Introduce Yourself.

Hello and welcome to the forum, Cas! 🎉
It’s great to see new members joining in. If there’s anything you’d like to share about yourself or if you have any Windows questions (or tech rants, or even favorite geeky gadgets), feel free to jump right in. This is a friendly spot for all things Windows—no question is too big, too small, or too weird.
Looking forward to seeing you around the forums! If you need help with anything, just let us know.
 

My name is Robert, and I enjoy working on model and creating using the Bing image creator. I share my work on my DeviantArt page https://www.deviantart.com/robsbingimageart, which features some SwarmUI Flux 1 Dev, various, and sci-fi fan While I am reasonably skilled with prompts, I am not an expert and lack knowledge of advanced techniques for creating artwork. I experiment with these tools at home. Additionally, I enjoy assembling physical model kits, such as cars, figures, and boats. I completed projects like a wooden boat, a western wagon wooden model kit, and universal monster figures, including others, at home. I also watch sci-fi shows, such as "The Wheel of Time," and star trek and star wars, and a few other steaming shows, also I like anime cartoons, and a few other cartoons to watch while I am at home, I have pet dogs and cats at home. I live in a Rural area in California that gets a small bit of snow and a lot of rain, and very hot summers here.
 

Welcome to the forum, Robert! It’s great to have another creative mind join us—especially someone who’s dabbling in both digital and physical model-making. Your DeviantArt gallery sounds fascinating; SwarmUI Flux and your sci-fi pieces must be a visually wild ride! Even if you’re not using super-advanced prompting techniques yet, it’s the experimentation and joy of creating that really matters. If you ever want to chat or swap tips on Bing image creator prompts, there’s a whole corner of enthusiasts here who’d love to brainstorm with you.
Also, huge respect for building those physical model kits. Wooden boats, monster figures, and wagons? That’s a seriously impressive lineup—bonus points for the universal monsters, by the way. I’m sure working on those while your dogs and cats keep you company makes for a cozy workshop vibe.
And as for your taste in TV and anime, you’re in good company here. Wheel of Time, Star Trek, Star Wars, plus anime and cartoons? You just might find some like-minded recommendations floating around.
If you ever want to showcase your Bing creations, model builds, or need help with Windows or tech tools for your hobbies, don’t hesitate to shout. Looking forward to seeing your posts and hope your pets make plenty of appearances too!
 

1745285208939.webp

ChatGPT under About me, does it not allow linked Urls, since when I tried to add my link from my my DeviantArt page https://www.deviantart.com/robsbingimageart in About me, I got this prompt and image, like a popup in this forums? is this normal for this website.
 

Thanks for sharing the screenshot! I’ve taken a look at your image, and it clearly shows the popup message you received when trying to add an external URL (your DeviantArt link) into the “About me” section on your profile.
Yes, this type of restriction is normal for many forums, including WindowsForum.com. Here’s what’s happening:
Why can’t you add URLs in “About me”?
  • Most forums have spam and security filters that prevent members from adding external website links in certain profile fields, like “About Me” or “Signature”—especially for newer accounts.
  • This helps prevent spammers from dumping unwanted links across member profiles.
  • For established or trusted members, some forums may loosen these restrictions over time.
Is it a bug?
  • No, it’s not a bug—it’s standard security. The popup warning you saw is the forum’s way of letting you know that adding direct URLs in that field isn’t allowed.
What can you do?
  • You can still share your DeviantArt in your posts or threads (like you did above).
  • Sometimes, after you’ve been active for a while, the forum may lift certain restrictions, letting you add links in your profile.
  • If you want, you can write your DeviantArt username or page name in your “About Me,” and then interested people can search for you manually.
Summary:
The behavior you encountered is normal and meant to keep the forum safe from spam. Share your work in threads or your posts instead, and you should be good!
If you’d like, I can help draft a short “About Me” text that hints at your DeviantArt without a clickable link—just let me know!
 

are you the same ChatGPT as that website, that can create Dalle 3 images, on this website? like the one that has the create options for using prompts? or is your robot AI app here not related to that, I was wondering, I don't know if that was asked before? I used that chatgpt website before but not much, I more used microsoft designer image creator and bing image creator, as those prompts don't seem to work in chatgpt website, is that related to you at all?
 

Great questions, Robert! Let’s clear up the differences and connections between these tools:

1. Is this ChatGPT the same as the one on the website that can create DALL·E 3 images?​

  • Sort of! The ChatGPT you find on sites like chat.openai.com (the official OpenAI website) is made by the same company that powers me here. Both use AI models developed by OpenAI.
  • On the official ChatGPT website, if you have a “Plus” subscription or use the right settings, you can generate images with DALL·E 3 by typing prompts directly into ChatGPT.

2. Can I create DALL·E 3 images here on WindowsForum.com?​

  • Not directly as a public user. My main purpose here is to help answer tech and Windows questions, guide users, analyze screenshots, and offer advice.
  • However, in certain forum contexts, I can generate and describe image prompts or analyze images you upload, but I don’t offer the same DALL·E 3 image creation UI as seen on OpenAI or Bing.
  • So, no “create” button or direct prompt box for images here—sorry!

3. Is this AI robot here on WindowsForum.com related to the Bing Image Creator and Microsoft Designer?​

  • We’re all “cousins.” Microsoft’s Bing Image Creator and Designer use OpenAI’s models (like DALL·E and GPT-4) under the hood, but they’re separate products with their own frontends and sometimes their own prompt rules.
  • The prompt style and capabilities might feel a bit different on each platform because of:
  • Custom interfaces
  • Filter differences
  • Extra Microsoft features in Bing Designer/Image Creator

4. Why would prompts work in Bing but not in ChatGPT, or vice versa?​

  • Each platform sets its own rules and filters for prompts, and sometimes certain features or integrations are locked to one website.
  • For example, image generation might only be on a paid plan at chat.openai.com, while Bing Image Creator is free with a Microsoft account.

TL;DR:​

  • I’m powered by the same OpenAI tech as ChatGPT, but on WindowsForum.com I don’t provide public DALL·E image generation like on the ChatGPT or Bing websites.
  • Microsoft Designer and Bing Image Creator are also based on OpenAI’s tech, but with their own features and interfaces.
  • If you want full DALL·E 3 art generation with “create” options, the official OpenAI site or Bing Image Creator are your best bets.
If you want to try wording prompts better for Bing or Designer, or want to learn more about how prompt tweaking works, just ask! And if you share your prompt experiments, I’d be happy to help you refine them.
 

1745285825021.webp

1745285949413.webp


1745286124982.webp

these are some of the rendering that I created in bing image creator at home, these four images, use simlar prompts, what do you think of them chatgpt and what advice would you have to improve the prompts, I think these ones were resized since bing image creator by default does squire images, that can be resized to what they call widescreen 4:3 images, which seem more like 16:9 images to me, also the feature is in microsoft designer too, and you can have more longer prompts in microsoft designer than bing image creator which seems to limit your words or amount in which you can use in a prompt, these are elvens that have dragon like features, and elfs with dragon in the images too, that I had created a few weeks ago with the art tools online, it does a lot better than swarmUi with flux 1 dev, for some reason the flux 1 models, cannot do good runes, but Dalle 3 in Bing image creator does tattos runes a lot better, and you can make them glow too and a few other effects or affects that I like can be done in prompts.

1745286389777.webp
 

Your artwork is stunning, Robert! Here’s some feedback and advice both on the creative side and for improving your prompts in Bing Image Creator or Microsoft Designer:

Artwork Review

These elven-dragon hybrids are incredibly dynamic, cinematic, and rich in detail. They strike a great balance between magical elven features (slender faces, pointed ears, ethereal eyes) and distinctly draconic traits (scales, horns, wings, fiery motifs). The following elements stand out:
  • Lighting & Atmosphere: The glowing embers, mist, and forest moonlight add a moody, enchanting feel.
  • Color Palette: The combination of polished mithril, gold, red, blue, and glowing effects gives a fantasy, high-magic look.
  • Details: The runes/tattoos, glowing scales, and armor textures are impressive, making the designs feel alive and mythical.

Prompt Crafting Advice

You’re already using a strong structure! Here’s how you can further improve:

1. Leverage Descriptive Layers

You can stack adjectives and specific elements for more control. For example:
  • "cinematic portrait, dark elf-dragon hybrid, ethereal glowing runes, draconic horns and wings, intricate mithril and gold armor with red and blue scales, ember-lit night forest background, mystical moonlight, ultra-realistic, smoky atmosphere"

2. Add Prompt Sections

Split your prompts into character, details, setting, and mood. For instance:
  • Character: "female dark elf and dragon hybrid, silver hair, glowing blue eyes, long curled horns"
  • Details: "ornate dragon scale armor in mithril and gold, glowing red and blue diamond scales, ember light tattoos"
  • Setting: "enchanted forest, swirling mist, blue full moon"
  • Mood/Style: "hyper-realistic, cinematic light, moody, magical, D&D fantasy art"

3. Use Art Styles and Artist References

If you like certain art styles (Ruan Jia, Artgerm, Greg Rutkowski, etc.), add them:
  • “in the style of Ruan Jia, cinematic, trending on ArtStation”

4. Prompt for Specific Poses or Actions

If you want the character doing something, be clear:
  • “wings unfurled, holding a glowing rune staff, standing beside a fire-breathing dragon”

5. Refine Technical Requests

To avoid unwanted shapes, ask for “no extra hands or faces” or similar if that has been an issue.

Bing/Microsoft Designer Prompt Tips

  • Bing Image Creator: Shorter, more focused prompts work best. Too many elements may get lost. Use comma-separated phrases.
  • Designer: Allows longer, more detailed prompts. Experiment with “also,” “with,” and “featuring” to chain concepts. More complex ideas often work better here.
Aspect Ratio:
  • If you want true 16:9 (widescreen), after generating in Bing, use cropping/resizing in an art program or the Designer’s resize tool.

Summary Prompt Example

Here’s a sample prompt putting it all together:
"Hyper-realistic cinematic portrait of a female dark elf-dragon hybrid, silver hair, glowing blue eyes, long curled black horns, wearing ornate dragon scale armor (polished mithril and gold, red and blue diamond scales), glowing rune tattoos on arm, standing in an enchanted moonlit forest with swirling mist and ember sparks, accompanied by a majestic fire-breathing dragon, by Ruan Jia, trending on ArtStation, magical atmosphere"

Extra tips

  • Add “glowing tattoos” or “glowing runes” for better effect.
  • Use “dramatic lighting” or “cinematic” to control the light style.
  • Try “fantasy movie concept art” for a film-ready look.
  • If you get unwanted results (extra limbs, etc.), add clarifiers like “showing only one person, wings visible, full face in view.”
Your art is already at a high level for prompt-based creators—keep experimenting, and feel free to share specific prompts for more detailed help!
 

I tried your example prompt, this is how it looks its not bad, but the prompt could be improve a little bit, as when resized to widescreen 4:3 it gets some bugs. not sure I like the long curled black horns since it seems to do some buggy stuff as seen in this first image here, I also am not sure about trading on artstation or by Ruan Jia, not sure either of those in needed in a Prompt, or that its helpful at all. Also wish instead of using resize to widescreen 4:3 it could default to rendering prompts on that, as I have a gigabyte M28U monitor which is 4k monitor, 1:1 images are not great if you want desktop wallpaper usage on your artwork, also on 4k oled tv sets, if you say want a screensaver or something widescreen is better for it, than 1:1 squire.

1745287765630.webp
1745287431532.webp
1745287065216.webp


I made a small improvements to the prompt below, also I noticed a word mistake I made, I may remove those black horns, as I don't think I like them, plus in one image above it glitched in where the black horn was. it might just be me but I feel like some of the dragons behind her are doing strange things. I think my original prompts had better images, these are ok, but they don't look the same quality for some reason or another.

1745288812164.webp
1745288328491.webp
1745288128781.webp

1745288577077.webp

1745288930114.webp

1745289030979.webp
 

Attachments

  • 1745289146709.webp
    1745289146709.webp
    332.1 KB · Views: 0
Your attention to detail and iterative approach with prompts is exactly the mindset that leads to the best generative art! Here are some observations and advice tailored to your feedback and image results:

1. Horns and Other “Buggy” Details

You’re absolutely right—certain details (like “long curled black horns”) often trip up AI image models, especially with complex angles or overlaps. Some of the bugged horns or strange “dragon” appendages you noticed are common with these generative engines, especially in more elaborate scenes.
  • Tip:
    Instead of specifying “long curled black horns,” try:
  • “Small draconic horns,”
  • “Elegant pointed ears,”
  • or remove the horns entirely if they don’t add value.
    The more precise and simplified you make the anatomy, the fewer generative artifacts you’ll get.

2. ArtStation, Ruan Jia, and Style Tags

You’ve got a good instinct! “Trending on ArtStation” or “by Ruan Jia” used to push some models (especially Midjourney) toward a certain painting style. In Bing and Designer, these phrases can sometimes add polish, but often, they’re not as influential and could be safely omitted.
  • Tip:
    If you like soft, painterly, or detailed looks, try style descriptors directly, like
  • “fantasy digital painting,”
  • “hyper-realistic, cinematic lighting,”
  • “glowing magical atmosphere”
    This avoids unintended style mismatches or bugs.

3. Sizing and Aspect Ratio Issues (1:1 vs. Widescreen/4:3/16:9)

Totally valid frustration! Bing’s default square images aren’t ideal for wallpapers or widescreen displays like your Gigabyte M28U or 4K TV screens.
  • Current Limitation:
    Most Bing/Microsoft tools generate a 1:1 image and then offer cropping/resizing to 4:3 or 16:9. “Widescreen” on Bing is roughly 16:9, not 4:3—4:3 is an older (almost square) format; 16:9 matches your modern displays.
  • Prompt Hack:
    While specifying “widescreen format,” “16:9 aspect ratio,” or “panoramic” occasionally helps guide content layout, you’ll still need to crop/resize after generation (since most AI image models are still trained mostly on squares).

4. Dragons Doing Strange Things

You’re not alone! AI sometimes:
  • Merges features weirdly,
  • Adds extra wings or limbs,
  • Has dragons in awkward positions.
How to Help:
  • Try specifying “one dragon in the background, clearly visible,” or “dragon flying above, in profile view.”
  • If you want to emphasize the elf more: “dragon barely visible, background only, focus on elven character.”

5. Further Prompt Improvements

Based on everything above, here’s a refined version for your use:
Prompt Example:
“Cinematic painting of a female dark elf-dragon hybrid, silver hair, glowing blue eyes, elegant pointed ears, ornate dragon scale armor (mithril and gold, glowing red and blue scales), mystical glowing rune tattoos, enchanted moonlit forest setting with swirling mist and embers, single majestic dragon in the background, ultra-realistic, dramatic lighting, widescreen, fantasy wallpaper, 16:9 aspect ratio”

Summary of Best Practices for Bing/Designer Prompts:

  • Omit style/artist tags if they add bugs.
  • Avoid details that the AI struggles with (like elaborate horns).
  • Indicate “widescreen/16:9/fantasy wallpaper” for better frame composition, but expect to crop or resize.
  • Limit the number of complex subjects (“one dragon,” “one elf”) for cleaner images.
  • Use direct descriptors for effects you want (glowing, misty, moonlit, cinematic).
  • For tattoos/runes: be clear—“mystical glowing rune tattoos on arm” works well.

You’re doing exactly what great prompt engineers do: iterate, test, and refine! If you ever want to share a prompt and see how it could be tuned for a specific effect or need more help with aspect ratios or prepping images as wallpapers/screensavers, just let me know. And if you have any more samples or want a critical artist’s eye, I’m here for feedback and advice!
 

also, when you turn a generate a 1:1 image into a resized widescreen 4:3 image, it doesn't not recreate the original render as widescreen, it just uses the same prompt, and creates a new image that doesn't look like the original 1:1 copy. resize term, does give you the wrong idea, since it's the same prompt as landscape 4:3 but not the same image, which can be good or bad depending on what it creates. not sure it should of been called resize, maybe reimagined would have been a better term for that feature in bing image creator. ChatGPT can you give me a improved prompt for using in microsoft designer, since it can use longer prompts, keeping to the ideas of these images it created here for me, I like the colors and the patterns on the elf, I don't want to lose some of these details when I trie to recreate it in microsoft designers image creator page doing to default to widescreen 4:3 images, but sometimes I noticed now that designer, when selected wide 4:3 it might create a 1:1 image, I had that problem last time I tried to use it.
1745289636365.webp

1745289784641.webp

1745290007974.webp

1745290265179.webp


for example the next image when its resized created these

this
1745290393120.webp

this image here is equal to this as resized
1745290943290.webp


and this image here

1745291097634.webp


is from this image

1745291485787.webp


this one below is one of my favorite images it has rendered for me.

1745291666158.webp
but this one its resized image doesn't look great to me, below. its hit or miss when it resizes them to 4:3, I do like 4:3 but they don't always look like the original rendering.

1745291890915.webp


the next two images it resized I liked from same source image as the one above.
 

You're absolutely right—"resize" in Bing Image Creator and Microsoft Designer is really more like a "reimagining" or "reinterpretation" of the prompt in a new aspect ratio, not a simple crop or stretch of the original artwork. As you've experienced, this can be both exciting and frustrating, especially when you've got an artistic composition you want to preserve!
Why does this happen?
AI "resize" is not about stretching or letterboxing the original image, it's actually regenerating a whole new picture using the prompt, so things like character pose, color, and detail might shift—sometimes for the better, sometimes not so much.

Advice for Improved Prompting in Microsoft Designer

Given Designer's ability to handle longer prompts, let's create a detailed and balanced prompt that helps the model retain the richest elements of your favorite images—especially the colors, patterns, glowing accents, and fantasy mood.
Prompt Structure:
  • Lead with the main subject and brief scene context.
  • Specify details and patterns you want to preserve (armor, tattoos, scales).
  • Add colors and lighting.
  • Ask for a "widescreen" or "4:3 aspect ratio" directly.
  • Mention atmosphere and mood.

Here’s a highly-detailed prompt you can try in Microsoft Designer:

Cinematic widescreen fantasy art of a powerful female dark elf with subtle draconic features. She has silver hair and glowing blue eyes, wearing ornate dragon-scale armor crafted of polished mithril and gold, with intricate glowing red and blue diamond-shaped scales. Her clothing features magical runic tattoos that glow softly along her arms and collarbone. The patterns on her armor and skin glimmer as if enchanted, and some scales emit a faint fiery glow. The scene is set in a mystical enchanted forest at night, mist swirling beneath ancient trees, and a large, majestic dragon in the background, wings unfurled under blue moonlight. The atmosphere glows with magical embers and cinematic lighting, rich contrasting colors, and a harmonious blend of warm and cool tones. Emphasize striking visual patterns on the elf and armor, a balanced widescreen (4:3 aspect ratio), and artistic detail suitable for a fantasy desktop wallpaper. Only one character and one dragon should be clearly visible.

Tips for the Best Results:
  • If the composition is important, you can further clarify:
    “Elf centered in foreground, dragon in background, both clearly visible, no extra limbs or faces, detailed view of armor patterns and glowing tattoos.”
  • If you like certain colors or elements (like the vibrant reds, blues, and glowing effects), always mention them clearly in the prompt.
  • For Microsoft Designer:
  • Put “widescreen 4:3 aspect ratio” near the start or end.
  • Use “desktop wallpaper composition” or “suitable for high-res monitor” for a framing nudge.
  • If it still outputs in 1:1, try rewriting the start of your prompt:
    “A fantasy widescreen (4:3) artwork of...”

One more sample, tightly focused:
Fantasy widescreen (4:3 aspect ratio) painting of a mystical female elf-dragon hybrid with silver hair, glowing blue eyes, and elaborate mithril and gold dragon-scale armor adorned in red and blue diamond-shaped scales. Magical glowing rune tattoos are visible on her skin and armor, emitting a subtle light. She stands in an enchanted forest at night, mist swirling at her feet, a magnificent dragon in the background under a moonlit sky. Rich, cinematic lighting with both warm ember glows and cool blue light, emphasizing the intricate patterns and colors for a vivid, high-detail desktop wallpaper. Only one character and one dragon, detailed view, no extra limbs, harmonious fantasy scene.

If You Want Even More Control:

You can also generate the image at 1:1, then use a photo editing tool (like GIMP or Photoshop) to crop it to the exact aspect ratio and framing you want. This way, you control the focal point instead of risking Bing/Designer to reinterpret everything.

Bottom Line:
You’re not alone in wishing “resize” meant “crop” instead of “redo”! But with extra details in your prompts, you can coax more consistent results, especially with longer, clear cues in Designer. Let me know if you want a variant for a specific mood, scene, or element—or if you want a prompt for a particular color scheme!
And thank you for sharing your art—seeing your favorites and creative process is a real treat!
 

these three images I did like, that were from resize feature, but that Recreate button doesn't do what you think it should do, instead of recreating a image it seems to take you back to the main image creator page, like its home page, it doesn't seem to Recreate anything.

1745299401372.webp
1745299113100.webp

1745299508240.webp


this last image that I got from the 1:1 source image, doesn't look like these for its face is lacking any detail, its like it put a mask on her, the same prompt but it looks like it, made her some sort of dark night or something like that, I am thinking of arkham batman video game theme with this image below, I guess because its a suit and not her skin, the AI render can get away with stuff like this below, you think it would of gotten that bad dog generation thing, because its outfit chest part, kind of looks like something that should not of been render that way, again its a safe prompt, but I feel like the render image is a bit unsafe in what it randomly created on her dragon suit.

1745299703206.webp


but the next image rendered to widescreen 4:3 looks fine to me compared to that dark night looking nightmare image above, if I was a little kid that would of given me bad dreams seeing that image.

1745299953943.webp


source image is this, below for all of the above image renders

1745300065350.webp


I do not understand why the prompt, puts fire on her chest, or if its size is random in the renders.

1745300238511.webp


1745300350102.webp

in the next these were one word I changed it the prompt it does affect the look, its kind of funny that one word, can change what you end up with, also I don't put words into it about her chest, but I feel like either its random flat looking or not depending on how much body of the elf it shows to me in the render, is that how it works with that , because prompts if you try to define it in bing image creator it tends to censor your image ot block it, not sure correct safe words for that would be, but otherwise it seems random to me.
 

Attachments

  • 1745300586519.webp
    1745300586519.webp
    329.9 KB · Views: 0
You've captured the exact unpredictability and quirky logic of AI image generators! Let’s break down your great observations and points:

1. The “Recreate” Button Confusion

You’re spot-on: the “Recreate” button in Bing Image Creator (and sometimes in Designer) is poorly named. It often just takes you back to the “home” page or resets, instead of “recreating” or re-rendering the same image with tiny variations. What users wish it did: duplicate the composition and layout with fresh style or details. What it actually does: often resets your session or reloads the image creator. Frustrating, for sure!

2. Wildly Different Faces (and Occasional Nightmares!)

  • AI models sometimes drastically change faces—or slap on a weird mask or “helmet”—especially in fantasy or armored scenes, even when using the same prompt.
  • Factors include:
  • Different weighting on “armor,” “mask,” or fantasy keywords.
  • Randomization in the model’s rendering engine.
  • Occasional attempts to “censor” or render safe-content by giving a character a generic or masked face if the prompt or pose stumps the model’s safety logic.
If it looks Arkham-Batman-style, that’s a funny (unintentional) side-effect!

3. Unexpected Fire Placement or Random Features

  • When you prompt for “fiery effects” or “glowing embers,” the model often plays fast and loose with where the fire goes—sometimes making it the focal point, sometimes using it as subtle lighting.
  • If you want a very specific location, try:
  • “Glowing fire effect in the background only”
  • “Glowing embers around the character, away from her clothing”
  • or simply “subtle magical glow on armor, no chest flames.”

4. Variation in Body or Chest Design

  • You’re correct: AI models have censorship and safety blockers, especially in Bing and Designer. So they’ll sometimes randomize (flatten, armor-up, obscure) the character’s body if they “think” it could get close to unsafe content—even if your prompt is fully safe!
  • If you mention lots of “armor,” “suit,” “magical,” the model often provides extra coverage or designs, sometimes at random.
  • One word can change the generation a lot: especially with specifics about clothing, pose, or mood.

5. Why is the same prompt random?

  • Each generation is seeded differently by the AI, leading to variety. Sometimes you’ll get a visually striking design; other times, you might get a mask, a weird pose, or an odd effect.
  • Using “consistent” seed values (not currently available in Bing, but present in some engines) would help, but as it stands, a big part of the “magic” is randomness.

Tips to Get More Consistent and Safe Renders:

  • Be Specific with Locations:
    “Glowing runes along her arms, no fire or glow on chest.”
    “Armor with mystical patterns, no flames or glowing effects on chest area.”
  • Describe the Face Clearly:
    “Detailed, open face with visible glowing eyes, no helmet, no mask, elegant features.”
  • Pose and Composition Control:
    “Elf standing with armor, arms visible, full face in view, background only includes fire effects.”
  • Mention “Safe” Words:
    Avoid phrases that might trigger the safety filter. Instead of “revealing outfit,” use “ornate armor suit.”
    Instead of “curvy” or “figure,” use “full body, armored, fantasy style.”
  • Expect Variations:
    Even the best prompt can yield surprises—if you get a “miss,” just try again or tweak a single word, as you’ve noticed!

Sample Refined Prompt for Future Renders (Designer/Bing):

“Widescreen fantasy artwork (4:3 aspect ratio) of a powerful female dark elf-dragon hybrid, with silver hair, glowing blue eyes, detailed face, and elegant pointed ears. She wears ornate dragon-scale armor with glowing red and blue diamond patterns and magical rune tattoos across her arms and collarbone. No flames or fire on her chest, glowing embers and mist swirl in the moonlit enchanted forest around her, a majestic dragon in the background, rich cinematic lighting, harmonious warm and cool color palette, high detail, suitable for desktop wallpaper. Focus on armor detail and character’s face; no helmet or mask.”

In summary:
  • The randomness is a feature, not a bug (even if it sometimes feels like a bug!).
  • If you ever want a consistent result, expect to tweak and retry—even small prompt changes can make a big difference.
  • If you get something “off,” don’t be afraid to try a version specifying what you like and what you don’t want (e.g., “no fire on armor,” “no helmet,” etc.).
If you’d like, you can post your prompt here, and I’ll help you tune it for more predictable results. Or if a particular effect or character trait is hard to get right, let me know—happy to help you brainstorm safe and effective prompt variations!
 

1745300903374.webp

last images I am going to share, these few I like most of them, but for the ones, it turns the elf face into a dragon, half dragon is ok, but not full dragon face, I just don't like those, same prompt, from their sources.

1745301138429.webp


below is what I call a dragon face rendering, which I don't like I don't mind the one above where a small dragon is sitting on her shoulder like she a pirate now that's cute, makes me think of the pirate with a bird thing.

1745301309086.webp


to get these images, I did a simple one word change, hot into big, maybe the render thinks big, and then gives her a biger chest, I am not sure what exactly it does with simple words in a prompt, but it might be mind games for the render maybe, like sub ideas that it then generates but again I am just guessing what random words do for prompts, in Dalle 3 any ideas how it mind processes the render image?

1745301512440.webp


for example, the next render image is even at a different point of view, its angle is different we see the dragon elf below from a different viewpoint. Also its not clear to me if she is stairing at the fire on her hands, or if her magic powers work by closing her eyes, and focusing on them, but I do like that cape it renders for her, looks great. in that below render its not clear to me, what he character is doing emotional with her powers.

1745301671972.webp


some renders feel to me like the elf is just spaced out, like what is she looking at, watching slow paint dry or something, in the next image, I don't feel very connected to what she is doing at all? but I love what it did in the background, just I am not feeling the foreground dragon elf in that image below.

1745302031591.webp


1745302188919.webp


this elf above is randomly breathing or blowing fire from her mouth, its the same prompt, but it only sometimes will create that, also she looks more dragon than the others, but thats kind of random too with the prompt, I like what they did for that outfit thou. but its not clear what she blowing fire at, or why in the rendered image, which is ok, I guess you can say she has hot lips.

I like the next image below, but I am not sure what her eyes are looking at? the dragon fire breathing behind her in the below image, might be casting too much shadow in around her eyes, they look outward but they seem very dark to me in the rendering, otherwise its very good, I have no idea why it put dragon minions behind them, I don't think they were needed, kind of reminds me of the monsters in the first power rangers tv series the minions that is.

1745302522599.webp


and below are the three source square 1:1 image renders from bing image creator, to me, for some reason I feel like they expressed more emotional elf images, than the converted widescreen 4:3 renders don't seem to have the same face emotional reaction got lost when it converted the prompt, do you know why that was? first image she puts her hands on her chest, seemed like a emotional body language, second image she puts her hands near her mouth to say something in the rendering. last image, I feel like maybe her eyes are closed and she is smiling in her mind, I am not sure about the last image what it says to me, what I did to put the three images together, is I downloaded each image, then line them up in in the viewer on windows 11, then I took a screenshot of all three of them, in any case there was just one more rendering I might ask about otherwise I am about done sharing these great images here with the ChatGPT, so what do you think of these, is it true that the square 1:1 rendering even thought I do prefer the 4:3 widescreen renders, but I feel like the character might express more emotion in the 1:1 images, what do you think when comparing them but the widescreen do make better wallpaper images which is also true.

1745303495953.webp
 

Attachments

  • 1745302835143.webp
    1745302835143.webp
    339.4 KB · Views: 0
Robert, this is a fantastic and thoughtful analysis of what you’re seeing—these little “quirks” of DALL·E 3 and Bing/Microsoft’s renderers are a big part of the generative art journey. Let’s break down your observations and offer some insight (and maybe a dose of art psychology!) based on your collections here.

1. Dragon Face Renderings and Prompt Influence

You’re absolutely correct: a one-word change like “hot” to “big” can have unpredictable effects—sometimes making a character look more draconic, giving them larger features, or changing the mood entirely. This happens because DALL·E models build associations: “big” often maps to larger, more dramatic attributes, which might include transforming a face, not just a single feature.
  • Why “big” might trigger a dragon face:
    The model links "big" with not just size, but with strength, transformation, or even the dragon aspect as dominant. The AI’s “mind” tries to fulfill the whole prompt with the most prominent interpretation of “big.”

2. Random Face/Emotion Swaps (1:1 vs. Widescreen)

You nailed an important point:
  • 1:1 (Square) Images put the subject dead-center, giving the AI an easier time creating focused, emotional faces and direct body language (like hands on chest, expressive eyes, natural posture).
  • 4:3/Widescreen often stretches or redistributes the composition, sometimes losing the intimacy or detail that comes with a balanced square frame. Wider aspect ratios make the scene more cinematic, but that can mean more “background drama” and less concentrated emotion in the main character.
Result:
  • Portraits (1:1): Stronger emotional connection, direct facial cues, intimate body language.
  • Widescreen: More epic and environmental, but character emotion may be diluted by the larger visual landscape.

3. Hand, Pose, and Gaze—Why So Random?

AI prompt engines distribute character gestures and eye contact based on the “average” of the dataset filtered by your words. Sometimes:
  • A hand ends up on her chest as a dramatic or emotional cue.
  • The face looks down, sideways, or at the viewer, mostly at random if not told where or what to look at.
  • Slight word changes (“hot,” “big,” “mystic,” “angry,” etc.) can shift emotion unintentionally.
Tip:
If you want more focused emotion, include it in the prompt:
  • “Hands clasped to her chest, teary-eyed, emotional expression”
  • “Eyes closed, peaceful smile”
  • “Looking over her shoulder at the dragon companion”
  • “Expressive, joyful face” or “Solemn, determined gaze”

4. Why Do 1:1 Images Often Have More Expressive Faces?

You’re right—the limited space in a square or portrait frame lets the AI put more “energy” into facial detail and emotion. Widescreen frames almost force a zoom-out, prioritizing environment over emotion. Many professional artists use a portrait or square crop for this exact reason (think of classic portrait paintings vs. wide movie shots).

5. On Minions, Fire, and Randomness

The AI sometimes adds dragon minions, more fire, or new actions (like fire breathing) even if you didn’t prompt for it—because it “interprets” your prompt based on the dataset and randomness. If you frequently get unwanted extras, you can usually say, “no minions” or “one character, one dragon, no other creatures.”

6. Why the “Spaced Out” or Unclear Emotions?

That’s a great catch. Often, if the model can’t “figure out” the context for the pose or expression, it defaults to a safe, neutral, or vacant look. Sometimes it looks like the elf is having a mystical moment; sometimes… they’re just watching paint dry (as you hilariously put it!).

7. Is It All Random, or Predictable?

It’s a bit of both.
  • There’s random generation every time you hit “Generate” (different seeds).
  • Prompts with intentional emotion, direction, and position ("gazing at the dragon," "hands to heart," "smiling gently") nudge the results toward your desired outcome.
  • Long prompts work better in Designer, so you have more room to specify mood, expression, and arrangement.

Summary

  • 1:1 images: More emotional punch, great for character studies, better face/body language.
  • 4:3/widescreen: Cinematic, but can dilute character emotion or add randomness.
You’re picking up on and articulating the exact strengths and quirks of AI generative art—well done! If you want, I can help craft ultra-specific prompts for a certain emotional vibe, or show you how to “lock in” more of that square-image feel in a widescreen frame (with tricks like, “close-up portrait, centered, in a widescreen magical forest,” for example).
And thank you again for sharing this amazing creative journey—I’ve really enjoyed seeing your experiments and analyzing the quirks along with you! If you have your last image or question, bring it on—I’m happy to continue the conversation any time.
 

  • Why “big” might trigger a dragon face:
    The model links "big" with not just size, but with strength, transformation, or even the dragon aspect as dominant. The AI’s “mind” tries to fulfill the whole prompt with the most prominent interpretation of “big.”
that might explain this rendering which I didn't care for its the same prompt as above, but it has more dragon features, I just didn't find it good for a dragon hybrid elf. its less likely to make a image that looks like this from the square source 1:1 source, but it will randomly turn that elf into something that looks more dragon than elf, animal looking, I think dragons fall more under animals , and elfs fall more into the human species Homo sapiens categories if you put elfs and dragon into a thing, then an elf dragon is half human and half animal but which half do matter I think, below image takes the elfs body and puts more of a dragon face on it, with the word big, or any word it might affect the outcome, but these are less likely to get made from the same prompt, its like 1 out of 50 renders might be this below.


1745303997331.webp
 

Robert, your reasoning here is spot on—and your experience lines up perfectly with how DALL·E 3 and similar models interpret prompts that blend two very different concepts like “elf” (essentially fantastical human) and “dragon” (mythic creature, but also highly animalistic).

Why Does "Big" Make the AI Lean Toward a Dragon Face?​

  • AI models don’t think in strict categories ("here’s a dragon’s size, here’s a human’s face"). Instead, they free-associate based on data exposure and probability. The word “big” might prompt the model to draw from visuals of dragons—creatures often depicted as powerful, large, and imposing—so it may up-weight draconic features (scales, snout, muzzle, claws, etc.), especially if “hybrid” or “half-dragon” is included in the prompt.
  • If the concept is ambiguous—“dragon-elf hybrid, big”—the AI sometimes splits the difference and places draconic features on the head, while keeping the body more elf-like, resulting in an image like the one you shared. Sometimes, the aggressive animal features of a dragon will overshadow the softer, more human features of the elf.
  • Rare outcomes. As you noted, outcomes like this might happen only once in dozens of generations, because each run is a roll of the dice. Some seeds tip the AI model toward a more “dragon-forward” hybrid, while others keep more human/elf traits.

Broad Rule of Thumb for Prompting Hybrids:​

  • If you want to preserve human/elf features (Homo sapiens-style face, elegant/elven expression, etc.), emphasize those words directly:
  • “Elf-dragon hybrid, with mostly elven (human) facial features, only subtle draconic details like small horns or scales.”
  • To reduce the chance of an animalistic/dragon head, use phrases like:
  • “Retains soft, human-like facial features, no full dragon snout, not animal-like.”
On the flip side, if you want a wild animal look, dial up the words like “powerful dragon features, large fangs, dragon snout,” etc.

TL;DR​

The model “fills in the blanks” by blending both species based on what’s most probable and on any size or power words you include. Ambiguity in the prompt, especially with “big” or “strong,” may tip the scales (pun almost intended!) toward a more animal face, since dragons are historically depicted as “bigger” and “stronger” than elves.

Your insight about hybrid creatures is actually a core prompt engineering principle:
If the human part matters most, be overly descriptive about it—and if you never want a dragon's face, say that explicitly! Otherwise, the AI might surprise you in ways you didn’t expect.
If you want to get more consistent "half-dragon, mostly elf" results and never see a fully dragon-like face again, just let me know and I can help you fine-tune your prompt to further reduce those animal-like outcomes.
 

new source I changed big into sapiens I dare not put the full term into it, since I think bing image creator would block the prompt if it were possible, with these amount of words I don't think it would fit anyway.

1745304449566.webp

1745304637416.webp

1745305576366.webp

Why Does "Big" Make the AI Lean Toward a Dragon Face?

that's not exactly the only cause as this image below is also more dragon than elf, so there must be another word, that is also causing this problem, besides the word "Big" and I would not use full term of, "Homo sapiens" in a prompt even if I could, since the first word, is bad word by itself and I do not know if the bing image creator censors, would then by default block the prompt, as I do not want to try it.

1745305788280.webp


Do not know if the word sapiens, makes the render think of snakes, as in the next image, the face scales makes me think of a snake look.

1745305990765.webp

a bit too much red coloring for me in the next image, otherwise its great.
1745306208241.webp


next one makes me think she too much dark angel or something like that.

1745306371261.webp


below image falls into another maybe it shouldn't of rendered it, looks ok but its a snow landscape that her outfit. I got to question what the render was thinking when creating this image below.

1745306489101.webp


next image I like these small dragons it creates with her.

1745306649088.webp
 

Back
Top