Generating Images
Ask an agent to create an image for you, right in the conversation.
You can ask an agent to create an image just by describing it in a normal message — there's no special syntax or separate button. For example: "Generate an image of a golden retriever wearing sunglasses on a beach." The agent creates the image and returns it in the chat.
Describing What You Want
Think of it like describing a photo to someone over the phone — the more concrete detail you give, the closer the result lands to what you pictured. A one-line description leaves a lot up to the agent to guess; a few extra details about the subject, setting, and style give it real anchors to work from. Descriptions can be up to about 4,000 characters, so there's plenty of room to be specific.
What to Include in Your Prompt
You don't need all of these every time, but a good description usually touches on a few of the following:
| Element | What it tells the agent | Example phrase |
|---|---|---|
| Subject | Who or what the image is actually of | "a golden retriever puppy" |
| Action or pose | What the subject is doing | "sitting on a porch, tail mid-wag" |
| Setting or background | Where the scene takes place | "on a wooden porch next to a red front door" |
| Style | The overall artistic look | "photorealistic", "watercolor painting", "flat vector illustration" |
| Lighting & mood | The feel of the light and atmosphere | "warm golden-hour light", "moody and dramatic" |
| Color palette | Which colors should dominate | "soft pastel tones", "high-contrast black and white" |
| Composition & angle | How the shot is framed | "close-up", "wide aerial view", "centered and symmetrical" |
Stacking a few of these together goes a long way — for example: "A golden retriever puppy sitting on a sunlit wooden porch, tail mid-wag, warm afternoon light, photorealistic, shallow depth of field." That single sentence covers subject, action, setting, lighting, and style all at once.
Vague vs. Detailed Prompts
A vague prompt like "draw a dog" leaves the breed, setting, pose, lighting, and style entirely up to the agent, so the result can land far from what you had in mind. Adding just a handful of details narrows that down a lot:
| Vague | Detailed | Why it helps |
|---|---|---|
| "draw a dog" | "a fluffy golden retriever puppy sitting on a sunlit porch, photorealistic" | Pins down the breed, pose, setting, and style |
| "a logo for my bakery" | "a minimalist circular logo for a bakery called 'Wheat & Co.', featuring a wheat stalk, warm brown and cream colors" | Gives a concrete subject, shape, name, and color palette to work with |
| "a picture of a city" | "a wide aerial view of a futuristic city skyline at night, neon lights reflecting on wet streets, cyberpunk style" | Adds viewpoint, time of day, mood, and style |
Aspect Ratio
You can request a specific aspect ratio by mentioning the shape or use case in your message. The agent will pick the closest fit — or you can be explicit, like "...as a 16:9 widescreen image" or "...in portrait orientation for a phone wallpaper". If you don't mention a shape, the model auto-selects the ratio that best suits your prompt.
| Ratio | Orientation | Best for |
|---|---|---|
1 : 1 | Square | Profile pictures, icons, product thumbnails, social media posts |
16 : 9 | Landscape | Widescreen banners, YouTube thumbnails, presentation slides, desktop wallpapers |
9 : 16 | Portrait | Instagram Stories, TikTok covers, mobile wallpapers, Reels thumbnails |
4 : 3 | Landscape | Presentations, blog post headers, classic photo prints |
3 : 4 | Portrait | Portrait photography, Pinterest pins, printed posters |
3 : 2 | Landscape | DSLR-style photography, editorial images, print layouts |
2 : 3 | Portrait | Book covers, event posters, full-body portraits |
2 : 1 | Wide landscape | Website hero banners, email headers, LinkedIn cover photos |
1 : 2 | Tall portrait | Infographic panels, mobile app splash screens |
19.5 : 9 | Ultra-wide landscape | Modern smartphone display backgrounds (landscape) |
9 : 19.5 | Ultra-tall portrait | Modern smartphone display backgrounds (portrait) |
20 : 9 | Cinema wide | Ultra-wide monitor wallpapers, cinematic scenes |
9 : 20 | Cinema tall | Tall digital signage, narrow banner ads |
auto | Auto | Model picks the best ratio for your prompt — the default if you don't specify |
For social content, 9:16 is the go-to for Stories and Reels, 1:1 for feed posts, and 16:9 for YouTube. For print, 3:2 or 4:3 matches standard photo paper sizes. When in doubt, leave it as auto and the model will choose a ratio that complements your prompt.
Resolution
Resolution controls how large and detailed the output image is. Two tiers are available:
| Resolution | Pixel density | When to use it |
|---|---|---|
1k (default) | ~1024 px on the long edge | Chat previews, quick mockups, social media drafts, internal use |
2k | ~2048 px on the long edge | Print materials, large banners, high-fidelity marketing assets, detailed artwork |
To request higher resolution, mention it in your message — for example: "...generate this in high resolution" or "...I need a print-quality version". If you don't specify, the default 1k resolution is used.
Higher resolution images take slightly longer to generate and may use more credits. For most chat and preview use cases, 1k is more than sufficient.
Prompt Examples by Use Case
Here are some ready-to-use prompt patterns for common scenarios. Copy and adapt them to your own subject:
| Use case | Example prompt |
|---|---|
| Social media post | "A flat-lay of a coffee cup and a croissant on a marble table, overhead shot, warm natural light, 1:1 square format" |
| Website hero banner | "A wide panoramic view of a mountain range at sunrise, soft mist in the valley, photorealistic, 16:9 landscape" |
| Product thumbnail | "A minimalist product shot of a navy blue water bottle on a white background, soft studio lighting, square format" |
| Phone wallpaper | "A serene Japanese garden at dusk, cherry blossoms falling, lantern light reflecting on a koi pond, 9:16 portrait" |
| Book cover | "A dramatic oil-painting style cover for a mystery novel — a fog-covered Victorian street, lone figure under a gas lamp, 2:3 portrait" |
| Illustration / icon | "A flat vector illustration of a rocket ship launching from a stylized planet, bold colors, minimal detail, square format" |
| Presentation slide | "An abstract geometric background for a tech presentation — deep navy blue, subtle gold grid lines, 16:9 widescreen" |
Style Reference Guide
Adding a style keyword to your prompt is one of the fastest ways to steer the look of an image. Here are common styles and what they produce:
| Style keyword | What it looks like |
|---|---|
| photorealistic | Looks like a real photograph — sharp, detailed, true-to-life |
| cinematic | Movie-still quality — dramatic lighting, shallow depth of field, film grain |
| watercolor painting | Soft, translucent washes of color with visible brush texture |
| oil painting | Rich, textured paint strokes — classic fine-art look |
| flat vector illustration | Clean, geometric shapes with bold fills — icon or logo style |
| pixel art | Retro 8-bit or 16-bit style with visible pixels and limited palette |
| pencil sketch | Hand-drawn look with visible graphite lines and shading |
| anime / manga | Japanese animation style — large eyes, expressive linework, vibrant colors |
| cyberpunk | Neon lights, dark cityscapes, futuristic dystopia aesthetic |
| minimalist | Very little detail — lots of negative space, simple shapes, muted palette |
| isometric 3D | Flat 3D perspective — common in UI mockups and infographics |
| stencil street art | Bold outlines and spray-paint texture — urban graffiti style |
You can combine styles for hybrid results — for example: "a cyberpunk cityscape in watercolor painting style" or "a minimalist flat vector illustration of a mountain, cinematic lighting."
How the Image Arrives
A generated image shows up in the chat the same way any agent-created file does — as a document card that previews inline and can be opened full-size. See Attachments & File Outputs for how these cards work. The image is also saved automatically to your workspace's Files, under the AI Generated tab, so you can find it again later even outside the conversation.
Content Moderation
All generated images pass through automatic content moderation before they are returned. If a prompt is flagged — for example, because it requests violent, explicit, or otherwise policy-violating content — the image is blocked and the agent will tell you rather than returning an unrelated image. Rephrasing the prompt or choosing a different subject will usually resolve it.
When It Isn't Available
Image generation only works when all of the following are true:
| Requirement | Where it's controlled |
|---|---|
| The agent is allowed to generate images | The Generate images toggle on the agent's Permissions tab |
| Your workspace's plan includes image generation | See Your Plan & Topping Up |
| An image model is configured by your admin | Admin → LLM Providers → Default image model |
If any of these isn't the case, the agent tells you it can't create the image instead of making something up.
Generating an image uses credits like any other agent action — see How Credits Work for how usage is billed.