AI Image Generation with DALL-E vs Gemini vs Stable Diffusion

A picture worth a thousand words is an adage applicable even today. Complex and multiple ideas can be portrayed in a single image. The trend is shifting from texts to pictures and motion graphics. Researchers say that images are very powerful and most choose images to understand the message because they are the entry point to stories – they add meaning, and they remarkably touch people’s psychological state, memory, and emotions. While we have larger MLs and LLMs, the popularity of image-generating AI tools has skyrocketed in recent times.

Midjoruney, Dall-E, Google Gemini, and Stable Diffusion (Stability) are a few image-generation software tools popular all over the Internet. While Midjourney has led the race so far, we thought that there is a grave need to compare the rest of the AI image generation tools – Dall -E vs. Google Gemini, vs. Stable Diffusion.

Let’s have a quick introduction before we compare AI image-generation software tools.

What is DALL–E?

DALL–E is an AI model that generates images or illustrations based on textual descriptions users put as a prompt. To build an image in line with the text, it translates billions of text chunks from all over the Internet into an abstract. This stored information is then utilized as a reference tool to describe information and finally to create prompt-oriented images. The DALL–E model is available through ChatGPT.

What is Google Gemini?

Google introduced the AI image generator tool through Gemini in 2024. Gemini is Google’s main suite of AI models, and it was equipped to produce images as per the prompts by users. Though Google Gemini is more in the news due to its historical inaccuracies and questionable responses, the AI image generator Gemini is known for providing illustrations/images that are very close to the user’s imagination.

What is Stable Diffusion?

Stable Diffusion is by Stability AI, a leading open-source generative AI (GenAI) company that aims to deliver breakthrough and open-access AI models that require minimal resources to build images, language, audio, and code. Stable Diffusion is the latest and most advanced T2I (Text-to-Image) that comprises 2 billion parameters.

AI Image Generation Tools: The Rising Popularity and Impact

Artificial Intelligence has made image-generating tools more efficient and accurate to the prompts. AI image generators are immensely popular among marketers and content creators to boost their content with eye-catching and engaging graphics.

Statistics reveal that just under 40% of marketers use Generative AI to create images for social media posts. Furthermore, 36% of them harness the power of AI image generators to build website images.

DALL–E vs Google Gemini vs Stable Diffusion – Comparison of AI Image Generators

To compare image generator AI tools, we decided to have a common prompt to run on these three different platforms. The purpose was to understand how these AI image generator tools pursue the texts and use their algorithms and models to build images.

We considered three general parameters to evaluate the AI image tools –

1. How well does the AI generator tool understand the prompt with details?
2. How much response time does it take to generate text-to-image results?
3. How was the main image created and what was put in the surroundings and background?

Experiment with Prompt #1 for AI Image generation

Create an image of an ornate, Victorian-era key lying on a weathered, wooden surface, with intricate, steampunk-inspired gears and mechanisms visible within its transparent, glass shaft.

The parameters

How well does the AI generator tool understand the prompt and detail it?

DALL – E (ChatGPT)

DALL -E understood the prompt the way we wanted. It precisely created exactly what we had thought of. The AI image generator quite well understood the instructions: inspired gears and mechanisms visible within its transparent, glass shaft. In fact, it also created a transparent mechanism in the shaft along with the head.

DALL-E perfectly created a Victorian-era key. The AI image generator also crafted an eye-catching weathered wooden surface. The copperish color was a perfect match with the prompt and the image we had created in our minds. Overall, it was a detailed image with sharp features and a royal design of the key.

What we loved

• Natural light created on the glass
• The wood
• The crescent of the glass at the shadow side

Google Gemini

Unlike DALL–E and Stable Diffusion, Google Gemini always provided multiple options. Each image would have a slightly different approach which means the user can have different choices from the prompt. However, in this case, the three keys were not impressive as they missed an essential part of the ‘transparent mechanism’ in the head. The AI image creator produced one key that suited our description; however, we did not find it impressive.

Here, only one key with the key head glass was up to the mark. The angle of the key was such that the detailing in the transparent mechanism wasn’t that visible. Overall, the AI image generator did not do a good job for us for this.

Stable Diffusion

Stable Diffusion made it all perfect, the way we wanted. It had a transparent mechanism and glass shaft. The key looked royal, but we were expecting the glass cover in the key’s head which it missed. Overall, the detailing and prompt-centered image was quite satisfactory.

Stable Diffusion did a marvellous job. The image had a bigger key with all the details visible. Though the transparent mechanism wasn’t that impressive, the AI generator tool did understand the prompt and produced the relevant illustration. Nevertheless, we observed that it missed a major part – the glass cover on the key head.

What we loved:

• The close-up of the key
• The Victorian design

Point counts:

• DALL – E: 1
• Google Gemini: 0
• Stable Diffusion: 0.5

How much response time does it take to generate text-to-image results?

The AI image generation tool, DALL-E took between 6-8 seconds. On the other hand, Google Gemini image AI-generating software took between 7 and 9 seconds to understand and create the image. Lastly, Stable Diffusion was quite fast as this image-generating AI tool took around 5 to 7 seconds to craft the prompt-cantered image.

Point counts:

• DALL-E: 1
• Google Gemini: 1
• Stable Diffusion: 1

How was the main image created and what was put in the surroundings and background?

DALL -E

The main image created was as per the prompt. There was nothing extra which means the AI image generator software followed the instructions strictly without adding anything that was not asked.

Google Gemini

Google Gemini AI generator provided a few options for the same prompt and that made this AI image-generating tool a bit more comprehensive and holistic. While three images created a wood, one image was presented with natural grass which caught our eyes. It looked soothing but then we did not ask for it.

Stable Diffusion

Like DALL-E, Stable Diffusion did a stupendous job by creating the main image and surroundings as per the prompt. The wood was exactly the way we wanted. It did not add anything extra that we did not ask for.

Point Counts:

• DALL-E: 1
• Google Gemini: 1
• Stable Diffusion: 1

Experiment with Prompt #2

Cinematic film still, close-up, photo of a gold-scaled dragon warrior in full plate armor, in a hyper-realistic fantasy style.

The parameters

How well does the AI generator tool understand the prompt?

DALL – E

The ChatGPT did a fantastic job by showing us the close-up of the dragon warrior. The colors used were vibrant and the gold-scaled creature looked impressive. The spikes on the body and armor were detailed and sharp. Nevertheless, what we observed was that this AI image generator took our prompt quite literally! The ‘cinematic film still’ phrase triggered DALL-E to create the clapperboard and show it in the picture. While it is impressive that the AI engine takes prompts seriously, we expected it to be understood as a cinematic scene rather than focusing on BTS!

Google Gemini

Google Gemini simply changed the armor suit with different designs for three of its results. Each one had a different zoomed-in/out approach. The gold-scaled dragon warrior’s face remained almost similar. The fourth image showcased a larger pose of the warrior with more elements in the background. All the images looked a bit dull compared to the rest of the tools.

Stable Diffusion

A single shot with a detailed dragon warrior was displayed. Impressively, the dragon’s head was blended with sallet and bevor to look more aggressive. The blue diamond went pretty well with a gold-plated pauldron and cuirass. The blue warrior robe looked attractive on the body.

What we loved:

• The close-up of the warrior
• Diamonds
• Warrior robe

Point count:

• DALL – E: 0.5
• Google Gemini: 0
• Stable Diffusion: 1

How much response time does it take to generate text-to-image results?

The AI image generator DALL-E took between 6-9 seconds. On the other hand, Google Gemini image AI-generating software took between 8 and 10 seconds to understand and create the image. Lastly, Stable Diffusion was quite fast as this image-generating AI tool took around 5 to 8 seconds to craft the prompt-cantered image.

Point counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 1

How was the main image created and what was put in the surroundings and background?

DALL -E

The main image created was as per the prompt. However, a noteworthy addition was the clapperboard as the AI engine took the prompt quite seriously. The phrase ‘cinematic’ was taken too literally and the AI image generator thought to include behind the scenes.

Google Gemini

Google Gemini AI generator showed four options for the same prompt and that made this AI image-generating tool a bit more comprehensive and holistic. Nevertheless, three of those images were almost the same with minor design changes in the armor suite. The fourth image was a zoomed-out picture of the warrior with a planet. The tool tried to show the battleground, but it looked more as an extraterrestrial region.

Stable Diffusion

Stable Diffusion was, so far, the best in creating the gold-scaled dragon warrior with a detailed close-up. The background showed a castle that fulfilled our purpose of having a war-like flair. Overall, this image was something we wanted through the prompt.

Point Counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 1

Experiment with Prompt #3

Create vibrant, explosive swirls of orange, yellow, pink, and blue paint cascade from the ceiling onto a polished grey floor in an art gallery, contrasting with monochromatic abstract paintings on white walls and creating a dynamic, energetic scene under bright, focused lighting.

The parameters

How well does the AI generator tool understand the prompt?

Google Gemini

For this time, Google Gemini went a bit advanced and created four different options rather than presenting similar images with trivial changes as in the past. However, the AI image generator missed the prompt’s instructions in two of its images. The tool showed a straight fall of orange, pink, and yellow paint but missed two essential things: swirl and the color blue. However, the rest of the two images did include the colors (including blue) mentioned in the prompt. But again, while including the blue color in two images, the tool could not focus on having monochromatic abstract paintings. Overall, each of the images created missed some or the other thing from the prompt. The upper right image did not show the floor and the depth of the image; instead, it appeared like the colors were oozing out of nowhere.

DALL-E (ChatGPT)

This tool created the most impressive image out of the prompt. Beautiful swirls of all the colors mentioned in the prompt were created. Not only that, the AI image generator won our hearts by showing the floor with spherical color balls. The tool did manage to show monochromatic abstract paintings on the white walls. This was the only tool that included bright and focused lighting in the image as instructed in the prompt.

Stable Diffusion

Stable Diffusion was quick to understand the prompt, however it failed to include blue paint with equal weightage as the rest of the colors. There was just a fraction of the blue color. The AI image generator also missed the monochromatic painting but managed to show the gray floor. It also failed to create bright and focused lighting which was a part of the prompt.

Point count:

• DALL – E: 1
• Google Gemini: 0
• Stable Diffusion: 0.5

How much response time does it take to generate text-to-image results?

Google Gemini created four options and took around 4-6 seconds in the first attempt. Similarly, one attempt was enough for the DALL-E AI image generator to create an image in 6-8 seconds. Lastly, Stable Diffusion took around 5-8 seconds to create the image out of the prompt in the first attempt.

Point counts:

• DALL-E: 1
• Google Gemini: 1
• Stable Diffusion: 1

How was the main image created and what was put in the surroundings and background?

Google Gemini

The tool not only failed in going literally for the prompt but also created the background with its own. The focused lighting was shown on the paintings on the wall whereas the prompt mentioned that it should be on the swirls of paints. Likewise, the two images below missed the paintings on the wall completely. Only the first image (left upper corner) came a bit close to the prompt but then it failed to create attractive scenes and monochromatic paintings.

DALL-E

This was the most impressive image generation from the prompt. ChatGPT created mind-boggling swirls coming from the top and smashing onto the gray floor. What we loved were the windows that showed natural ambient light focusing on the fall. Plus, the AI image generator did manage to include monochromatic paintings on the white walls.

What we loved:

1. Swirls
2. The overall image details
3. The windows and ambient lights
4. The colored balls on the floor
5. The depth of the field with paintings
6. The ceiling

Stable Diffusion

The proportion of the swirls from the ceiling and the rest of the room did not match and hence it looked way too artificial. Also, the tool failed to create a typical monochromatic painting because both paintings have a trace of some other colors/shades. The tool managed to get the gray floor and ceiling but overall, the picture did not match our expectations though it was better than Google Gemini.

Point Counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 0.5

Experiment with Prompt #4

In the shadow of the last sun, a fisherman had fallen asleep, and he had a furrow along his face, like a sort of smile.

The parameters

How well does the AI generator tool understand the prompt?

Google Gemini

We tried with very short prompt and straightforward instructions despite that Google Gemini completely failed to parse through the instructions and generated quite irrelevant images. All the images did not show the fisherman’s face and hence there was no chance to evaluate the furrow and the smile. Two of the images were more like a silhouette with no detailing. The fourth image was completely off the track showing a fisherman sleeping in the boat with no trace of the last sun. One of the images focused more on the canoe and the sea with a tiny fisherman.

DALL-E

ChatGPT chose to show us a close-up of a fisherman. The tool did manage the last sun and the man sleeping peacefully. The furrow created along his face was impressive and a little smile created out of it was noteworthy.

What we loved:

1. The last sun
2. The sun rays
3. The shades and shadows on the fisherman
4. The fabric and hat of the man

Stable Diffusion

Stable Diffusion did a way better job than Google Gemini; however, it was no match with the image created by the DALL-E AI image generation tool. The image showed a fisherman lying in his canoe and there was the last sun on the horizon. The picture shows that he is sailing in the middle of the ocean. However, due to the silhouette-type image, the furrow and the smile created were not visible.

Point Counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 0

How much response time does it take to generate text-to-image results?

DALL-E ChatGPT took 4-6 seconds to create the best image that aligned with the prompt in one attempt. Stable Diffusion spent 5-7 seconds to understand the prompt and create the image of a not-so-detailed fisherman. Google Gemini not only took 8-10 seconds (the lengthiest time) but also disappointed with the outcomes.

Point counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 0

How was the main image created and what was put in the surroundings and background?

Google Gemini

Google Gemini created a dull background with no detailing. The most surprising was the downright corner image where the setting sun was omitted and replaced by a fairy-tale scene of a house, canoe, fisherman, and flowers in the front yard. The rest of the images did have seas in the background with some sort of light from the setting sun but none of them were impressive. The images had more darkness where detailing of the fisherman was not feasible.

DALL-E

This was the most impressive AI image generation from the prompt. ChatGPT followed the instructions to the length and breadth. A sharp and clear picture of a fisherman with a furrow was significant. The background showed the setting sun with enough ambient light to highlight the silhouette of canoe and fishing nets. Also, the ambient light created on the face of the fisherman was impressive as it clearly defined the creases on the face and the fine fabric of the man’s shirt.

What we loved:

1. Close-up of the fisherman
2. The overall flair of dusk
3. Sharp features on the face
4. The furrow and smile it made
5. The shirt’s fabric and detailing
6. The expression on the face

Stable Diffusion

Stable Diffusion produced a nice sky with the setting sun on the horizon. The water showed a reflection of the sunlight. However, the size of the fisherman and his canoe could have been a bit better to look natural. The silhouette effects hampered the purpose of the prompt wherein a furrow and a smile created out of it were missing completely. The tool did not miss showing a fishing rod dropping out of the canoe. Overall, the image was nice, but it missed the purpose.

Point Counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 0.5

Experiment with Prompt #5

Anime girl, girl knight, blunt bangs, hime cut, pointy ears, pearl opal, very aesthetic, masterpiece, best quality, hyper-detailed, ultra-detailed, UHD, perfect anatomy, sword, dazzling, transparent, waving sword, burnished silver, steel armor, shining armor, dazzling armor, detailed Illustration, official artwork, wallpaper, official art, extremely detailed eyes and face, beautiful detailed eyes, blue eye.

The parameters

How well does the AI generator tool understand the prompt?

Stable Diffusion

Stable Diffusion neatly produced the girl knight with a detailed view of almost everything that the prompt mentioned such as sharp ears, ultra-detailed, sword, steel and shining armor, and blue eyes. It seems the AI image generator tool has not missed a single piece of instruction. We actually asked for too much and tried to confuse the AI engine by asking the same thing in different ways and avatars. However, Stable Diffusion did its job. The shot it produced was a close-up of a blond girl who looked like a confident knight.

Google Gemini

Google Gemini first produced only two images, unlike its model which gives us four options. It stopped after generating two images and asked us to ‘Generate More’. However, if we talk about the first two AI-generated images, they show the girl knight from the front and in a portrait pose. One image showed the sword (that did not appear like one thought, it was more like a Star Wars beacon), and the other image missed it. Surprisingly, the AI image generator tool Google Gemini showed one image in which the girl had horns. Both the images missed pointy ears.

The first attempt could not produce all four images.

In the second attempt, Google Gemini gave up and prompted a message: I can’t generate these images. Enter a new prompt to generate more images.

DALL-E (ChatGPT)

This model of AI image generator produced an image that was straight from the prompt. However, to our surprise, it showed the image vertically. When corrected locally, the image was in landscape mode. The image showed a girl knight but it was more of a childish character. It did produce a shining sword. DALL-E seems to love diamonds and that is what was shown in this image as well. The girl wore armour and had pointy ears as per the prompt. She was dressed nicely with a flower in the crown.

Point Counts:

• DALL-E: 0.5
• Google Gemini: 0
• Stable Diffusion: 1

How much response time does it take to generate text-to-image results?

DALL-E for its ChatGPT work took 4-6 seconds to produce a landscape image of the girl knight whereas Google Gemini took two attempts (exceeding 10 seconds in total) to produce two images and that too were not up to the mark. Stable Diffusion took 6-8 seconds to produce a near-to-perfect image from the prompt.

Point Counts:

• DALL-E: 1
• Google Gemini: 0
• Stable Diffusion: 1

How was the main image created and what was put in the surroundings and background?

Stable Diffusion

The main image created by the AI image-generating tool Stable Diffusion was up to the mark as it showed everything in detail and as per the prompt. The background was a beautiful sky with clouds that matched with the main character’s shades. The shadow and reflection of the sky and the sunlight were visible on the girl knight’s armor.

Google Gemini

It was a plain background with no detailing for the girl knight. The sword looked more like a laser beacon and the character looked more like a beast with horns on the head. The tool also missed sharp ears and attractive blue eyes. There was nothing that attracted us in the image.

DALL-E

The detailing in this picture was dramatic and not as real as we found in Stable Diffusion. It was more cartoonish in DALL-E’s case. The girl looked Chinese and kid rather than a knight with boldness. The image showed graphical diamonds that appeared more like lens flairs in some places. We also observed that it was a tile image where you can find two more copies of the main characters in the background. We are not sure what it means because it was not in the prompt.

Point Counts:

• DALL-E: 0.5
• Google Gemini: 0
• Stable Diffusion: 1

The Total Counts: DALL-E vs Gemini vs Stable Diffusion

If we look at the point counts for all the questions for all the AI image generators, we get this score:

Prompt 1

Prompt 2

Prompt 3

Prompt 4

Prompt 5

The Final Results: DALL-E vs Gemini vs Stable Diffusion

In our comprehensive testing of AI image-generating tools, DALL-E, Google Gemini, and Stable Diffusion were evaluated using the same prompt. DALL-E emerged as the top performer, scoring 13.5 out of 15, excelling in image detailing, adherence to prompt instructions, and the quality of backgrounds.

Stable Diffusion followed with a score of 11, demonstrating strong image quality and precision, though slightly less consistent in background elements. Google Gemini, scoring 3, stood nowhere in accurately following prompt instructions and producing detailed images.

Verdict

DALL-E stands out as the most reliable tool for generating high-quality, detailed images that align closely with given prompts. However, our findings are based on specific testing parameters and individual experiences may vary. We recommend exploring each AI tool to determine which one best suits your unique needs.

Social Hashtags

#AIImageGeneration #DALLEvsGemini #StableDiffusionAI #AIArtBattle #AIComparison #ImageGenerationTech #AIImageCreators #CreativeAI