Just saw the announcement about vector generation in Firefly. I've been testing it this morning, and it's a significant step, though with some clear constraints right out of the gate.
My initial tests show it's solid for generating simple logos, icons, and basic shapes with clean lines. The output is genuinely scalable SVG, which is the main point. However, I noticed a few things:
* The control is pretty basic compared to dedicated vector tools. You're heavily reliant on prompt engineering to get specific styles.
* Complex illustrations with layered details can get messy. It seems to work best for concepts that are inherently graphic and simple.
* The integration with Adobe Express is smooth, which is great for quick social media graphics or draft assets.
Has anyone else tried pushing it beyond simple icons? I'm particularly curious about:
* Using the output in automated design workflows. Can we reliably pull these SVGs via API for batch processing?
* How well does it handle typography or lettering within the vector?
* Any workarounds for getting more consistent, repeatable styles?
It feels like a powerful feature waiting for more granular controls. For no-code automation, having a text-to-vector endpoint could be a game-changer for generating dynamic marketing assets.
Your point about automated workflows and the API is a good one. I've been running tests specifically on that. I can confirm the API does return clean SVG markup, but there's a reliability issue for batch jobs: the style consistency you mentioned isn't there yet. Two identical prompts can yield vectors with different path structures, which breaks downstream processing if you're automating asset generation.
I tried the typography. It's predictably weak. The system treats text as just another shape to approximate, so letterforms are often malformed or merged. For any logo requiring precise typography, you're still better off handling that layer separately in Illustrator.
What we're seeing is a strong v1 for ideation, but not for production pipelines. The lack of seed control or style anchors means you can't guarantee repeatable outputs, which is a non-starter for system integration. It feels like they prioritized the scalable output format over the parametric controls needed for serious use.
data is the product