MediaPipe On-Device Text-to-Image Generation Solution Now Available for Android Developers
- Google has released a new on-device text-to-image generation solution called MediaPipe Image Generator, available for Android developers.
- The Image Generator task can be used in three ways: text-to-image generation based on text prompts using diffusion models, customizing a foundation model via MediaPipe Diffusion LoRA fine-tuning on Vertex AI, or using pre-trained models provided by Google.
- The Image Generator can create new images by passing in a text prompt, the number of iterations for generating, and a seed value.
- The plugin system allows the diffusion model to accept a condition image along with a text prompt as inputs, enabling the creation of new images using specific structures extracted from the condition image.
- The Image Generator can be customized using LoRA weights to teach the model about new concepts, such as objects, people, or styles.
- The generated LoRA weights can be deployed on-device or through Vertex AI's one-click deployment for optimized server inference.
- Google plans to expand the capabilities of on-device image generation and encourages developers to share their creations using the hashtag #MediaPipeImageGen and tagging @GoogleDevs.