SundaySky · Lead UX/UI Designer, AI & creative features
AI image generator
From stock limitations to smart AI: visuals that fit your story.
Pendo usage data · last 90 days
Goal
Enable users to instantly generate AI-based image placeholders directly in the platform, keeping them in flow without searching for external assets.
The challenge
Video creators struggled with quick access to branded visuals. Uploading images was a hurdle, Getty assets often missed the mark, and users paused creation because a visual was missing. We needed a lightweight, brand-consistent way to reduce that friction.
Discovery
3 inputs01. Target users · persona
Video creators who need a visual now, not a search result later.
Video creators needed quick visual assets to keep their projects moving. They were often blocked by missing images, wanted fast inspiration, and preferred simple flows over complex creative tools.
02. Benchmark & inspiration
Powerful tools everywhere, all of them too heavy for a quick creation flow.
We reviewed AI visual tools including Canva, Microsoft Designer, Wix, Tome and Adobe Express. Most were powerful but too complex for our needs, which pointed to a minimal, lightweight beta that fitted our platform's quick creation flow.
03. User interviews
Users paused creation when a visual was missing, and blamed the library, not themselves.
- Users often paused creation because a visual was missing.
- Uploading their own images felt like a blocker.
- Users couldn't find what they were looking for in Getty Images.
- Users wanted inspiration and control.
- Style variety mattered, but it had to be easy to choose.
What discovery told us
- Users paused work when visuals were missing.
- Uploading images felt slow and disruptive.
- Getty search was hard and often irrelevant.
- AI lowered friction and felt easy to try.
- Users expected brand-aligned results: icons, video, illustrations.
What made it hard
- Writing prompts was intimidating.
- The AI Copilot button was hard to find.
- Results were inconsistent.
- Limited asset types were available.
- A hard balance between simplicity and control.
Design
Flow and 3 solutionsMain user flow
Prompt, generate, choose, place, without leaving the scene.
Design system
A modular panel that behaved like the rest of the platform.
- A modular UI panel, consistent with the platform.
- Visual style presets.
- Lightweight interactions, no page reloads or deep modals.
- Auto-fallbacks for failed generations.
- Reusing the existing Copilot in a secondary window rather than a progressive prompt panel.
Solution
Solution 01. Image generator dialog
Decision: presets instead of a blank prompt box
The dialog pairs a prompt field with visual style presets (close-up, 3D illustration, photograph, black and white, cinematic, icon, aerial), so a user who doesn't know how to write a prompt can still get a usable result. Generated images can be added straight to a folder or to Your Media.
Solution 02. Copilot prompt assistant
Decision: reuse the Copilot rather than build a prompt panel
The existing Copilot writes and refines the prompt conversationally, then inserts it, so prompt-writing stops being a blank-page problem. It reused a component users already had rather than adding another one to learn.
Solution 03. Fit to placeholder, save to library
Decision: land the result in the scene, not in a download
The chosen image fits the placeholder directly on the canvas and can be saved to the library for reuse: keeping the user in the creation flow rather than sending them out to manage files.
Analyse
3 measures01. Adoption & usage · Pendo
Strong adoption at launch, then a slide, because the results weren't good enough to come back for.
Initial adoption showed strong interest, but usage slowed over time as users found results less relevant. “Regenerate” was the top action. Only 49 users (5.4%) used the Copilot, though it worked well when they did. Few users saved images to the library.
Takeaway. Users wanted quick retries, but missed the Copilot and didn't reuse images, because it was hard to find and didn't match the product.
Pendo usage data · last 90 days
02. User journeys · Pendo path analytics
Most users completed the flow, then stopped at the preview.
Most users who started the flow completed it, with drop-off concentrated after the image preview. A short median session time indicates focused intent. Conclusion: the flow is intuitive; the problem is dissatisfaction with the generated results.
Pendo path analysis. Click to enlarge
Pendo path analytics
03. User feedback · post-release interviews
Easy to use, confirmed. The Copilot felt hidden.
Post-release interviews confirmed ease of use, and that the Copilot felt hidden and underused. Three requests came up repeatedly:
- Generate directly from the Copilot.
- Access previous generations.
- Preview images in the context of the video.
Interview findings. Click to enlarge
Post-release interviews
Takeaway
Users liked the feature but wanted clearer, stronger results that match the brand. The flow was not the problem; the output was. That reframed the next round of work around result quality, Copilot discoverability and reuse, rather than around the interface.