How to generate images with AI

AI can make an image from just a description. Type "watercolor painting of a fox in a library" and you get one in about 10 seconds. You can use this for greeting cards, social media posts, illustrations for a blog, or just to play around. Most of the popular tools are free to try.

Where to make AI images (free and paid)

Google Gemini (free, most generous)

• Go to gemini.google.com. Sign in with Google.
• Type "create an image of..." or "make a picture of..."
• Gemini uses its Imagen model. Quality is excellent.
• Free tier is fairly generous (daily limits but high).

ChatGPT (free with limits)

• Go to chatgpt.com. Sign in.
• Type "create an image of..." in any chat.
• Uses OpenAI's image generator (the latest version is excellent).
• Free tier limited to a few images per day. Plus ($20/month) raises the limit significantly.

Microsoft Copilot (free)

• Go to copilot.microsoft.com. Sign in with Microsoft.
• Click the image icon and describe your image.
• Uses OpenAI's image model under the hood. Free with daily image credits.
• Bing Image Creator is the same service from a different URL.

Midjourney (paid, $10/mo and up)

• Best for serious image makers. Goes through Discord (a chat app) or the new web app.
• Most artistic results. Steeper learning curve.
• No free tier as of 2026.

Adobe Firefly (free trial then paid)

• Adobe's commercial-safe image generator. Trained only on licensed content.
• Free trial credits, then bundled with Adobe Creative Cloud.
• Best for business use because Adobe takes responsibility for the source content.

Claude

Claude does not generate images. It can analyze images you upload, but if you want to create one, use a different tool.

Step 1: Make your first image

1. Open Gemini, ChatGPT, or Copilot.
2. Type a prompt. Example: "Create a watercolor painting of a small fishing boat at sunset on Monterey Bay."
3. Wait 5 to 20 seconds.
4. Click the image to view full size. Right click (or long press on phone) and choose Save to download.

Step 2: Write good image prompts

Image prompts work differently than text prompts. The trick is being descriptive about both what is in the image and how it looks.

The recipe

[subject] + [setting / background] + [style] + [mood / lighting] + [camera or art details]

Example: building a good prompt step by step

• Bad: "A dog."
• Better: "A golden retriever puppy."
• Better still: "A golden retriever puppy lying on a wooden porch."
• Best: "A golden retriever puppy lying on a sunny wooden porch in late afternoon, watercolor painting style, soft pastel colors, warm and cozy mood."

Useful style words to know

• Painting styles: watercolor, oil painting, acrylic, pencil sketch, ink drawing, charcoal
• Eras: 1950s illustration, vintage photo, Art Deco, Art Nouveau, mid-century modern
• Famous styles: "in the style of Studio Ghibli," "in the style of a Pixar movie," "like a children's book illustration"
• Photo styles: photorealistic, 35mm film, golden hour, soft focus, professional headshot
• Cartoon styles: cartoon, comic book, Saturday morning cartoon, anime
• Mood: cozy, dramatic, peaceful, mysterious, cheerful, melancholy

Step 3: Iterate

Just like with text, the first image is rarely the best one. Try:

• "Same image but more colorful."
• "Try again with the dog smaller and the porch wider."
• "Make the lighting warmer."
• "Try one more, completely different style."
• "Change the puppy to a black lab."

You usually need 2 to 4 tries to land on something you love.

Common uses worth knowing about

Greeting cards and invitations

"Create a birthday card cover for a 5 year old who loves trucks. Watercolor style, cheerful colors."

Social media posts

"Create a square Instagram post graphic with the words 'Open Saturdays 9am to 4pm' over a friendly cartoon illustration of a coffee mug. Warm coffee shop colors."

Blog post illustrations

"Create a friendly illustration of an older man and his small dog sitting in a sunny living room reading a newspaper. Watercolor style."

Visualizing ideas

"I want to remodel my kitchen with butcher block counters and sage green cabinets. Show me what that could look like."

Logo brainstorming (not final logos)

"Suggest 4 logo concepts for a small dog grooming business called Paws and Bubbles. Simple, friendly, no text."

For an actual final business logo, get a designer. AI logo ideas are good starting points only.

What AI image generators are bad at

• Text inside images. Has gotten much better but still misspells often. For any image with words, double check.
• Hands and fingers. Better than 2023 but still occasionally weird. Look for 6 fingers, fused hands.
• Specific real people. Most tools refuse to generate images of named celebrities or public figures.
• Exact replicas. "Make my dog look exactly like this photo" is hit or miss. Style and composition transfers well; exact resemblance does not.
• Specific brand logos. Most tools refuse or produce off-brand approximations.

Editing an image you already made

Once you have an image you mostly like, ask for changes:

• "Same image but change the sky to sunset."
• "Remove the person in the background."
• "Add a small dog sitting next to the woman."
• "Change the dress color from blue to green."

ChatGPT and Gemini both handle these edits well in the same chat. Adobe Firefly has the most precise editing tools.

Can you use AI images commercially?

The terms of service for ChatGPT, Gemini, Copilot, and Firefly all grant you commercial use of images you generate. Adobe Firefly explicitly indemnifies (legally protects) you because it was trained on licensed content.

Caution areas:

• Real people. Generating images of real celebrities or public figures for commercial use is a legal risk.
• Copyrighted characters. Do not generate Mickey Mouse, Marvel characters, etc. for business use.
• Trademark logos. Do not approximate someone else's logo.
• Copyright registration. US Copyright Office currently says AI-only images cannot be copyrighted by you. If you need exclusive rights, get human-made art.

Common beginner mistakes

• Prompts too vague. "A landscape" gets a generic landscape. "A coastal landscape near Big Sur at golden hour, photorealistic, soft focus" gets something good.
• Giving up after one image. Make 3 to 5 variations. Pick the best. Edit from there.
• Trying to put text in images. Better to add text afterward in Canva, Photos, or any image editor.
• Generating images of real people. Both legally and ethically questionable. Tools often refuse, and you should not be asking.

Where to put your AI images after

• Print and frame: save the highest resolution version, send to Walgreens / Costco / Mpix.
• Social media: save and upload, but stick within each platform's allowed sizes.
• A printed book or report: upgrade to a paid tier for high-resolution downloads (free tiers often output lower resolution).
• Editing: open in Canva (free) to add text, crop, or combine with other elements.

Video walkthrough

Video by Kyle Behrend on YouTube