[enhancement]: Access to closed-source models #8633
Description
Is there an existing issue for this?
- I have searched the existing issues
Contact Details
No response
What should this feature add?
This feature should add the ability to configure and use external API providers (such as fal.ai, Replicate, and similar services) within InvokeAI to access closed-source models like nano-banana, Kontext Pro, and other proprietary models.
Users should be able to:
- Add and manage API keys for external providers through the InvokeAI interface
- Select external models from these providers alongside locally installed models in the UI
- Configure provider-specific settings if any (model parameters)
- Seamlessly generate images using external models with the same workflow and canvas features available for local models
- View model metadata and capabilities for external providers
This would essentially restore part of the functionalities that was previously available through InvokeAI's paid tier, but allow users to bring their own API credentials from providers of their choice.
The implementation should focus on provider-level API key management rather than model-specific keys. While it could theoretically support individual API keys per model (one for Google, one for OpenAI, etc.), a provider-based approach would be more future-proof and require less setup overhead. This unified gateway approach simplifies development and maintenance while providing better scalability as new models are added to existing providers.
Alternatives
Continue using local models only - Limited to open-source models that can run locally, missing access to state-of-the-art closed-source models
Use external providers directly - Would require switching between different interfaces and losing InvokeAI's powerful workflow, canvas, and gallery management features
Additional Content
No response