Summary

Add the ability to configure custom API endpoints for image generation and a "Test Connection" feature to verify the setup.

Problem to solve

Currently, Presenton supports a limited set of image generators (GPT, Nano Banana). This limits users who want to use more advanced, cost-effective, or specialized models like Bytedance-Seedream available through third-party providers like Meganova.

Additionally, I recently experienced a bug where Nano Banana generated completely black images, causing a loss of paid credits. The current system doesn't allow users to verify if the integration is working correctly before committing to a full generation, which can lead to wasted resources.

Proposed solution

I suggest adding a "Custom Provider" option in the settings where users can:

1. Define a Custom Endpoint: Allow inputting a base URL (e.g., https://api.meganova.ai/v1/images/generations).
2. Custom Model Name: A field to specify the model string (e.g., Bytedance/seedream-5-0-lite-260128).
3. Authentication: A field for the API Key.
4. Test Button: A "Test Connection/Generate Test Image" button. This would trigger a simple request to ensure the API key and endpoint are valid and that the output is rendering correctly, preventing credit loss from bugs.

Example API Implementation (compatible with Meganova/OpenAI-like structures):

// Request example for third-party providers
const data = {
    "model": "Bytedance/seedream-5-0-lite-260128",
    "prompt": "A modern minimalist smartwatch on a clean white background...",
    "size": "1920x1920",
    "provider_options": {
        "image": referenceImages // Support for reference images would be a plus
    }
};

Alternatives considered

Using the current built-in providers, but they lack the variety of styles and the competitive pricing/performance found in newer models like Seedream. Manual generation outside of Presenton and importing files is a workaround, but it breaks the integrated workflow of the app.

Use case

A user wants to create a presentation with a very specific artistic style only available via Bytedance models. They configure their Meganova API key and endpoint in Presenton, click "Test" to ensure the connection is live and images aren't coming out black, and then proceed to generate all presentation assets seamlessly within the app.

Impact

• Who would benefit: Power users, designers, and anyone looking for better cost/performance ratios in AI image generation.
• Severity: Medium/High (due to the risk of losing credits/money when a provider fails or is misconfigured).
• Workflow impact: Significantly improves reliability and gives the user freedom to choose their preferred AI provider.

Additional information

As mentioned, I previously reported a bug where images from Nano Banana were generated as solid black. This resulted in lost credits. A "Test" feature is crucial to ensure that if a provider has an issue, the user finds out through a controlled test rather than losing credits during their actual work.