Leonardo AI

Leonardo.Ai built its identity around giving creative professionals fine-grained control over image generation that general-purpose tools did not offer — fine-tuned models trained on specific art styles, ControlNet-style pose and depth conditioning, canvas tools for inpainting and outpainting, and a model training pipeline that lets users teach the system their own visual aesthetic. The platform launched in late 2022 and grew rapidly through a community-oriented model where users could share their trained models and generation presets with other members. That social layer turned Leonardo into more than a generation tool; it became a marketplace of creative workflows, with popular user-trained models accumulating thousands of downloads.

The technical stack reflects that diversity of features. A single generation request on Leonardo might pass through a custom fine-tuned model checkpoint, apply one or more ControlNet conditioning signals, run through the core diffusion sampling loop across dozens of steps, and then apply upscaling in a second pass. Each of these stages is GPU-bound, and the platform serves them across a distributed cluster that must handle both the latency-sensitive interactive web app and the throughput-oriented API tier simultaneously. The model zoo — many of which are community-contributed — adds a model-loading and caching dimension that simpler platforms running a single model do not have to manage.

Infrastructure problems at Leonardo manifest in ways that reflect this complexity. The image generation queue can back up during peak hours, with tokens debited from a user's account but the generation never completing. The fine-tuning pipeline — where users submit datasets to train custom models — is particularly sensitive to GPU availability, and training jobs can be paused or delayed for hours when compute is constrained. The API may return successful job-submission responses followed by webhook events that never fire. Canvas operations like inpainting can time out mid-edit, leaving the canvas in an inconsistent state. In severe incidents, the web app loads but the generation modal returns blank outputs with no error message.

Outage.gg aggregates real-time reports from Leonardo.Ai users. If your generations are failing, training jobs are stalled, or the API is not delivering results, the live status page shows whether the issue is widespread or localised.