Integration WordPress AI Client SDK

[Aunshon]

Summary: WP AI Client SDK Limitations for Our Use Case

We evaluated the WordPress AI Client SDK for integration into our WordPress plugin but ultimately decided not to proceed due to several functional limitations and incompatibilities with our requirements.

Official reference:
https://make.wordpress.org/ai/2025/11/21/introducing-the-wordpress-ai-client-sdk/
https://github.com/WordPress/wp-ai-client

Our Requirements

1. Video generation from images

   • Input: one or more images
   • Output: video
   • Video behavior and instructions provided via prompt

2. Image generation from multiple images

   • Input: 2–4 images (e.g., one model image + product images)
   • Output: a generated image
   • Transformation/composition instructions provided via prompt

Findings & Limitations

1. No Video Generation Support

   • The SDK currently does not support video generation.
   • There is no API surface for passing images + prompt to generate video output.

2. Image Generation Is Prompt-Only

   • Image generation works only for text → image use cases.
   • Supplying image inputs using:
     • withFile()
     • withMessageParts()
   • results in “not supported” errors.
   • Multi-image → image generation is not supported, even though the API structure suggests it might be.

3. Model & Preference Issues

   • Changing the model or model preferences frequently results in:
     • 404 errors, or
     • "not supported" responses
   • This makes it difficult to reliably target models that may support advanced multimodal generation.

Conclusion

Due to:

• Lack of video generation support
• Inability to generate images from multiple input images
• Unstable behavior when switching models or preferences

The WP AI Client SDK does not currently meet our plugin’s requirements. For now, we’ve decided not to adopt the SDK and will revisit it once multimodal (image + image, image → video) generation is officially supported.