Is your organization falling behind with AI?

Back to Swarm Blog

Image Battle: When AI Makes the Art, Who Wins?

Tags:
SHARE THIS ARTICLE
JUMP AHEAD
SPREAD IT

In a few short years, artificial intelligence has evolved to a point where it can recognize patterns, draw, and even conjure images.

What began as an obscure experiment is now massively adopted as a mainstream creative force.

AI image generators like DALL·E and Midjourney now captivate the public with convincing visuals at the click of a button, sparking both delight and doubt about the future of art and the creative industry.

How We Got Here: A Timeline of AI Image Generation

AI art may seem like an overnight sensation, but its roots run deep.

An early milestone was Google’s DeepDream in 2015, which produced weird, dream-like visuals and revealed the “mind” of a neural network.1

Image by Google Research

In 2014, Ian Goodfellow and colleagues developed Generative Adversarial Networks (GANs), two dueling neural nets that could learn to create realistic images by training on real photos.2

GANs sparked a wave of AI art in the late 2010s, and a GAN-generated portrait even sold at a Christie’s auction.3

Portrait of Edmond de Belamy, 2018, Generative Adversarial Network print, on canvas. Image by Christie's.

But the true eruption came with large-scale text-to-image models.

OpenAI’s DALL·E (2021)4 and its successor DALL·E 2 (2022)5 showed that AI could draw anything described in words.

That same year, the open-source Stable Diffusion6 and the independent Midjourney7 put similar capacity into public hands.

Suddenly, anyone could type a phrase and get a custom image in seconds.

From experimental, fuzzy outputs a few years ago, we now have sophisticated AI tools delivering high-resolution art on demand.

ImageBattle.ai: Putting AI Models Head-to-Head

With a boom of image generators and wrappers, one challenge is figuring out which model is best for specific tasks.

Enter ImageBattle.ai, a vibe coding project by Philipp Kandal, Grab’s CTO and one of our community members at Swarm.8

Screenshot from ImageBattle.ai

This tool lets users directly compare multiple image models side by side using the same prompt.

Kandal created ImageBattle after finding that existing comparison sites gave scores but did not allow easy visual side-by-side judging.

The platform presents a gallery of outputs from different generators for each prompt so you can have Midjourney, DALL·E, Stable Diffusion, and others “battle” for your subjective favor.

Essentially an evolving snapshot of the state of AI art, this tool helps designers, product builders, and creators identify the optimal image generator for their needs.

ROI, Reach, and Render Speed: The Business Case for AI Visuals

Let’s look at some successful use cases across various industries.

Global payments network Klarna claimed a $6 million reduction in image production costs, using genAI tools like Midjourney, DALL·E, and Firefly for image generation. They also accelerated their image development cycle from 6 weeks to 7 days.9

The ability to cheaply produce endless variations of product imagery for different brand colors, contexts, or cultures is a game-changer for digital marketing.

In the video game industry, studios similarly use AI for concept art and assets.10

Generators can draft characters or environments much faster than humans, which is a huge time-saver, especially for smaller, independent developers.

Brand marketers and design agencies have embraced AI image generation for rapid content creation.

In 2022, Heinz released a campaign asking AI to “draw ketchup”, and DALL·E 2 produced images of ketchup bottles unmistakably resembling Heinz’s product.11

Image by Heinz

This stunt went viral and showed how generative AI can be used in branding: in this case, affirming Heinz’s ketchup is so iconic that even a computer knows it.

The best stories are how even solopreneurs and small businesses benefit from these tools.

A solo indie author can use Midjourney to create cover art, and a one-person startup can generate marketing visuals.

AI image generators have democratized visual content creation, letting individuals and small businesses produce graphics that can compete with better-resourced teams.

For bootstrapped creators and entrepreneurs, this newfound capability delivers real ROI in saved cost and time, if not in direct revenue.

Creation Without Consent: The Legal Fog Around AI Images

If an AI creates a new image that subtly remixes countless copyrighted pieces, is it a new creation or a copyright infringement?

With excitement over AI art comes serious concerns about ethics and authenticity. These image generation models are trained on millions of images scraped from the web, including copyrighted photos and artwork gathered without permission.

Many artists argue that it is theft, and in early 2023, a group of illustrators sued the makers of Stable Diffusion and Midjourney, accusing them of unlawfully using protected artwork to train their AI.12

Meanwhile, the U.S. Copyright Office has taken the stance that purely AI-generated images with no human editing cannot be copyrighted, because there is no human creator in the loop.13

What Happens When Anyone Can Fake Reality?

AI image generation has advanced so fast that images can be so realistic and easily mistaken for real photos, a problem that bad actors could exploit.

In 2023, an image of the late Pope Francis in a stylish white puffer coat went viral, with many convinced it was a real photo. It was a fabrication made with Midjourney, and commentators called it the first major instance of mass-scale AI image misinformation.14

Image by leon @skyferrori (X)

Similar fake visuals of public figures or events have popped up, sometimes causing panic before being debunked. As the tech improves, doctored images and videos will become harder to spot, potentially fueling propaganda or scams.

The Hidden Stereotypes in AI-Generated Imagery

As AI models learn from human-made images and text, they can reflect and amplify societal biases present in that data. For example, one study found that an AI tool often portrayed people in stereotyped ways: productive people are male, people cleaning are female.15

Images by Nita Shatiku (Threads)

This happens not out of malice, but because the models are mimicking the imbalances in their training material.

These developers are working on fixes, such as fine-tuning models or filtering training data, but users of AI images should also stay vigilant against AI unwittingly reinforcing stereotypes and offensive tropes.

Humans Optional? How AI Is Rewriting Creative Work

Graphic designers, illustrators, photographers, and other people working in the creative industry are watching AI’s rapid progress.

Concerns of creative job displacement are not unfounded: in China, a gaming studio laid off part of its art team and now relies on AI for illustrations (keeping only a few staff to polish the AI outputs).16

Icon, claiming to be the world’s first AI Chief Marketing Officer, plans, creates, and runs thousands of ads per week. It can study ad campaigns from websites, competitors, and customer reviews, and ad account performance.17

The prospect of companies using AI in place of creative humans raises tough questions.

Optimists argue that AI will create new opportunities and that human originality will still stand out. But in the short term, it’s clear that generative AI is already reshaping the creative job market, and professionals and policymakers alike are scrambling to keep up.

As AI continues to affect creative industries, some efforts are being made to guide its responsible use.

OpenAI Launches Image Generation Guide

In April 2025, OpenAI launched GPT-Image-1, its most advanced image generation model, now accessible through OpenAI’s API.

To support users in using GPT-Image-1 to its fullest potential, OpenAI released a comprehensive Image Generation Guide.18

This guide provides detailed instructions on using the model’s capabilities, including generating diverse styles, adhering to custom guidelines, and accurately rendering text within images.

Image Provenance: The Movement for Creative Ownership

There are ongoing industry efforts around image provenance, building technologies that tag AI-generated images to verify their origin.

The Coalition for Content Provenance and Authenticity (C2PA)19 and Adobe’s Content Credentials20 embed digital ‘watermarks’ into AI-generated images, helping creators and audiences verify authenticity.

The Artist Strikes Back: New Tools to Protect Art from AI Training

As awareness grows about how AI models are trained on massive datasets—often without artists’ consent—new technologies and initiatives are emerging to protect creative work.

  • Glaze: Developed by the University of Chicago’s SAND Lab, Glaze applies subtle alterations to artworks, invisible to the human eye but effective at confusing AI models.This technique helps prevent AI from learning and replicating an artist’s distinctive style without permission.21
  • Nightshade: Also from the SAND Lab, Nightshade takes a more aggressive approach by ‘poisoning’ training data, embedding tiny distortions that, if included in a model’s dataset, could cause AI outputs to behave unpredictably when trying to replicate protected art.22
  • Have I Been Trained?: This tool allows artists to search popular AI training datasets and check if their work has been included. If found, they can submit opt-out requests to have their work removed from future training sets.23
  • Cara: A rising social networking platform for artists, Cara automatically applies “NoAI” tags to uploaded artworks, signaling that the content should not be used in AI model training. Cara also integrates Glaze for additional protection.24

The Swarm Method for AI: Guidelines for Responsible AI Image Use

As an AI-native company, Swarm is excited about generative image tech, but we also set ground rules to use it responsibly.

Here are a few principles we follow aligned with our company values:

  • Commitment to craft excellence by provenance. At Swarm, we place a premium on the value and ownership of work. We respect human creators and intellectual property.
    • We use image models that have clear usage rights or licenses.
    • We avoid directly mimicking a living artist’s style without permission. If we ever use someone’s original art or likeness in AI, we make sure we have proper rights.
    • If an image is AI-generated, we label it clearly so viewers are not misled.
  • Enabling human collaboration, not replacement. We use AI image generation as a tool, not as a way to cut people out of the creation process.
    • AI image generation is a starting point. Designers refine and review AI-generated assets to ensure they meet our quality standards and brand values.
    • We actively check AI outputs for bias or inappropriate elements before using them.
    • We never use AI images to mislead, and we avoid generating harmful or disallowed content.
  • Embracing ambiguity by staying ahead of the curve. At Swarm, we lean into curiosity and a spirit of learning when we are faced with new tools.
    • We believe in helping our community use tools that will enable them to achieve their goals. This means we dabble in new technologies and share how they could be helpful to our consultants.

These guidelines help our team stay on top of the latest advancements while keeping our focus fixed on what matters most: awesome humans doing impactful work.

Endnotes

  1. Mordvintsev, Alexander et al. (2015). "Inceptionism: Going Deeper into Neural Networks." Google AI Blog. https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html
  2. Goodfellow, Ian et al. (2014). "Generative Adversarial Nets." NeurIPS Conference. https://arxiv.org/pdf/1406.2661
  3. Sutton, Benjamin. (2018). “The first portrait generated by artificial intelligence to be sold at auction was purchased for more than 40 times its high estimate.”https://www.artsy.net/article/artsy-editorial-first-ever-portrait-generated-artificial-intelligence-auction-sold-40-times-estimate
  4. OpenAI. (2021). "DALL·E: Creating Images from Text." OpenAI Blog. https://openai.com/blog/dall-e
  5. OpenAI. (2022). "DALL·E 2." OpenAI Blog. https://openai.com/dall-e-2
  6. Stability AI. (2022). "Stable Diffusion v1 Release." Stability AI Blog. https://stability.ai/news/stable-diffusion-public-release
  7. Rose, Janus. (2022). “Inside Midjourney, The Generative Art AI That Rivals DALL-E.” Vice. https://www.vice.com/en/article/inside-midjourney-the-generative-art-ai-that-rivals-dall-e/
  8. Kandal, Philipp. (2024). "About ImageBattle.ai." ImageBattle.ai. https://www.imagebattle.ai/about
  9. Klarna. (2024). “AI helps Klarna cut marketing agency spend by 25% and run more campaigns.” https://www.klarna.com/international/press/ai-helps-klarna-cut-marketing-agency-spend-by-25-and-run-more-campaigns
  10. Takahashi, Dean. (2024). "How Invoke Uses AI for Ethical Image Generation in Gaming." VentureBeat. https://venturebeat.com/ai/how-invoke-uses-ai-to-power-ethical-image-generation-for-games-kent-keirsey-interview/
  11. Coggan, Georgia. (2022). "Heinz Asked AI to Draw Ketchup – and It Worked." Creative Bloq. https://www.creativebloq.com/news/heinz-ai-draw-ketchup
  12. Xiang, Chloe. (2023). “Artists Are Suing Over Stable Diffusion Stealing Their Work for AI Art.” Vice. https://www.vice.com/en/article/artists-are-suing-over-stable-diffusion-stealing-their-work-for-ai-art/
  13. Recker, Jane. (2023). "U.S. Copyright Office Says AI Art Can't Be Copyrighted." Smithsonian Magazine. https://www.smithsonianmag.com/smart-news/us-copyright-office-rules-ai-art-cant-be-copyrighted-180979808/
  14. Golby, Joel. (2023). "I thought I was immune to being fooled online. Then I saw the pope in a coat." The Guardian. https://www.theguardian.com/commentisfree/2023/mar/27/pope-coat-ai-image-baby-boomers
  15. Tiku, Nitasha. (2023). "These fake images reveal how AI amplifies our worst stereotypes." The Washington Post. https://www.washingtonpost.com/technology/2023/11/01/ai-bias-stereotypes-photos/
  16. Zhou, Viola. (2023). "AI Already Taking Illustrator Jobs in China." Rest of World. https://restofworld.org/2023/ai-china-video-game-layoffs-illustrators/
  17. Davison, Kenna. (2025). https://www.linkedin.com/posts/kennandavison_introducing-icon-the-worlds-first-ai-cmo-activity-7321217146494017537-G4DC
  18. OpenAI. (2025). "Introducing GPT-Image-1 and the Image Generation Guide." OpenAI Blog. https://openai.com/index/image-generation-api/
  19. Coalition for Content Provenance and Authenticity (C2PA). (2023). "About C2PA." C2PA Official Website. https://c2pa.org/
  20. Parsons, Andy. (2025). “Adobe Content Authenticity, now in public beta, helps creators secure attribution.” https://blog.adobe.com/en/publish/2025/04/24/adobe-content-authenticity-now-public-beta-helps-creators-secure-attribution
  21. Shan, Shawn et al. (2023). "Glaze: Protecting Artists from Style Mimicry by AI." University of Chicago. https://people.cs.uchicago.edu/~ravenben/publications/pdf/glaze-usenix23.pdf
  22. Shan, Shawn et al. (2024) "Nightshade: Prompt-Specific Poisoning Attacks on Text-to-Image Generative Models.” https://people.cs.uchicago.edu/~ravenben/publications/abstracts/nightshade-oakland24.html
  23. Have I Been Trained? (2023). "Search Your Artwork in AI Training Datasets." HaveIBeenTrained.com. https://haveibeentrained.com/
  24. Cara. (2024). "Cara: Social Network for Artists, Protecting Against AI Scraping." Cara App Official Site. https://cara.app/

Head image by Lennon Villanueva. Mixed media collage using the original Abbey Road album photograph and AI-generated images sourced from public online platforms.

Pia Besmonte Ligot-Gordon
Head of Editorial
Pia is a published author and educator who serves as Swarm's Community and Growth Lead, where she crafts content and community experiences that inform and empower fractional tech builders in a rapidly evolving landscape.
AI
Agents
Digital Transformation
SHARE THIS ARTICLE
SWARM INTELLIGENCE