Mim- Transmit meaning.

Mim — An experimental social network for sharing “meaning,” not images Mim is a new type of semantic communication social network that shares the “meaning” of an image rather than sending the image itself. AI analyzes images and converts them into semantic representations such as captions, prompts, and tags, then reconstructs images on the receiving device. Mim aims to create a new way of sharing information with a strong focus on privacy and data minimization.

■ Image Understanding

• SigLIP2: Extracts tags and estimates features to generate captions and prompts. In Pro Mode, it is also used to calculate semantic fidelity.
• Qwen3.5-VL-0.8B: Directly generates captions, prompts, and tags from images, enabling more natural descriptions.
• Vision Framework: Supports lightweight image understanding through object recognition. When no model is installed, only lightweight image understanding via Vision Framework is available, and advanced semantic analysis is not supported.

■ Image Reconstruction

• Stable Diffusion 1.5: Supports img2img and generates high-quality reconstructed images in the timeline.
• Stable Diffusion 1.5 (LCM): Optimized for fast, low-latency reconstruction.
• Image Playground: On supported devices, it can be used for lightweight image generation when no model is installed. When no model is installed, available generation features are limited, and reconstruction methods and quality may vary depending on the environment.

■ Post Modes

• Light Mode: Privacy-focused (meaning-centered)
• Strong Mode: Easier to reconstruct (includes more guiding information) Reconstruction quality varies depending on the model, mode, and source image.

■ Social Features

• Follow feature
• Comment feature

■ Pro Mode You can view metrics suited for research and analysis:

• Semantic fidelity
• LPIPS-based perceptual distance evaluation
• Prompt generation time
• Image generation time
• Memory usage for each step

■ System Requirements

• Requires iOS 18.4 or later
• iPhone 15 Pro or later is recommended
  • A supported device is required to use Image Playground
• Stable Diffusion behavior has not been verified on some devices

■ Notes Generated images are experimental and may not exactly match the original images. Results may vary depending on the model and generation environment. Model providers do not guarantee output quality.

■ GitHub (Detailed Documentation) https://github.com/lube8163-lab/mim-ios/tree/main