I'm trying out Leonardo AI for some design mockups at my new job. The image generation is pretty cool, but I've hit a wall trying to export assets for our devs.
When I need to hand off a UI component, the export formats feel limited. I couldn't find options for things like SVG with clean layers, or specific resolutions for different devices. Is this a common issue? How are other teams handling this workflow, especially if you're integrating these assets into a live project?
For a pro tool, I was expecting more control over the output. Maybe I'm missing a setting?
Still learning
The export limitations you're hitting are a common pain point when moving from an AI generation tool into a production pipeline. The core issue is that these platforms are optimized for creation, not for asset pipeline integration. You aren't missing a setting.
For a professional workflow, you'll likely need to use an intermediary tool. I process all generated assets through a dedicated vector editor to rebuild the SVG structure and enforce layer naming conventions before export. For multiple resolutions, a simple script using ImageMagick or Sharp to batch generate sizes from the highest-quality output is far more reliable than relying on the AI platform's built-in options.
This adds a step, but it gives you deterministic control over the final output. What's your current handoff process like with your developers? Do they require specific naming conventions or folder structures for the assets?
Yep, that's the wall everyone hits. These AI image tools are fantastic ideation engines, but their export is basically a "save as" feature. You're not missing anything.
Think of the raw output as a first draft. You'd never hand a developer a messy PSD without layers labeled, right? Same principle. We run everything through a quick Figma cleanup for SVG structure and layer sanity before it touches the repo.
The built-in resolution presets are also notoriously unreliable for exact pixel-perfect needs. We just take the highest quality PNG it can spit out and feed it into a simple script. It's an extra step, but at least the output is deterministic.