Open source

This week's picks lean toward local, self-hosted, and pipeline-integrated generation across image, video, and audio modalities.

• 1

  LTX 2.5HF Models

  Lightricks· 1,665 likes

  Lightricks' LTX-2.5 is a diffusion-based model that generates video from image, text, audio, or combined inputs, supporting a range of pipelines including image-to-video, video-to-video, and audio-to-video synthesis. Builders working on multi-modal video workflows may find it worth evaluating given its native support for audio conditioning, which is less common among open video generation models and enables tighter audio-visual synchronization without separate post-processing steps.

  diffusion-single-fileimage-to-videotext-to-videovideo-to-videoimage-text-to-videoaudio-to-video
• 2

  Anima 2.9BHF Models

  Gazingstars123· 317 likes

  Anima-2.9B is a 2.9B-parameter text-to-image diffusion model fine-tuned and layer-expanded from the Anima base, targeting anime and illustration generation. Builders using ComfyUI or Forge-Neo can drop it in without custom nodes, and LoRA training support via a dedicated standalone trainer makes it straightforward to adapt for specific styles or characters.

  diffusion-single-fileanimacomfyuitext-to-imageenbase_model:circlestone-labs/Anima
• 3

  MochiDiffusionGitHub

  MochiDiffusion· 7,926 stars

  Mochi Diffusion is a native Mac app for running Stable Diffusion and FLUX.2 Klein locally, built in Swift and optimized for Apple Silicon via Core ML and the Neural Engine. Builders targeting on-device image generation on macOS can use it as a reference for how to wire up diffusion pipelines to Apple's hardware acceleration stack.

  aneappleapple-siliconcoremlmacosneural-engine
• 4

  H3_Character_Sheet_GeneratorHF Models

  PoopMan333· 83 likes

  A ComfyUI workflow built on MiniMax H3 that takes up to nine reference images and produces a multi-panel character turnaround sheet by extracting frames from a single video generation pass. Because all views come from one pass rather than separate generations, facial features, colors, and costume details stay consistent across the sheet without manual correction.

  minimax-h3comfyuiworkflowcharacter-sheetturnaroundcharacter-consistency
• 5

  sdnextGitHub

  vladmandic· 7,316 stars

  SD.Next is a self-hosted WebUI that consolidates image and video generation, captioning, and media processing into a single interface built on Stable Diffusion, Diffusers, and Transformers. Builders who want a unified local environment for experimenting across multiple generative pipelines without stitching together separate tools may find it worth a look.

  ai-artcaptiondiffusersgenerative-artpythonpytorch