Feature Request: Image Generation API (Text-to-Image / Image-to-Image)

[iazrael]

Feature Request

Description

I would like to request support for image generation capabilities in oMLX, specifically:

1. Text-to-Image (T2I) - Generate images from text prompts
2. Image-to-Image (I2I) - Transform/modify existing images based on text prompts

Motivation

Currently, oMLX supports:

• Text LLMs
• Vision-Language Models (VLM) for image understanding
• OCR models
• Embedding and reranker models

However, there is no support for image generation models. Adding this would make oMLX a more complete local AI inference server, covering the full spectrum of multimodal capabilities.

Proposed API Design

Following OpenAI's DALL-E API specification, the endpoints could be:

Example request:

Potential MLX Models to Support

• Stable Diffusion XL (SDXL) variants
• Stable Diffusion 3 (if available in MLX format)
• FLUX models
• Other diffusion models from mlx-community

Benefits

1. Unified API: Single server for text, vision, and image generation
2. Apple Silicon Optimization: Leverage MLX's performance on M-series chips
3. Offline Capability: Local image generation without external API calls
4. Privacy: Keep generated images entirely on-device

Additional Context

I noticed the project already has excellent VLM support via . Image generation could potentially leverage similar patterns from the MLX ecosystem (e.g., for diffusion models if available).

Thank you for considering this feature request!