Open source

This week's picks center on building and controlling image and video generation pipelines with modular, local, or efficient tooling.

• 1

  Sulphur 2 BaseHF Models

  SulphurAI· 1,331 likes

  Sulphur-2-base is a text-to-video and image-to-video generation model built on the LTX 2.3 architecture, with native support for the full range of LTX 2.3 formats and a bundled GGUF prompt enhancer for refining inputs before generation. Builders already working in the LTX ecosystem may find the drop-in distillation LoRA and diffusers compatibility useful for integrating it into existing pipelines with minimal rework.

  diffusersgguftext-to-videoendpoints_compatibleconversational
• 2

  ComfyUIGitHub

  Comfy-Org· 114,368 stars

  ComfyUI is a node-based GUI and backend for running diffusion models, letting you wire together image and video generation pipelines visually as a graph. Builders who want fine-grained control over inference workflows without writing a full pipeline from scratch will find its modular node system a practical starting point.

  aicomfycomfyuipythonpytorchstable-diffusion
• 3

  SANA WM_bidirectionalHF Models

  Efficient-Large-Model· 99 likes

  SANA-WM (Bidirectional) is a 2.6B-parameter image-to-video diffusion transformer that generates 720p, minute-scale videos from a conditioning image and text, with explicit 6-DoF camera trajectory control per frame. Builders working on camera-controlled scene synthesis may find the hybrid linear attention design worth evaluating, as it combines frame-wise Gated DeltaNet with periodic softmax attention blocks to handle the long context of minute-scale video without the memory overhead of full attention throughout.

  diffuserssafetensorstext-to-videoimage-to-videocamera-controlworld-model
• 4

  LocalAIGitHub

  mudler· 46,456 stars

  LocalAI is a self-hosted, GPU-optional engine that runs LLMs, image generation, video, audio, and vision models through a local API. Builders who want to run image or video generation pipelines without cloud dependencies or GPU requirements may find it a practical starting point.

  agentsaiapiaudio-generationdecentralizeddistributed
• 5

  Lens TurboHF Models

  microsoft· 77 likes

  Microsoft's Lens-Turbo is a text-to-image diffusion model built on the Lens architecture, which focuses on rethinking training efficiency for foundational generative models. Builders looking to experiment with a research-backed, diffusers-compatible pipeline may find it worth evaluating as a more compute-efficient baseline for text-to-image work.

  diffuserssafetensorstext-to-imageendiffusers:LensPipeline