What AI Image Generator Should YOU Be Using??

Introduction and Overview

In this video, the speaker discusses various AI image generators and aims to determine the best tool for specific use cases. The speaker mentions several popular tools such as Mid Journey, Dolly 3, Firefly Image 2, Stable Diffusion XL, Google's generative search experience, and Idiogram.

Evaluating Accuracy of AI Image Generators

• The first criterion evaluated is accuracy.

• Prompt adherence is tested by providing specific prompts to each tool.

• Mid Journey performs reasonably well in following the prompt for a photo of a green bus floating in space.

• For a more complex prompt involving a sitting artist with a bucket hat painting a canvas of a three-headed monster, Mid Journey falls short in accuracy.

• Using the "style raw" option in Mid Journey improves adherence to the prompt but still has limitations.

• Dolly 3 inside Chat GPT demonstrates better accuracy than Mid Journey for both prompts.

• Bing's Image Creator with Dolly 3 also produces accurate results for both prompts.

Evaluating Other Criteria

Creativity and Realism

Illustrations, Logos, Vectors

Textures and Background using Text and Images

Censorship in Images

Usability of User Interfaces

Pricing Comparison

Conclusion and Final Thoughts

Comparing Image Generators

The speaker discusses different image generators and compares their performance.

SDXL vs. Leonardo

• SDXL is used as part of the image generation pipeline.

• Leonardo is used to compare against SDXL.

• Both generators produce good results with green buses floating in space.

Mid Journey vs. Leonardo

• Mid Journey performs better than Leonardo in adhering to the prompt.

• Both generators include a sitting artist and a bucket hat, but neither generates a three-headed monster.

Firefly Image 2

• Firefly Image 2 has a more cartoony style.

• It also generates green buses floating in space.

• Like Leonardo, it misses the prompt's requirement for a three-headed monster.

Google's Generative Search Experience

• Google recently introduced the ability to generate images directly from search queries.

• It successfully generates green buses floating in space, although one appears as a VW bus instead.

• When given the prompt for an artist sitting with a bucket hat painting a canvas of a three-headed monster, it initially fails but succeeds on the second attempt.

Idiogram

• Idiogram is another image generation tool.

• It produces cartoon-like images of green buses but not in space.

• When given the more complicated prompt, it partially succeeds by including elements like sitting and bucket hats but fails to accurately depict the three-headed monster.

Ranking and Creativity Assessment

The speaker ranks the image generators based on accuracy and creativity.

Accuracy Rankings

1. Dolly Three (9)

2. Google (7.2)

3. Mid Journey Raw (5.5)

4. SDXL (6.5)

5. Firefly Image 2 (6.5)

6. Idiogram (6.7)

Creativity Assessment

• Mid Journey generates visually appealing images with vibrant colors and interesting elements.

• Even with a one-word prompt like "Beauty," Mid Journey produces creative results.

• Raw's images are impressive but lose points for including random letters.

• Overall, Mid Journey receives a high creativity ranking of 9, while Raw gets a slightly lower score due to the inclusion of random letters.

The speaker mentions that they will pick up the pace in the video and provide prompts, results, and scores for the remaining image generators.

Comparison of Image Creativity

The speaker compares the creativity of different AI models in generating RGB images and images related to the prompt "Beauty".

Mid Journey vs. Dolly 3 (Image Creator)

• Mid Journey's RGB images are creative with contrast and depth, giving it a slight edge over Dolly 3.

• Dolly 3 (Image Creator) lacks creativity and produces colorful swirls instead of impressive images.

• Dolly 3 in chat GPT is more creative than Dolly 3 as an image creator, possibly due to additional context provided by chat GPT.

Leonardo vs. Adobe Firefly vs. Google

• Leonardo's RGB images are great, similar to mid Journey but with slightly less contrast.

• Images generated by Leonardo for the prompt "Beauty" are diverse compared to Adobe Firefly and Google.

• Adobe Firefly lacks creativity and produces similar unimpressive images for both prompts.

• Google generates four unique images but lacks vibrant colors in its RGB image.

Idiogram's RGB Images

• Idiogram generates four completely different and creative RGB images.

• The contrast is not as deep as mid Journey, but the overall quality is impressive.

Comparison of Realism in Generated Images

The speaker evaluates the realism of AI-generated images using a prompt featuring a couple holding hands in front of the Eiffel Tower.

Mid Journey's Realistic Images

• Mid Journey produces realistic images that could fool some people, although there may be minor flaws like floating lights.

Due to time constraints, further sections will be summarized in subsequent responses.

Evaluating Realism of AI-Generated Images

In this section, the speaker evaluates the realism of AI-generated images using different AI models.

Evaluating Dolly 3

• The speaker finds that the images generated by Dolly 3 do not look super realistic and have a Pixar-like quality.

• When using Bing's Image Creator, the images generated by Dolly 3 appear slightly more realistic than before.

Assessing Chat GPT's Dolly

• Some images generated by Chat GPT's Dolly exhibit weirdness in facial features, indicating they are AI-generated. However, one image stands out as more realistic than others.

• Leonardo also generates relatively realistic images but still has some issues with facial details and hand positions.

Analyzing Firefly Image 2

• Firefly Image 2 produces fairly realistic images, although it ignores the prompt instruction to include the Eiffel Tower in front. Facial details are lacking in some cases.

Examining Google's Output

• The speaker finds that neither of the images generated by Google is close to being realistic.

Reviewing Idiogram Results

• The images produced by Idiogram have various issues, such as disproportionate figures and lack of detail in people's faces. The Eiffel Tower appears somewhat acceptable due to blurring.

Summary of Realism Scale

• Mid Journey Raw is considered the most realistic model, followed by Firefly 2 and Mid Journey without using raw data. Other models did not meet expectations in terms of realism.

Evaluating Illustrations

In this section, the speaker assesses how well AI models generate drawings and illustrated work.

Using Nii Mode with Mid Journey

• The speaker switches to Nii mode in Mid Journey for generating illustrations. The prompt used is "anime girl with braids in the neon streets of Tokyo."

• The generated images are not described in detail, but it can be inferred that they were evaluated based on their resemblance to the given prompt.

Overall Summary

The speaker evaluates the realism of AI-generated images using various AI models such as Dolly 3, Chat GPT's Dolly, Leonardo, Firefly Image 2, Google, and Idiogram. They find that Mid Journey Raw produces the most realistic results among these models. Additionally, they briefly touch upon evaluating illustrations using Nii mode in Mid Journey.

Style Raw for Illustrations

The speaker discusses the use of Style Raw for creating illustrations and mentions that it doesn't perform well in this regard. They rate the performance of different tools in generating illustrations.

Performance of Different Tools

• Chat GPT: The images generated by Chat GPT are coherent and solid, but lack contrast compared to Mid Journey. Rating: 7/10.

• Bing Image Creator: Similar style to Chat GPT, with decent images but still lacking contrast. Rating: 7/10.

• Leonardo: Images have good contrast and depth, comparable to Mid Journey. Rating: 7.8/10.

• Firefly Image 2: All images have deep dark contrast and are solid. Rating: 7.5/10.

• Google AI Generated Images: Adherence to prompt is not great, but illustrations are good. Rating: 6.5/10.

• Idiogram: Images are pretty good at creating illustrations, rating varies between 6.8 and 7/10.

Overall Performance

The speaker concludes that all the tools tested are decent at creating illustrations, with Mid Journey, Leonardo, and Firefly Image 2 performing particularly well in terms of contrast and quality.

Logos and Vectors Generation

The speaker explores the performance of different tools in generating logos and vectors.

Performance of Different Tools

• Mid Journey (non raw): Solid vector images with a simple flat design. Rating: 8/10.

• Style Raw (Mid Journey): Comparable performance to non raw version. Rating: 8/10.

• Dolly 3 (Chat GPT): Images are good, but not as simple as desired for a logo. Rating: 7.5/10.

• Leonardo: Generated images are not exactly what was asked for in terms of simplicity. Rating: 6/10.

• Adobe Firefly: Solid performance with better results than DOI. Rating: 7.8/10.

• Google AI Generated Images: Nailed the request for simple flat vector logos. Impressive performance, rating at 8.3/10.

• Idiogram: Similar style to Firefly, solid performance in generating logos and vectors. Rating: 7.8/10.

Overall Performance

The speaker suggests that Mid Journey (non raw), Style Raw (Mid Journey), and Google AI Generated Images are the top performers in generating simple flat vector logos.

The ratings provided by the speaker are subjective and based on their own preferences and impressions during testing.

Design and Textures

This section discusses the ability of different AI models to create textured tile backgrounds. The prompt used is "colorful circuitry". The models tested are Mid Journey, Dolly 3 with Chat GPT, Leonardo, Adobe Firefly, Google, and Idiogram.

Mid Journey

• Mid Journey can create tilable images that seamlessly tile when placed side by side.

• It receives a rating of 10 for its tiling capabilities.

Dolly 3 with Chat GPT

• Dolly 3 claims to create tilable images but fails to do so effectively.

• The resulting images have visible seams and do not seamlessly tile.

• It receives a rating of 0 for its tiling capabilities.

Adobe Firefly

• Adobe Firefly can create tilable images that seamlessly tile when placed side by side.

• It receives a rating of 10 for its tiling capabilities.

Google

• Google struggles to create any sort of tiled image effectively.

• It receives a rating of 0 for its tiling capabilities.

Idiogram

• Idiogram attempts to create tiled images but has visible seams and does not pass the tiling test effectively.

• It receives a rating of 0 for its tiling capabilities.

Text in Image

This section explores the ability of AI models to generate text within an image. The prompt used is "a penguin holding a wooden sign that says subscribe to Matt wolf".

Mid Journey and Mid Journey Raw

• Both versions of Mid Journey fail to generate accurate text within the image.

• They receive a rating of 0 for their text generation capabilities.

Dolly 3 with Chat GPT

• Dolly 3 successfully generates text within the image, but there are some typos.

• It receives a rating of 7.5 for its text generation capabilities.

Dolly with Bing's Image Creator

• Both versions of Dolly successfully generate text within the image, with minor typos.

• They receive a positive rating for their text generation capabilities.

Conclusion

In summary, Mid Journey and Adobe Firefly perform well in creating tilable images, while Dolly 3, Google, and Idiogram struggle with tiling. When it comes to generating text within an image, Dolly 3 and Dolly with Bing's Image Creator show promising results, although there is room for improvement.

Accuracy of Text in Generated Images

The speaker discusses the accuracy of text in generated images and compares different AI models.

Mid Journey AI Model

• The speaker mentions that the Mid Journey AI model spelled their last name correctly but got their first name wrong.

• The generated image had some errors, such as adding an extra "F" in the word "subscribe" and incorrectly stating the speaker's name as "Matt Whitwell."

• Overall, the accuracy of text in images generated by Mid Journey is not great.

Firefly 2 AI Model

• The speaker rates Firefly 2 slightly better than Mid Journey in terms of image quality.

• While the images look good visually, they still struggle with accurately representing text.

• Some letters are correct, but overall, it falls short in accurately depicting words.

Changes to Prompt for Google AI Model

• To get better results from the Google AI model, the speaker had to modify the prompt by typing "create an image of a penguin holding a wooden sign that says subscribe."

• Quotations around phrases or adding additional words did not yield satisfactory results.

Idiogram AI Model

• Idiogram was able to generate images with accurate text consistently.

• However, it struggled with generating more than one word at a time.

• Despite this limitation, it performed well in following prompts and producing correct text within images.

Summary:

The accuracy of text in generated images varies across different AI models. Mid Journey and Firefly 2 have limitations when it comes to accurately representing text. On the other hand, Google and Idiogram perform better in terms of generating correct text within images. However, Idiogram struggles with generating multiple words simultaneously.

Censorship Comparison

This section compares the censorship levels of different AI models.

Censorship Levels

• Google: Google censors certain words and may not generate images of celebrities.

• Mid Journey: Mid Journey censors some words but still generates images of trademarked characters.

• DALL·E: DALL·E does not seem to be heavily censored and generates a variety of images, including trademarked characters.

• Idiogram: Idiogram does not appear to be heavily censored and generates a wide range of images, including potentially controversial ones.

• Firefly: Firefly is more heavily censored compared to other models and rejects many prompts with specific IP or people's names.

Usability Comparison

This section evaluates the usability of different AI models.

Usability Ratings

• Mid Journey: Mid Journey has a less user-friendly experience as it requires using Discord commands within the platform.

• DALL·E in ChatGPT: DALL·E in ChatGPT offers a good user experience similar to using ChatGPT, allowing for easy conversations and prompt customization.

• DALL·E in Bang's Image Creator: DALL·E in Bang's Image Creator is straightforward but lacks advanced features like aspect ratio changes or additional inpainting options.

• Leonardo: Leonardo provides extensive customization options, intuitive user interface, and prompt generation assistance, making it highly versatile and user-friendly.

• Firefly 2: Firefly 2 offers a range of features, including aspect ratio changes, style matching, and effects, with a simple and intuitive user interface.

• Google: Google's usability is familiar to those who have used the search engine but can be confusing at times due to specific wording requirements.

The remaining sections will be summarized in subsequent responses.

Pricing Comparison

This section provides a comparison of the pricing plans for different image generation platforms.

Pricing Plans

• Dolly 3:

• Lowest plan is $10/month.

• Dolly 3 requires a Chat GPT Plus membership, which costs $20/month. Making it the most expensive option.

• Free usage available with limited features and slower generation speed.

• Leonardo:

• Offers both free and paid plans.

• Free plan provides a generous amount of daily tokens for image generation.

• Paid plan costs $10/month, allowing for more image generation compared to mid Journey.

• Firefly:

• Offers both free and paid plans.

• Free plan includes 25 image generations per month.

• Paid plan costs $5/month and provides 100 monthly generative credits.

• Google:

• Completely free to use.

• Idiogram:

• Currently completely free to use.

Price Rankings

• Most Expensive: Dolly (5/10)

• Best Value: Leonardo (7.5/10)

• Completely Free Options: Google (10/10), Idiogram (10/10)

Image Generation Capabilities

This section discusses the strengths of each platform in terms of generating accurate, creative, realistic, or specific types of images.

Accurate Generations from Prompts

• Recommended Platforms: Dolly, Google

• Usability Issues with Google's Platform

Creative Images

• Recommended Platforms: Mid Journey, Leonardo, Stable Diffusion XL

• Honorable Mention to Dolly and Firefly

• Mid Journey excels at realism

• Good Illustration Options: M Journey, Leonardo

• Best Vector Generation: Google followed by Mid Journey and Idiogram

Textures and Backgrounds

• Platforms that Generate Tiling Textures: Mid Journey, Stable Diffusion, Leonardo

• Firefly Version 2 also generates tiling textures

Text in Images

• Platforms Capable of Adding Text: Google, Idiogram, Dolly versions

Censorship Concerns

• Least Censored Platform: Stable Diffusion

• Other Options with Less Censorship: Leonardo, Idiogram, Mid Journey, Google

Usability Rankings

• Best Usability: Leonardo

• Good Usability: Firefly, Dolly 3 (Chat GPT interface)

Summary and Rankings

This section provides a summary of the different image generation platforms and their rankings based on various factors.

Summary of Platforms

• Ideogram:

• Free to use.

• Uncensored.

• Good for text and images.

• Mid Journey:

• Not free but versatile.

• Great for creative images, textures, logos, illustrations.

• Leonardo:

• Not free but least censored.

• Performs well in various aspects except for text inside images.

Overall Rankings (Out of 100)

1. Leonardo (75.5/100)

2. Tie between Mid Journey and Ideogram (both around mid-range scores)

3. Dolly 3 performs the worst due to high cost and limitations.

The rankings are subjective based on the given criteria.

Comparing AI Image Generators

In this section, the speaker provides a final wrap-up of comparing various AI image generators and their best and worst use cases.

Final Thoughts on AI Image Generators

• The AI image generators discussed in the video are accurate and provide good illustrations.

• These tools can also generate text inside images.

• It is important to consider the specific use case when choosing an AI image generator.

Conclusion and Next Steps

The speaker concludes the video and suggests next steps for viewers.

Conclusion

• The speaker had a lot of fun creating this video but it was one of the more intense ones.

• Viewers' results may differ from what was shown in the video.

• The comparison provided should give more clarity on which tools to use in different situations.

Next Steps

• Viewers are encouraged to check out "Future Tools," where the speaker curates cool AI tools on a daily basis.

• Future Tools also includes curated AI news.

• There is a free newsletter available that sends subscribers the coolest tools and news directly to their email inbox.

Final Remarks

The speaker wraps up the video with final remarks and encourages viewers to engage with the channel.

Final Remarks

• The speaker thanks viewers for tuning in and nerding out about AI image generators.

• This video may be longer than usual, but it aims to be thorough in providing information.

• If viewers enjoyed the video, they are encouraged to give it a thumbs up.

Channel Engagement

• Viewers who haven't already subscribed are invited to consider subscribing to the channel.

• The goal is to reach one million subscribers by the end of 2024. Subscriptions would greatly help achieve this goal.

Closing

The speaker concludes the video.

Closing

• The speaker says goodbye and ends the video.