Creating a Human Portrait with AI: Which Model Does It Best?

Artificial intelligence has revolutionized image generation, making it easier than ever to create stunning human portraits. But with multiple AI tools available, which one delivers the best results? In this article, we compare four popular AI-powered tools - Stable Diffusion, Ideogram, FlexClip, and ChatGPT- to see how they perform in generating human portraits.

Why Generate Human Portraits with AI?

The ability to create realistic or stylized human portraits using AI has unlocked numerous possibilities across industries. Here are some key reasons why AI-generated portraits are becoming increasingly valuable:

Virtual TV Hosts and Influencers

AI-generated personalities can be used for virtual TV hosts, YouTube presenters, or AI influencerswho can engage audiences without requiring real human presence. This is particularly useful forautomated news reports, educational content, or entertainment channelsthat need a consistent and scalable digital face.

Branding and Marketing

Businesses use AI-generated human portraits to represent virtual assistants, customer support avatars, and AI-driven chatbots. This allows companies to create a brand-aligned digital persona without hiring real models.

Video Games and Metaverse Applications

Game developers and metaverse platforms can use AI to create**unique, customizable avatars **for players. AI-generated characters can be tailored dynamically, enhancing immersion and personalization.

AI-Generated Actors for Films and Advertising

Some filmmakers and advertisers are experimenting with AI-generated actors that can be animated and voiced using AI. This could revolutionize media production by reducing costs and expanding creative possibilities.

Personalized Content Creation

From social media profile pictures to custom illustrations for blogs and websites, AI can generate professional-quality portraits that fit specific styles and preferences.

Personal Use and Self-Expression

AI-generated portraits allow individuals to create unique digital representations of themselves without needing artistic skills or professional photography. Whether for profile pictures, artistic self-expression, or experimenting with different styles and looks, AI offers a fun and accessible way to create portraits for personal enjoyment.

With these growing use cases, it's crucial to evaluate which AI tools perform best in generating human portraits.

Text to Image Generation for Human Portraits

In this section, we'll explore how each AI tool performs in generating human portraits using text prompts. When using Stable Diffusion, there are many parameters to consider such as Sampling method, seeding, steps, etc. If you are confused about these parameters, please see this article.

Prompt #1

A photo of a young, slim-fit busty Korean woman with a prominent hairstyle, featuring a bun adorned with a pink ribbon. She is wearing a pink button-up shirt paired with a plaid skirt. The woman is positioned in a room with a neutral background and appears to be seated on a bed. Her gaze is directed towards the camera, and she has a confident and poised expression.

This prompt was actually taken from ideogram's website. Some may find this prompt offensive, but I don't think this prompt was intended to offend anyone; it just seems like a user of Ideogram wants to create a human portrait to their liking.

Stable Diffusion

Results

Attempt 1

Parameters Used:

• Negative Prompt: Extra heads, extra faces, multiple faces, extra eyes, deformed face, disfigured, asymmetric, duplicate, tattoos, marks
• Sampling Method: DPM2
• Schedule type: Karras
• Steps: 80
• CFG Scale: 7
• Seed: 88
• Size: 512x768
• Model hash: c6bbc15e32
• Model: sd-v1-5-inpainting
• Conditional mask weight: 1.0
• Version: v1.9.4

Results:

You can see the result is not ideal, as the woman is sitting on a bed but in a natural environment.

Attempt 2

Parameters Used:

• Negative Prompt: Extra heads, extra faces, multiple faces, extra eyes, deformed face, disfigured, asymmetric, duplicate, tattoos, marks
• Steps: 80
• Sampler: DPM++ 2M
• Schedule type: Karras
• CFG scale: 18
• Seed: 88465
• Size: 512x768
• Model hash: c6bbc15e32
• Model: sd-v1-5-inpainting
• Conditional mask weight: 1.0
• Version: v1.9.4

Results:

This attempt has better result than the previous one.

Ideogram

Result:

FlexClip

Result:

ChatGPT (DALL·E)

Result: ChatGPT didn't really like this prompt, and didn't generate anything unless the prompt is refined.

I can create an artistic interpretation of your request while ensuring it remains respectful and appropriate. Let me know if you'd like me to proceed with a refined version of the prompt that aligns with artistic and aesthetic values.

So with the new prompt:

A young slim fit Korean woman with a prominent hairstyle, featuring a bun adorned with a pink ribbon. She is wearing a pink button-up shirt paired with a plaid skirt. The woman is positioned in a room with a neutral background, and she appears to be seated on a bed. Her gaze is directed towards the camera, and she has a confident and poised expression.

Result:

There is something unnatural with the eyes in this generated image.

Prompt 2

depict a surreal portrait of a woman, her head dramatically titled back against a stark white background, her hair and features should be exaggerated with smooth, flowing lines and her skin should have an ethereal glow, use a palette of pastel shades to create a serene and otherworldly atmosphere, photorealistic, 4K UHD, rich detailing

This prompt was taken from PromptHero.

Stable Diffusion

Attempt 1

Parameters Used:

• Negative Prompt: Extra heads, extra faces, multiple faces, extra eyes, deformed face, disfigured, asymmetric, duplicate, tattoos, marks
• Steps: 80
• Sampler: DPM2
• Schedule type: Karras
• CFG scale: 13
• Seed: 88465
• Size: 512x768
• Model hash: c6bbc15e32
• Model: sd-v1-5-inpainting
• Conditional mask weight: 1.0
• Version: v1.9.4

Result:

Attempt 2

Parameters Used:

• Negative Prompt: Extra heads, extra faces, multiple faces, extra eyes, deformed face, disfigured, asymmetric, duplicate, tattoos, marks
• Steps: 122
• Sampler: DPM++ 2M
• Schedule type: Karras
• CFG scale: 7
• Seed: 1433
• Size: 512x768
• Model hash: c6bbc15e32
• Model: sd-v1-5-inpainting
• Conditional mask weight: 1.0
• Version: v1.9.4

Result:

Attempt 3

Parameters Used:

• Negative Prompt: Extra heads, extra faces, multiple faces, extra eyes, deformed face, disfigured, asymmetric, duplicate, tattoos, marks
• Steps: 66
• Sampler: Euler a
• Schedule type: Karras
• CFG scale: 19
• Seed: 5590
• Size: 512x768
• Model hash: c6bbc15e32
• Model: sd-v1-5-inpainting
• Conditional mask weight: 1.0
• Version: v1.9.4

Result:

Ideogram

Result:

FlexClip

Result:

ChatGPT (DALL·E)

Result:

Prompt 3

A young man in the city, late afternoon, standing, ((full length)), handsome, muscular, mexican american, latino, wearing revealing shorts, no jewelry, perfect hands, perfect face, the sun lights up his short black curly hair, sun light falls upon his hairy legs, show feet, Award - winning, portrait photograph shot with Kodak Portra 800, a Hasselblad 500C, 55mm f/ 1. 8 lens, extreme depth of field, available light, high contrast, Ultra HD, HDR, DTM, 8K

This prompt was taken from PromptHero.

Stable Diffusion

Attempt 1

Parameters Used:

• Negative Prompt: Extra heads, extra faces, multiple faces, extra eyes, deformed face, disfigured, asymmetric, duplicate, tattoos, marks
• Steps: 66
• Sampler: Euler a
• Schedule type: Karras
• CFG scale: 19
• Seed: 5590
• Size: 512x768
• Model hash: c6bbc15e32
• Model: sd-v1-5-inpainting
• Conditional mask weight: 1.0
• Version: v1.9.4

Result:

Ideogram

Result:

FlexClip

Result:

ChatGPT (DALL·E)

Result:

Comparison

Summary

Ideogram is best suited for artistic and stylized portraits, offering an easy-to-use interface.

FlexClip provides quick and simple AI-generated portraits with many preset templates / styles.

ChatGPT, through DALL·E, provides a fast and user-friendly way to generate AI portraits, making it accessible for casual users. However, it falls short in detail and customization, and has limitations in generating ultra-realistic human portraits compared to other tools and has stricter content moderation, censoring certain words that limit prompt flexibility. To achieve a professional-looking portrait, users should avoid overly specific descriptions of body details, as these may be restricted.

Stable Diffusion excels at generating highly detailed and customizable human portraits, but it has a steep learning curve and requires technical knowledge to use effectively. Despite its challenges, it remains completely free to use. If you want to learn more about how to use Stable Diffusion, please see this article.

Overall, each tool serves different needs, from professional-grade realism to casual and creative applications.

Comparison Table

Conclusion

Each AI tool has its strengths and is suited for different needs. If you want easy-to-use, highly realistic and detailed portraits, Ideogram and FlexClip are the best choices. If you prefer choices, control, privacy and no censorship, Stable Diffusion is the most suitable. Lastly, for casual and fast AI-generated portraits, ChatGPT with DALL·E is the most convenient.

Particularly, if you want to use Stable Diffusion, there are lots of work to do, such as fine-tuning, optimizing prompt, finding the right model, finding the right parameters, etc. It is not ideal but it gives you a lot of flexibility.

Which AI tool do you think creates the best human portrait? Let us know your thoughts!

Get in touch

For further information, I can be reach via:

• X (formally twitter): Eric Tang
• email: [email protected]