sign successfully captures the intended message on the first attempt, showcasing strong adherence to text prompts.

Challenges with Longer Text Prompts

• When testing longer text prompts (e.g., plane writing "subscribed"), results vary; initial attempts yield partial success but miss complete adherence to longer phrases.

• Subsequent trials show improvement in visual representation but still struggle with capturing all requested text accurately.

Prompt Adherence Comparisons

• The discussion highlights how different models handle complex prompts. For instance, MidJourney often misses multiple elements from intricate requests.

• While Flux shows promise in adhering closely to detailed prompts (like depicting specific scenarios), its performance remains variable compared to competitors like DALL-E 3 and MidJourney.

AI Art Generation: A Comparison of Models

Overview of Dragon Imagery

• The speaker discusses various dragon images, highlighting a normal dragon in cowboy boots eating chips and comparing it to more complex prompts.

• Demonstrates the capabilities of Dolly 3, which successfully generates a three-headed dragon eating nachos while wearing cowboy boots and watching TV, showcasing its prompt adherence.

Limitations and Capabilities of AI Models

• Mentions that while current models cannot generate not safe for work content, they can theoretically create copyrighted images. An example is given with SpongeBob SquarePants high-fiving Super Mario.

• Tests generating celebrity images like Tom Hanks hugging Kanye West; results show some discrepancies but overall recognizable outputs.

Prompting Techniques for Better Outputs

• Emphasizes the flexibility of generating various imaginative scenarios without restrictions (e.g., Donald Trump eating a donut).

• Discusses methods to enhance prompts using AI tools or specific keywords (like "4K" or "HD") to improve image quality.

Advanced Prompting Strategies

• Suggests being as detailed as possible in prompts for better results; mentions users pushing boundaries with creative descriptions.

• Highlights an example where typography creatively forms the word "cat," demonstrating the potential of imaginative prompting.

Future Developments in AI Art Generation

• Introduces Flux One's upcoming text-to-video capabilities, indicating a shift towards more dynamic content generation.

• Compares Flux One's performance against Mid Journey and Dolly 3; notes that while improvements are evident, it hasn't surpassed them yet.

Conclusion on Model Performance and Potential

• Concludes that Flux One shows promise by combining strengths from existing models but still has room for improvement in realism and prompt adherence.

• Predicts that as Flux evolves, it could integrate features from all leading models, making it a versatile tool for creators.

• Notes the open-source nature of Flux will allow developers to refine it further, enhancing its capabilities over time.

New Open Source Model Competing with Stable Diffusion

Excitement for New Developments

• The speaker expresses enthusiasm about a new open-source model that appears to outperform Stable Diffusion, indicating a competitive landscape in AI tools.

• A follow-up video is planned to share further insights and improvements related to the new model as the speaker learns more.

Acknowledgments and Resources

• Gratitude is extended to Miguel, also known as Angry Penguin, for assistance in creating the video and explaining the capabilities of Flux.