How Do I Create Images With Ai

General
1

I’ve been seeing amazing AI-generated images online and I’d really like to learn how to make my own for personal projects and social media. I’m confused by all the different tools, prompts, and settings, and I’m not sure which platforms are beginner-friendly or free to try. Can someone explain the basic steps, which AI image generators to start with, and any tips for getting good results without a lot of technical knowledge

2

Short version: pick one tool, learn prompts, tweak settings, repeat.

Here is a simple path.

  1. Pick a tool
    Easiest options to start with
    • DALL·E (inside ChatGPT / Bing Image Creator)
    • Midjourney (Discord)
    • Stable Diffusion (for more control, but more setup)

If you want fast and simple, use Bing Image Creator or DALL·E.
If you want “pro” control, go Stable Diffusion later.

  1. Learn prompt basics
    Structure your prompt like this
    • Subject: “portrait of a woman”
    • Style: “cinematic lighting, 35mm photo, depth of field”
    • Details: “freckles, soft light, realistic skin, studio background”
    • Quality words: “high detail, 4k, sharp focus”

Example
“a cozy living room interior, warm lighting, lots of plants, large window, soft shadows, ultra detailed, high resolution, photo realistic”

Avoid vague prompts like “cool art of a guy”.
Add what you want, remove what you do not want.

  1. Use negative prompts
    Some tools have a “negative prompt” box. That tells the model what you do not want.
    Example negatives
    “blurry, low quality, distorted hands, extra limbs, text, watermark, logo”

  2. Play with key settings
    Main ones you will see in tools like Stable Diffusion
    • Steps: 20–30 is fine for starters
    • CFG / guidance: 6–9. Higher means it listens more to your words.
    • Sampler: try Euler or DPM++ 2M Karras, then compare.
    • Resolution: 512x512 or 768x768 to start. Higher sizes need more power.

On Midjourney
• “–ar 16:9” for landscape
• “–ar 9:16” for phone wallpaper
• “–v 6” or whatever latest version is

  1. Use references
    Upload a photo and say what you want changed, like
    “Use this photo of me, make a stylized portrait, studio lighting, blue background, cyberpunk style.”

Reference images help keep style or face consistent.

  1. Start simple, then iterate
    Do this loop
    • Prompt once
    • See what is wrong
    • Edit prompt to fix only one or two things
    • Generate again

Example
First try: “anime knight on a horse”
Result: armor looks weird, background boring
Second try: “anime knight on a horse, detailed silver armor, sunset battlefield, dramatic clouds, high detail”

  1. Where to use for social media
    • Portraits and avatars
    • Backgrounds for stories
    • Thumbnails for videos
    • Simple posters with text added later in Canva or Figma

Avoid putting text in the AI prompt. It messes up words a lot.
Generate clean art, then add text in a design tool.

  1. Basic “starter prompts” you can copy
    • “hyper realistic portrait of a young woman, soft natural light, shallow depth of field, detailed skin, 85mm photography, neutral background, high detail, 4k”
    • “isometric cozy bedroom, pastel colors, small desk with pc, plants, fairy lights, clean lines, vector style, high detail”
    • “minimalist flat illustration of a person working on a laptop, light background, modern color palette, clean shapes, vector art”

  2. If you want to go deeper later
    Search for
    • “prompt engineering stable diffusion guide”
    • “midjourney prompt cheat sheet”
    • “stable diffusion automatic1111 tutorial”

For now, pick one tool, try 10–20 prompts, and keep notes on what words give you the results you like.
That trial and error phase teaches you faster than any long guide.

3

Skip the “learn everything” trap. Think in use-cases first, tools second.

You said “personal projects and social media,” so I’d split it like this:

1. Figure out what you want the images for

Different goals need different habits:

  • Profile pics / portraits

    • Use tools with face control: DALL·E (through ChatGPT), Lensa, or apps that do “AI avatars.”
    • Start from your own photo and tell it the style: “same pose, Pixar style, bright colors, clean background.”

  • Backgrounds, vibes, aesthetics for posts

    • Use something that’s fast and forgiving: DALL·E, Bing Image Creator, Ideogram, Canva’s AI image.
    • Think “mood + scene” more than “super technical prompt”:
      • “cozy fall cafe, overhead shot, warm orange lighting, steam from coffee, clean composition.”

  • Thumbnails / social banners

    • Honestly, the workflow is:
      1. Generate a cool background in AI.
      2. Add text and layout in Canva/Figma.
    • AI is terrible at usable text. I slightly disagree with @stellacadente here: you can experiment with text in the image, but be prepared for wonky letters and redo the important text manually.

2. Don’t start with Stable Diffusion

Everyone treats Stable Diffusion like “real pro mode.” It is, but also it’s a distraction when you’re new. Installing, models, checkpoints, LoRAs, samplers, VAE… that stuff is fun later, but right now it will eat your motivation.

For a beginner who just wants cool stuff for socials:

  • Use: DALL·E / Bing Image Creator / Midjourney if you don’t hate Discord.
  • Ignore: custom models, ControlNet, etc., until you’ve made at least 100 images and know what you wish you could control better.

3. Think in “recipes” instead of prompts

Instead of trying to write the perfect mega prompt, make small reusable “recipes”:

  • “Moody photo look”

    • [subject], dramatic lighting, high contrast, film photography, grain, 35mm, shallow depth of field, muted colors

  • “Cute flat illustration”

    • [subject], flat vector illustration, pastel colors, simple shapes, minimal shading, white background, clean design

Just swap the subject:

  • “a person reading a book”
  • “a couple cooking together”
  • “a gaming setup on a desk”

You’ll get consistent style without rethinking from scratch each time.

4. Use images as training wheels

A hack that beginners skip:

  • Upload a reference image and tell the model what to change:
    • “Use this photo, make the lighting softer, pastel color palette, add plants in the background, Instagram aesthetic.”
  • Or: “keep the composition but change it to watercolor painting”

This teaches you faster than prompt theory because you see how style words actually affect something you recognize.

5. Make a “failed images” folder

When something comes out weird, don’t just delete it:

  • Save the ones that are almost right.
  • Look at them later and reverse engineer:
    • “Why is this one better than the others? Did I mention ‘soft light’? Did I remove ‘hyper realistic’?”

You’ll start building your own mental dictionary of what words do, instead of relying on generic prompt lists.

6. Keep your settings dumb-simple

On tools that expose settings:

  • Steps: leave default unless images look super mushy.
  • CFG / guidance: mid-range like 7 is fine, but don’t obsess. The prompt matters more than wiggling from 6.5 to 7.5.
  • Aspect ratio: match your use
    • Stories / Reels: 9:16
    • Posts: 1:1 or 4:5
    • Banners: 16:9

You don’t need to understand samplers on day one. Treat advanced settings as “later” DLC.

7. Build a tiny workflow, not a giant skillset

Something like:

  1. Brainstorm idea in plain language.
  2. Turn it into a short, clear prompt: “[subject], [style], [lighting / colors], [camera or art type].”
  3. Generate 4 versions.
  4. Pick the best 1–2 and regenerate “variations” if your tool has it.
  5. Download, then tweak in Canva / phone editor (crop, text, filters).

Repeat this every time you post. You’ll accidentally get good.

8. When you’re less confused

When the tools stop feeling like magic and more like “okay, I sorta know what happens,” then Stable Diffusion or Midjourney prompt rabbit holes are worth diving into. Until then, focus on:

  • One easy tool
  • One or two favorite styles
  • One consistent workflow for your socials

You don’t need to master “AI art” in general. You only need to master “AI art that works for your posts.”

4

Skip the “perfect prompt” obsession for a second and think in workflows, not tools.

@vrijheidsvogel and @stellacadente already covered the big three (DALL·E, Midjourney, Stable Diffusion) and how to talk to them. Let me zoom in on what actually gets you from “cool idea in my head” to “usable image on my feed” with less frustration.

1. Start from your post, not from the AI

Before you even open a generator, answer:

  • Is this an Instagram post, story, banner, or YouTube thumbnail?
  • Do you need text on it?
  • Is your face or brand color involved?

Then reverse it:

  • Need text?
    Generate a clean background only, then add text later in something like Canva. Trying to get AI to render perfect words is still a lottery.

  • Need consistent vibe across a carousel?
    Reuse the same “style block” every time: same colors, same art type (e.g., “flat illustration, pastel colors, white background”) and just swap subjects.

This avoids the classic beginner mistake: a feed that looks like 10 different artists did it.

2. Use “prompt autocompletes”

Instead of reinventing prompts, build a mini library like this:

  • Photo-real “personal brand” look
    [topic], modern office, natural window light, neutral background, soft shadows, high detail, realistic photography

  • Aesthetic “quote background”
    minimalist abstract shapes, soft pastel gradient, clean design, empty space for text, high resolution

You just attach this to whatever idea you have:

  • “person working at laptop” + photo-real block
  • “self care quote background” + aesthetic block

Faster than crafting full prompts each time.

3. Steal from yourself, not from random “prompt lists”

Those mega prompt lists floating around are noisy. Better:

  1. Generate a batch of images.
  2. When something looks close to what you love, copy that exact prompt into a note.
  3. Strip out the subject so you have a generic template.

Within a week you will have 5 to 10 “house style” prompts that feel like you, rather than generic “AI art.”

4. Light disagreement: references are great, but don’t rely on them too early

You’ll see a lot of “just upload a reference and tell it what to change.”
Useful, yes, but if you lean too hard on it from day one, you never learn how style words behave on their own.

I’d split it:

  • First 30 to 40 images: no references at all, only words
    This trains your intuition: “cinematic,” “flat,” “vector,” “soft lighting,” “dramatic,” etc.

  • After that: pull in references for consistency
    For example: “Use this photo, turn it into a clean flat illustration, brand colors blue and orange, simple background.”

You get both skills: pure prompting and guided edits.

5. Treat each tool like a personality

Rough mental model that keeps expectations realistic:

  • DALL·E / Bing: “Well behaved photographer & illustrator”
    Great for clear scenes, decent faces, straightforward prompts.

  • Midjourney: “Stylish concept artist”
    Fantastic for mood, lighting, and pretty images, sometimes less literal.

  • Stable Diffusion: “Tinker-friendly engine”
    Incredible control once you know what you are doing, but early on it turns into a configuration hobby instead of an art helper.

Since you are making stuff for personal projects + socials, it is totally fine if you never touch Stable Diffusion at all.

6. Where the product title ”“ fits in

If you are compiling all this knowledge somewhere, a title like “How Do I Create Images With AI” actually helps. Pros and cons of framing it this way for your own notes, blog, or mini guide:

Pros

  • Super clear for beginners who search exactly that phrase.
  • Broad enough to cover DALL·E, Midjourney, Stable Diffusion, and whatever comes next.
  • Easy to extend into sections like “for social media,” “for personal branding,” “for thumbnails.”

Cons

  • Very generic; you may want a subtitle that narrows it, like “How Do I Create Images With AI for Social Media Posts.”
  • Competes in a search space full of generic tutorials, so your unique workflows and examples matter more than the title alone.
  • Does not hint at style or niche, so if you focus on “cozy aesthetic” or “techy infographics,” that needs to be in headings, not just the title.

Use it as a base, then add your twist in subheadings and examples.

7. How to actually get better, fast

Here is a 3-day mini plan that fits around normal life:

Day 1: Portraits & avatars

  • Pick one person (you, a friend, or a fictional character).
  • Make:
    • 3 realistic versions
    • 3 stylized (anime / comic / flat illustration)
  • Keep only your favorite 2 and save the prompts.

Day 2: Backgrounds for text posts

  • Create 10 simple backgrounds in one style for quotes.
  • Criteria:
    • Space for text.
    • Not too busy.
    • Same color palette across all.

Day 3: One full “campaign”

  • Idea: one topic, like “productivity tips” or “book recs.”
  • Generate:
    • 1 cover image for a main post.
    • 3 to 5 supporting images that match the style.
  • Assemble the posts in Canva, add real text, schedule or publish.

By the end of that you will understand way more than reading 20 tutorials. What @vrijheidsvogel and @stellacadente outlined will suddenly feel obvious instead of overwhelming, because you have your own experiments to anchor it to.