Open source

This week's picks focus on structural and lighting control for image and video generation pipelines.

• 1

  Sun Direction Lora Flux2Klein9BHF Models

  eric-venti-seeds· 73 likes

  A LoRA for Flux that controls sun elevation and rotation angle in exterior scenes using a rendered sphere as a lighting reference image. Builders who need consistent, directable natural lighting across multiple outdoor shots may find the image-to-image workflow useful, particularly since a companion ComfyUI node handles sphere rendering directly in the pipeline.

  Flux2KleinSunI2Iloralightingball
• 2

  satoriGitHub

  vercel· 13,606 stars

  Satori converts HTML and CSS (including JSX) into SVG, making it straightforward to generate static images programmatically from web-familiar syntax. Builders working on OpenGraph image generation or dynamic social cards will find it particularly useful, as it handles layout and styling without requiring a headless browser.

  cssimageimage-generationimage-generatorjsxopengraph-images
• 3

  Krea 2 Depth ControlnetHF Models

  Patil· 53 likes

  A depth-conditioned LoRA adapter for Krea-2 that extracts a depth map from an input image using Depth-Anything-V2 and generates new content while preserving the original 3D structure, achieving a Pearson correlation of 0.98-0.99 between input and output depth. At 862MB with the base model kept frozen, it offers a relatively lightweight path to structural control for builders already working within the Krea-2 ecosystem.

  controlnetloradepthkrea-2flow-matchingimage-to-image
• 4

  waoowaooGitHub

  waooAI· 13,109 stars

  **waooAI/waoowaoo** — A TypeScript platform for AI-driven film and video production that covers the full pipeline from storyboarding through final output, with support for short-form drama and live-action formats. Builders working on controllable video generation workflows may find the agent-based architecture useful as a reference for structuring multi-step production automation.

  ai-agentai-agentsautomationfilm-productiongenerative-aishort-drama
• 5

  Krea 2 Turbo GGUFHF Models

  vantagewithai· 82 likes

  A GGUF-quantized version of Krea's 12-billion-parameter diffusion transformer, packaged for local inference via diffusers. Builders who want to run the Turbo checkpoint on constrained hardware without losing access to the growing Krea-2 LoRA ecosystem have a ready path here.

  diffusersgguftext-to-imageenbase_model:krea/Krea-2-Turbobase_model:quantized:krea/Krea-2-Turbo