Configure Media Models and Shared API Keys
Desktop Settings → Media Generation manages image and video routes, provider credentials, and local model discovery. A provider can reuse its supported AI Services key or use a separate Media Generation key.
Check the provider’s credential source
Open Media Generation → API Providers and find the provider you want. An AI Services Key badge means the app is using a supported key already configured in AI Services. You do not need to copy that key into this panel. If no credential is available, configure the provider before requesting generation.
Use a separate media key when needed
Add a key in the provider’s Media Generation field when you intentionally want to override the shared AI Services key. Review the displayed credential source after saving. An available key does not guarantee access to every model; provider permissions and account access still determine which requests can run.
Choose a model for the intended operation
In the image or video settings, choose the generation route and provider, then a model that supports the operation you need. Image generation, image editing, text-to-video, and image-to-video may offer different models and controls. Set the size, aspect ratio, and other options offered by the selected model.
Refresh the model catalog
Use Refresh models from API where available to recheck the provider’s catalog. For local models, use Refresh Models to scan installed models and any additional model folder. Managed downloads and complete models in the Hugging Face cache can be discovered without downloading another copy.
Match local models to their runtime
Check whether a model is available through Local or ComfyUI. A model detected for ComfyUI may require Install for Local before it can use the separate Local route. If a model disappears after moving files, check the optional model folder and refresh again. Review any loading or refresh error before changing the selected route.
Generate a small sample
Return to a generation workflow and request a small sample using the chosen route. Check the result and any reported provider error. If the request fails, verify the credential source, model access, and the runtime or local-model status. Save a working default only after confirming it produces the output you need.