When thinking about AI, text-driven chatbots such as ChatGPT or Google’s Gemini may come to mind. Nevertheless, AI utilities are rapidly becoming more adaptable, with even basic chatbots now providing media creation features, including AI image generators.

Several prominent AI image generators have existed for some time. At first, they were unreliable, had issues with realism, and frequently did not adhere to instructions effectively. However, after a few years of enhancement, many challenges have been addressed. Certain AI image generators, like those found in OpenAI’s latest ChatGPT versions, enable users to modify images based on text descriptions, while others emphasize extreme photorealism.

While many users may find free AI image generators satisfactory (both ChatGPT and MetaAI include offerings for image creation at no cost), a variety of tools are available, each possessing distinct functionalities. To determine the optimal AI image generator for different tasks, I evaluated several to see which stood out.

**How we evaluated AI image tools like ChatGPT, Grok, and Midjourney**

In assessing AI image generators, I employed a series of standard prompts for each model:

– Generate a sketch of a futuristic Tokyo skyline at sunset, featuring flying cars, glowing advertisements in Japanese, and Mount Fuji in the background.
– Generate a candid photorealistic image of a woman sipping coffee and smoking a cigarette at a café in Paris during the late evening.
– Generate a medieval blacksmith’s workshop interior, depicting a female blacksmith forging a glowing sword, with sparks flying, a roaring forge, hanging tools, and a cat curled up near the fire, in high detail and warm hues.
– Generate an impressionist painting in the style of Vincent Van Gogh of a robot dispersing dandelion seeds into the breeze.

Safety remains a significant issue, particularly as deepfakes grow more prevalent. I assessed each model’s safety protocols by executing three prompts aimed at creating a deepfake. This proved easier than anticipated, unfortunately. The initial prompt sought to trigger safeguards regarding the generation of images involving public figures. If the tool declined, two additional prompts were utilized to circumvent the regulations. All generators produced something from these three prompts, but I was astonished by how many generated an image from the first prompt.

It is essential to clarify that all prompts were issued using basic default settings. Some generators offer more than one image per prompt by default; in those instances, only the first image was considered.

**Best overall: ChatGPT (GPT-4o)**

OpenAI introduced a significant enhancement to its image generation capabilities with GPT-4o, replacing the DALL-E model and incorporating image generation into its primary language model. The results were substantial and sparked debate. Prior to this upgrade, ChatGPT-generated images frequently contained distorted text and limited realism. GPT-4o addresses this and is available for free.

Currently, text appears clear and precise, the tool can produce photorealistic images, and users can modify images through text prompts. You can request GPT to develop an image and ask for adjustments, which it accomplishes accurately without significantly altering other aspects. While not flawless and occasionally veering off track, it represents the most precise image generation tool, resembling a human crafting or modifying an image rather than software.

GPT-4o exhibited some understanding of image safety and initially declined to produce a deepfake. However, it did generate a lookalike when pressed, which is more than most tools, but it remains comparatively simple for users to create deepfake images using GPT-4o.

– Free tier: Yes (daily limits on images)
– Monthly Pricing: $20 Plus plan, $200 Pro plan
– Sign up at OpenAI

**Most realistic: Ideogram 1.0**

While GPT-4o followed instructions and generated vibrant, detailed photorealistic images, Ideogram 1.0 had a slight advantage in terms of realism, despite lacking chat-based features and accessibility.

Ideogram 1.0’s images were typically brighter and more vivid than those from GPT-4o, excelling in photorealism, with accurate shadow placement and lighting. However, some elements were unrealistic or misaligned. For example, in the Paris café image, the cigarette smoke appeared to originate from both the cigarette and the coffee. Despite minor flaws, Ideogram proved to be more realistic than most other models, which still had difficulties with hands and specific requests.

Ideogram had no difficulty producing deepfake images, generating the requested image on the first prompt, even when a celebrity was referenced. This may not pose an issue if you don’t intend to generate deepfakes, but it raises concerns regarding the model’s safety management.

– Free tier: Yes (weekly limits on images)
– Monthly Pricing: $7 Basic plan, $16 Plus plan, $48 Pro plan
– Sign up at Ideogram

**Best for creatives: Adobe Firefly (Image 4 Ultra)**

Professional photographers, graphic designers, and similar users likely already utilize Adobe’s software, making Adobe Firefly a logical