Visual agents

When words alone aren’t enough

Most of the time, the right output is text — a draft, a summary, a plan. But sometimes the work you need is visual: a slide image, a chart for a report, a mockup of a layout idea.

VDF AI ships three agents purpose-built for visual work. Each one specializes in a different kind of artifact. This page is a short guide to getting great results from all three.

Three agents, three jobs

Image Generator

What it’s for

Producing an image from a description, in seconds. Common uses:

• A hero image for a blog post or slide.
• Marketing-page illustrations.
• Internal docs that benefit from a visual.
• Concept art for an early product idea.
• Social-media imagery.

What a good prompt looks like

A good image prompt names four things:

1. Subject. What’s in the image.
2. Composition. How the subject is arranged.
3. Style. Photographic, illustrated, flat, painterly, technical.
4. Mood or palette. The feeling, or the color direction.

A short example:

A wide illustration of a small team gathered around a glowing laptop in a softly lit office. Flat, modern editorial illustration style. Warm palette, no faces visible.

You can compress this — “a flat editorial illustration of a small team around a laptop, warm palette” — but the more elements you name, the more consistent the result.

Patterns that produce sharper images

• Name what you don’t want. “No text in the image.” “No faces.” “No corporate stock-photo feel.” The Image Generator listens to negatives well.
• Specify orientation. Wide, tall, square. Saves a regeneration when you need a specific aspect.
• Use reference cues for style. “Editorial illustration.” “Flat geometric.” “Soft photographic.” Style cues are powerful.
• Iterate small. First prompt sets the shape. Refinement prompts adjust palette, mood, composition.

What to expect

Image generation takes a few seconds to a minute, depending on the size and the model. You’ll usually see one image. If you ask for variations, you can get several at once.

The Image Generator is good at:

• Editorial and illustrative styles.
• Conceptual imagery.
• Abstract or symbolic visuals.

It’s less good at:

• Photo-realistic depictions of specific real people or brands.
• Text inside images (use a layout tool for that).
• Highly technical diagrams (use Chart Generator or HTML Mockup Generator).

Chart Generator

What it’s for

Turning data into a chart. The Chart Generator picks the chart type that fits the data, unless you tell it otherwise. Common uses:

• A trend chart for a report.
• A breakdown chart for a presentation.
• A comparison chart for analysis.
• Multiple charts at once for a dashboard-style summary.

What a good prompt looks like

A good chart prompt names three things:

1. The data. Either attached as a file or pasted into the prompt.
2. What the chart should show. “Monthly revenue trend.” “Customer count by segment.”
3. Optional: the chart type. “As a bar chart.” “As a line chart with markers.”

A short example:

Here’s our monthly customer count for the last 12 months. Produce a line chart with markers showing the trend. Highlight the three months with the steepest growth. Title: “Customer growth — past 12 months.”

The Chart Generator handles a lot from the data alone, but explicit framing (what to highlight, what to title) produces a more usable chart.

Patterns that produce sharper charts

• Provide clean data. A table with clear column headers beats a CSV dump of messy export.
• Name the takeaway, not just the data. “Show the contrast between Q1 and Q4” produces a sharper chart than “make a chart of this.”
• Let the agent pick the type when you’re unsure. It usually picks well, and you can always ask for a different type after.
• Iterate on style. First get the data right, then adjust labels, colors, and emphasis.

What to expect

You get a chart you can preview, download, and embed. Most charts are produced in a few seconds.

The Chart Generator is good at:

• Standard chart types — bar, line, pie, scatter, stacked.
• Multi-series charts.
• Charts with annotations and highlights.

It’s less good at:

• Highly custom layouts (use a design tool).
• Interactive dashboards (the chart is a static image).
• Very large datasets (millions of points; aggregate first).

HTML Mockup Generator

What it’s for

Producing a clickable HTML mockup of a layout idea. Common uses:

• Sketching a marketing landing page before sending to design.
• Showing a stakeholder what a feature might look like.
• A clickable prototype for an internal tool.
• A table-of-contents page for a multi-section document.

What a good prompt looks like

A good mockup prompt names four things:

1. The page’s purpose. What is this layout for.
2. The main sections. What goes on the page, in what order.
3. The visual style. Modern, minimal, dense, playful.
4. The audience. Marketers, developers, executives.

A short example:

A clickable landing page for an internal tool that helps engineers manage on-call schedules. Sections: hero with a clear value statement, three feature highlights, a simple onboarding form. Style: clean, dense, technical. Audience: engineers.

The HTML Mockup Generator can take a lot more prompt than the other two visual agents. Be generous with detail.

Patterns that produce sharper mockups

• Describe the page in sections, top to bottom. “Hero, three columns of features, a CTA, a footer.” The agent follows the structure.
• Specify what the user should be able to click. “The CTA should link to a placeholder confirmation page.” Mockups are clickable; tell the agent what to link.
• Reference a style you know. “Like a modern SaaS landing page.” Style references are powerful here too.
• Iterate at the page level, then at the section level. First get the page structure right. Then refine individual sections.

What to expect

You get an HTML file you can preview in a browser, download, and share. The mockup is clickable but not fully functional — it’s for design and review, not for production.

The HTML Mockup Generator is good at:

• Landing pages, dashboards, marketing pages.
• Multi-section layouts with clear visual hierarchy.
• Quick design exploration before committing to a real implementation.

It’s less good at:

• Pixel-perfect implementations of a specific design system.
• Complex interactive applications.
• Anything that needs to actually work — it’s a mockup.

A common workflow that combines all three

For a polished internal proposal:

1. Use the Image Generator for a hero image.

   An editorial illustration that captures the proposal's mood.

2. Use the Chart Generator for the data argument.

   One or two charts that make the numerical case.

3. Use the HTML Mockup Generator for the "what we'd build" section.

   A clickable layout showing the proposed feature.

4. Compose the document with all three.

   Image at the top, charts in the middle, mockup at the end. A proposal that reads like it took a week.

A few patterns that work for all visual agents

Iterate, don’t re-prompt

When the first output isn’t right, ask for a specific change rather than starting a new prompt. “Make the palette warmer.” “Change the chart type to bar.” “Move the CTA to the top of the page.” Refinements are faster than restarts.

Don’t over-describe

A prompt with twenty constraints often produces a worse output than a prompt with five. Pick the constraints that actually matter to you.

Save outputs as you go

When a visual is “good enough to ship,” save it. The next refinement might overshoot.

Pair visual outputs with text

A great image without context isn’t useful. A great chart without a takeaway sentence underneath isn’t useful. Always pair visual output with one or two sentences of text that frame what the visual shows.

Where to go next

• Agent library — the full catalog including visual agents.
• Chatting with agents — refinement patterns that apply to visual work too.
• Tool catalog — the underlying generation tools.
• Creating your own agent — combine visual generation with team-specific knowledge.