Skip to main content
AI Designer

Generate an Image

Create a single-image design (post, ad, thumbnail, banner) from a text prompt with Krumzi's AI Designer.

Written By Andrei Terteci

Last updated 10 days ago

Single images are the bread and butter of Krumzi: Instagram posts, ads, YouTube thumbnails, LinkedIn graphics, banners, quote cards. Describe it, pick a size, send. Here's the full flow.

Before you begin

  • Image design comes with every plan, Starter included.

  • Credits are usage-based: you're charged after the turn for the work the AI actually did. A first generation costs more than a small tweak; complex briefs cost more than simple ones.

Generate your image

  1. Open AI Designer from the sidebar.

  2. In the type selector, switch from Auto to Image. On Auto, Krumzi picks the format and size from what you write, so pick Image when you want to choose the size yourself.

  3. Pick a size. Open the size picker to see 44 presets grouped by platform (Instagram, Facebook, LinkedIn, X, YouTube, TikTok, Discord, Pinterest, Bluesky, Product Hunt), each with a little aspect-ratio icon and its pixel dimensions. Or use the custom size bar at the bottom to type any width and height, with an aspect-ratio lock and an orientation swap.

  4. Set your options. Toggle Brand Kit if you have one, attach any photos or logos you want placed, and turn on Fast if you want the design back quicker (about 2x the credits for that turn, on Pro, Studio and Agency). For a design that has to be right, turn on Smart, next to Fast: Krumzi's most capable model plans the idea and builds the design, for about 2 to 3x the credits. Smart is on Pro, Studio and Agency, and it starts switched on during a free trial. See "Use Smart Mode for Your Hardest Designs". There's no model to choose.

  5. Describe the design and send. For example:

New seasonal menu post for a coffee shop. Headline "Fall Flavors Are Here", subtext "Pumpkin Spice Β· Maple Latte Β· Cinnamon Mocha". Warm, cozy, one big latte photo.

What you'll see

Your design opens in the chat editor, and the AI gets to work in the chat on the left. On a new design it decides the idea before it builds anything: the concept, the wording and the main visual. While it thinks, the status line reads Thinking it through and its thoughts scroll in a small box underneath, ending with a line that starts The idea:. That happens with Smart on or off.

You'll watch it plan the concept: like a designer, it sketches three different ideas, drops the weaker ones and builds the strongest. Then it adds elements to the canvas step by step (headline, image, shapes, background).

At the end you watch it look over its own work. The status reads Looking over your design, and any problem it finds gets a tag on the canvas, like Hard to read, Overlapping or Cut off. Then comes Fixing what I spotted, each tag flips to Fixed, and it takes One last look. If there's nothing to fix, you just see the look. When the turn ends, the thoughts fold into a closed Thought for 40s line (with your own number of seconds) that you can click to open.

When it's done, the reply tells you what it made and offers the idea that came second in one line, like "Want the type-led version instead?" Say yes and it reworks the design in that direction (your first version stays in your versions). No brand set up yet? Krumzi picks a look that fits your brief and names it in the reply, so you can steer it. You can press Stop at any point and keep whatever it's built so far.

AI photos arrive while it works. When your design needs a photo the AI generates, Krumzi doesn't sit and wait for it. It lays out the text and shapes while the photo is being made (usually 15 to 60 seconds), and the photo appears in place when it's ready. If a photo can't be made in time (90 seconds), Krumzi replaces it or adjusts the layout, and a photo that didn't arrive is never charged. If one fails right at the end, Krumzi tells you it left it out and didn't charge you. Pressing Stop also stops a photo that's still being made, at no charge.

What the AI can put on the canvas

  • Text with real typography choices from the full Google Fonts library, including mixed-weight headlines.

  • Images: your uploads, stock photos, AI-generated artwork (5 credits per image on every plan, charged only when it arrives), or real images pulled from a URL you pasted. Any image can get its background removed (5 credits) or be clipped to a circle, pill, arch, or blob.

  • Shapes, icons, emojis, QR codes, and annotations like arrows and highlights.

  • Backgrounds: solid, gradient, mesh, transparent, or a full-bleed photo with an overlay for legibility.

  • Effects: shadows, glows, blur, glass, filters, and blend modes.

Every one of those is a normal editable element afterward. Nothing is baked into a flat image.

After generation

Refine in chat

Keep talking: "make the headline bigger", "try a darker background", "move the logo bottom-right". Every change the AI makes saves automatically, just like your own edits, and every reply is kept as a version you can go back to. See "Refine Your Design".

Edit by hand

Click any element to move, resize or rotate it, and double-click a text to retype it. Colors, fonts and removing things are a request to Krumzi. For panels, layers and exact control, click Open full editor in the header on a laptop.

Export

Hit Download (it's called Export in the full editor) and choose PNG, JPG, WebP, or PDF, at 1Γ— to 4Γ— the canvas size, with an optional transparent background for PNG and WebP.

Tips

  • Give the AI your real headline. It designs around the words, so placeholder copy produces placeholder layouts.

  • Attach the actual product photo instead of describing it. The AI can see it and place it. Krumzi won't generate a look-alike of your own building, team or product, because people who know the real one would notice. Without your photo it designs around the words or uses atmosphere, and it may ask you for the photo.

  • Making several posts in one style? Turn on Batch and let the AI produce up to 10 variations from one brief.

  • Set up Brand Mind once and every image starts on-brand without you repeating colors and fonts.