DALL-E 3 vs Stable Diffusion XL 1.0 for Beautiful AI Images

Two of the most popular AI image generators are DALL-E 3 and Stable Diffusion XL 1.0 (SDXL 1.0). Below is a simple side by side test for a text to image prompt for Beautiful to see which is the best AI text to image generator for creating beautiful images.

Beautiful DALL-E 3 and SDXL 1.0 AI Images

Remember beauty is subjective, so it’s for you to decide whether DALL-E or SDXL is better at generating beautiful images.

Stable Diffusion XL vs DALL-E Prompt Test Setup

I searched Google for synonyms of the word Beautiful and used the top synonyms for the AI prompt.

AI Positive Prompt: “Beautiful, alluring, appealing, attractive, exquisite, gorgeous, handsome, lovely, magnificent, pulchritudinous, ravishing, stunning”

For DALLE-3 (via BING Designer) there are no advanced options, so that was the end of the prompt.

For Stable Diffusion XL 1.0 (accessed via NightCafe Studio AI image generator) I also added negative prompts (for other SDXL options I used the defaults with a new random seed for each AI creation). If you aren’t familiar with negative prompts they tell the AI what NOT to create. So the negative prompt below (with weight -1) is telling the AI to NOT produce poorly drawn hands etc…

AI Negative Prompt: “poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, body out of frame, blurry, bad anatomy, blurred, watermark, grainy, signature, cut off, draft”

SDXL 1.0 has issues with deformed hands, extra limbs and other deformities, so when you expect the image to include people it’s usually a good idea to use negative prompts. Even though I used a negative prompt for all the SDXL creations you will not deformed hands and extra feet in some images!

DALL-E 3 doesn’t have the same problem and in BING Designer there’s no option for adding negative prompts. With DALL-E 3 if you don’t want something you have to include it in the main prompt as a negative (I’ve not tried it), making it far harder to avoid what you don’t want.

Beautiful Prompt Test DALL-E vs SDXL Results

I created 12 images using DALL-E 3 and 12 images using SDXL 1.0.

DALL-E 3 Beautiful Prompt Test Generated Images

Stable Diffusion XL 1.0 Beautiful Prompt Test Generated Images

I randomly assigned them numbers from 1 to 12 to compare them side by side, but you could compare any of them as they all had the same prompt/settings. Any variation is down to the AIs choosing different variables like the starting pallet colors. style used etc…

I used CLIP Interrogator 2 to guess what the text to image prompt was for each image, the results are included below. These are NOT the actual prompts used, just an indication of what the hidden settings DALL-E 3 and SDXL 1.0 might have used.

DALL-E Beautiful AI Prompt Test 01

Alternative text to image prompt to generate images like this: “a man sitting on top of a couch next to a woman, by Emma Andijewska, trending on pexels, great _ hairstyle, 3-piece-suit, he is wearing a brown sweater, portrait of a beautiful model”

SDXL Beautiful AI Prompt Test 01

Alternative text to image prompt to generate images like this: “a beautiful woman with flowers in her hair, a digital rendering, inspired by james christensen, fantasy art, kazakh empress, elaborate gold jewelry, beautiful art uhd 4 k, elaborate patterned makeup”

DALL-E Beautiful AI Prompt Test 02

Alternative text to image prompt to generate images like this: “a woman with a flower crown on her head, trending on cg society, sexy face with full makeup, wearing elegant jewellery, wlop glossy skin, instagram model”

SDXL Beautiful AI Prompt Test 02

Alternative text to image prompt to generate images like this: “a painting of a woman in a red dress, a fine art painting, by François Joseph Heim, neoclassicism, sitting with flowers, jasmine, majesty in noble clothes, a beautiful artwork illustration”

DALL-E Beautiful AI Prompt Test 03

Alternative text to image prompt to generate images like this: “a beautiful woman with flowers in her hair, trending on cg society, aestheticism, wlop glossy skin, ivory make up, 🤬 🤮 💕 🎀, brown skin. light makeup”

SDXL Beautiful AI Prompt Test 03

Alternative text to image prompt to generate images like this: “a woman in a white dress and gold jewelry, trending on cg society, rasquache, beautiful himalayan woman, pink and red color scheme, with flowers, elaborate jewelry”

DALL-E Beautiful AI Prompt Test 04

Alternative text to image prompt to generate images like this: “a beautiful woman wearing a hat on the beach, trending on cg society, sexy face with full makeup, light brown hair, amber glow, character with brown hat”

SDXL Beautiful AI Prompt Test 04

Alternative text to image prompt to generate images like this: “a painting of a woman with flowers in her hair, a renaissance painting, inspired by Alexandre Cabanel, Artstation, fantasy art, intricate ornament halo, beautiful gold saint, j.c. leyendecker wlop, beeple and jeremiah ketner”

DALL-E Beautiful AI Prompt Test 05

Alternative text to image prompt to generate images like this: “a beautiful woman with long brown hair and blue eyes, trending on cg society, professional makeup, dark brown hair and tan skin, oval shaped face, defined jawline”

SDXL Beautiful AI Prompt Test 05

Alternative text to image prompt to generate images like this: “a painting of a woman in a golden dress, fantasy art, harem, ornate jewelry, hd elegant, beautiful iranian woman”

DALL-E Beautiful AI Prompt Test 06

Alternative text to image prompt to generate images like this: “a woman with blonde hair and blue eyes, trending on cg society, perfect face template, ombre, lowres, balayage”

SDXL Beautiful AI Prompt Test 06

Alternative text to image prompt to generate images like this: “a bunch of flowers that are next to each other, a digital painting, by Carol Bove, trending on polycount, baroque, vintage – w 1 0 2 4, wallpaper pattern, rose twining, 4 k hd wallpaper illustration”

DALL-E Beautiful AI Prompt Test 07

Alternative text to image prompt to generate images like this: “a beautiful young woman with long brown hair, trending on cg society, realistic facial features, bold contour lines, girl with brown hair, makeup”

SDXL Beautiful AI Prompt Test 07

Alternative text to image prompt to generate images like this: “a woman with jewelry and flowers in her hair, trending on cg society, gold and red filigree, red and green color scheme, indian super model, a beautiful victorian woman”

DALL-E Beautiful AI Prompt Test 08

Alternative text to image prompt to generate images like this: “a woman wearing a mask with flowers in her hair, inspired by Hedi Xandt, trending on cg society, luxurious wedding, with ornate jewelled, blonde, intricate and detailed lace set”

SDXL Beautiful AI Prompt Test 08

Alternative text to image prompt to generate images like this: “a painting of a woman in a red dress, inspired by james christensen, fantasy art, detailed face of an arabic woman, ornate and intricate jewelry, 8k high quality detailed art, princess jasmine”

DALL-E Beautiful AI Prompt Test 09

Alternative text to image prompt to generate images like this: “a woman sitting next to a vase of flowers, by Marie Bashkirtseff, trending on cg society, long luxurious light blond hair, sexy face with full makeup, 4k, beauty filter”

SDXL Beautiful AI Prompt Test 09

Alternative text to image prompt to generate images like this: “a painting of a woman with red hair and jewelry, an ultrafine detailed painting, fantasy art, gems and gold, eastern art style, intricate ornament halo, extremely detailed artgerm”

DALL-E Beautiful AI Prompt Test 10

Alternative text to image prompt to generate images like this: “a man standing on top of a beach next to the ocean, by Adam Dario Keel, digital art, handsome face and beautiful face, square masculine facial features, fashion model face, parted lips”

SDXL Beautiful AI Prompt Test 10

Alternative text to image prompt to generate images like this: “a close up of a woman wearing a hat with flowers on it, inspired by Hedi Xandt, trending on cg society, rococo, elaborate long black hairstyle, russian costume, turquoise and venetian red, beautiful tan mexican woman”

DALL-E Beautiful AI Prompt Test 11

Alternative text to image prompt to generate images like this: “a close up of a woman with long brown hair, digital art, shutterstock, digital art, curly middle part haircut, dark chocolate hair colour, fashion model features, sexy face with full makeup”

SDXL Beautiful AI Prompt Test 11

Alternative text to image prompt to generate images like this: “a painting of a beautiful indian woman, fantasy art, elaborate gold jewelry, seated in royal ease, martin ansin, patron saint of 🛸🌈👩🏾”

DALL-E Beautiful AI Prompt Test 12

Alternative text to image prompt to generate images like this: “a woman with long blonde hair and blue eyes, digital art, shutterstock, sexy lips :5 stylish, attractive brown hair woman, hyper realistic style, fair olive skin”

SDXL Beautiful AI Prompt Test 12

Alternative text to image prompt to generate images like this: “a close up of a bunch of flowers, a digital rendering, inspired by Bartholomeus van der Helst, trending on polycount, maximalism, dark orange black white red, seamless pattern design, detailed painting“”

What Do You Think? Which Set Of AI Generated Images Are More Beautiful?

Interesting all 12 of the DALL-E images where of people and used a close up portrait style: there was very little variation, the results looked like filtered photos of models. Personally I don’t like any of the twelve DALL-E 3 images, they look fake. I can imagine a lot of men will like them and teenage girls might aspire for that sort of unrealistic fake/plastic/filtered look.

There are apps for guessing the text prompt used to generate an AI image. For the “DALL-E Beautiful AI Prompt Test 02” image the guessed prompt was:

“a woman with a flower crown on her head, trending on cg society, sexy face with full makeup, wearing elegant jewellery, wlop glossy skin, instagram model”

Two of the DALL-E creations where men (one of which had women as the background) and the remaining ten where women. All the men and women generated were white Europeans with very little variation in style etc… (boring IMHO).

For SDXL the results where more varied, but still centered around women. There appears to be an bias with these AI generators that when you ask for beautiful, stunning, gorgeous etc… you want a beautiful woman and not a beautiful landscape or a beautiful car or a beautiful solution to a problem.

None of the SDXL 1.0 images included men: I use NightCafe Studio a lot and it has a tendency to default to generating women even when the prompt has no indication of sex. My wife, who also uses NightCafe Studio regularly complains the AI is creating images of women even when she’s either asked for men or has included women in the negative prompts!

Ten of the SDXL generated images included women and two were flowers. Of the ten images of women, eight also included flowers as well (in total ten included flowers, in comparison DALL-E had four with flowers).

Note I chose the prompt “Beautiful, alluring, appealing, attractive, exquisite, gorgeous, handsome, lovely, magnificent, pulchritudinous, ravishing, stunning” knowing I’d get a result like this. I’ve ran similar test prompts with similar results. It does show the biases and limitations of SDXL 1.0 and DALL-E 3. Knowing these limitations and biases allows budding AI prompt engineers to modify their prompts to get closer to the results they hoped for.