Open source

This week's picks center on lowering the cost and complexity of video and image generation workflows.

• 1

  KromossHF Models

  lodestones· 264 likes

  Kromoss is a full fine-tune of MOSS-Image-Base at 1024px, using a Qwen3.5-9B text encoder and FLUX.2 small VAE decoder for text-to-image generation via the diffusers MOSSImagePipeline. Builders looking to cut inference costs will find the bundled DMD2 turbo LoRA useful, which brings generation down to 8 steps at CFG 1 while reportedly achieving the best PartiPrompts score among the included options.

  diffuserssafetensorsmoss-imagetext-to-imageenbase_model:OpenMOSS-Team/MOSS-Image-Base
• 2

  FastVideoGitHub

  hao-ai-lab· 4,737 stars

  FastVideo is a Python framework for accelerating video diffusion model inference and post-training, with support for techniques like distillation. Builders looking to reduce generation latency or fine-tune video models more efficiently have a focused toolkit here rather than assembling individual components from scratch.

  diffusersdiffusion-modelsdistillationinferencepost-trainingvideo-generation
• 3

  youtube-automation-agentGitHub

  darkzOGx· 4,660 stars

  A self-hosted Node.js agent that handles a YouTube channel end to end, from topic research and script writing to visual generation, video assembly, SEO optimization, and publishing with a human approval step. Builders exploring multi-step agentic pipelines for video content will find it a concrete reference for chaining LLMs (Gemini, OpenAI, DeepSeek, and others) with media generation and platform publishing in a single workflow.

  ai-agentsautomationcontent-automationcontent-creationcontent-strategygoogle-gemini
• 4

  ComfyUI-LTXVideoGitHub

  Lightricks· 4,214 stars

  Custom ComfyUI nodes for running Lightricks' LTX-2 video generation model, covering both text-to-image and image-to-video workflows. Builders looking to integrate a DiT-based video generation pipeline into their ComfyUI setups without building custom nodes from scratch will find a ready-made starting point here.

  comfyuidiffusion-modelsditimage-to-videoimage-to-video-generationtext-to-image
• 5

  Moss Image EditHF Models

  chinoll· 94 likes

  Moss-Image-Edit is a LoRA for the MOSS-Image family of models that enables text-guided editing of colors, outfits, lighting, seasons, and visual style on existing images. Builders already using MOSS-Image-Base or MOSS-Image-SFT can drop it into a ComfyUI workflow with minimal setup, and bilingual prompt support (English and Chinese) broadens its practical reach.

  safetensorsloraimage-editingmoss-imagecomfyuiimage-to-image