Leonardo adds SDXL 0.9 and Prompt Magic v3

Generating an AI image that actually looks like a finished piece of art usually requires two things: fighting your prompt window and cleaning up weird textures in Photoshop afterward.

Leonardo AI just rolled out a major update to fix that workflow, integrating Stability AI's SDXL 0.9 alongside its own Prompt Magic v3 pipeline, plus handing out 2,000 token credits to sweeten the deal.

At the core of the upgrade is native 1024x1024 base resolution, aiming to kill off the blurry edges and awkward compositions that plagued earlier open-source generations.

Two encoders, two stages

The real upgrade here is under the hood. SDXL 0.9 uses two pre-trained text encoders at once, OpenCLIP-ViT/G and CLIP-ViT/L. That allows the engine to parse complex, layered prompts far more accurately without losing the plot.

Instead of trying to dump everything onto the canvas in a single pass, the model uses a two-stage pipeline. The first stage generates the latent structure of the image at target dimensions, and a dedicated high-resolution refinement module takes over to inject fine texture, skin micro-details, and realistic lighting.

Leonardo's Prompt Magic v3 ties into this system directly. It acts as an orchestrator across existing models, tweaking dynamic range and color accuracy so you do not have to write twenty negative prompts just to get natural shadows.

Because SDXL 0.9 is trained natively at 1024x1024, it also preserves coherence across multiple aspect ratios without distorting anatomy or perspective.

Why it matters

For digital artists, concept designers, and creators who do not want to manage local ComfyUI or WebUI setups on high-end GPUs, this is a very practical upgrade. Getting production-grade detail and stable 1024-pixel resolution straight from a browser canvas lowers the barrier to entry significantly, and the 2,000 free tokens make it an easy experiment to test right now.

Strategically, Leonardo AI is playing this smart. Instead of burning millions trying to train a foundation model from scratch to challenge Midjourney, it is focusing on pipeline orchestration and user interface. By slapping its proprietary Prompt Magic layer on top of Stability AI's bleeding-edge open weights, Leonardo gets the best of both worlds.

The real question is sustainability. Two-stage generation and dual encoders require serious compute, which naturally translates into higher token consumption once the free tier runs dry. If pricing scales too steeply, power users might eventually drift back to self-hosted SDXL rigs.