The Real Story Of [Bug]

The sudden rejection of agents.defaults.imageGenerationModel during OpenClaw doctor validation isn’t just a quirk - it’s a full-blown roadblock. Developers once set primary image models like xai-images/grok-imagine-image-pro with a clean config, but now the tool strips the key entirely, forcing inference fallbacks. This means no custom models get a seat at the generation table.

At its core, this bug contradicts OpenClaw’s documented config spec, where imageGenerationModel under agents.defaults is a validated, required field. The fix command - doctor --fix - removes the key but leaves no trace, breaking a core feature.

Here is the deal: the error shows up as ‘Unrecognized key’ on validation, yet the model still works if you skip doctor. The fix cleans the config but strips what users explicitly set - like choosing a preferred model.

But here’s the catch: OpenClaw 2026.3.13 ignores the official model definitions in some cases, even when providers like xai-images are fully configured. This creates a silent failure: no image generation, no fallback, no warning besides a blank output.

Despite this, the model itself still exists and functions if you bypass the doctor step. Yet the documentation promises stability - this breaks trust.

Is your model really missing? Run openclaw doctor, then --fix, but know the key may vanish. To keep control, define imageGenerationModel explicitly in your config.

Are you letting configuration quirks silently kill your creative flow?