Xb1n0ry Streamlines Reference Workflows With ComfyUI-KleinRefGrid

ComfyUI-KleinRefGrid replaces multiple reference image nodes with a single custom component that stitches up to four pictures into one unified grid. xb1n0ry created this utility to streamline the image conditioning process for the Flux Klein 9B model.
Local creators often waste valuable workspace managing separate encoding and injection steps for reference materials. This package bundles the gallery, stitching, and data routing into one interface, keeping project files cleaner and easier to share.
Unified reference grid generation
- Converts up to four reference pictures into a fixed 2000 × 2000 pixel canvas at runtime.
- Provides an internal image gallery with drag-and-drop uploads and thumbnail previews.
- Automatically fills empty slots with black tiles to maintain consistent dimensions.
- Exposes a numerical strength slider to adjust the influence of input materials.
- Embeds the entire gallery layout directly into saved workflow files.
Practitioners building character portfolios can maintain consistent outfits and facial structures across multiple generations without training custom adapters. Small studios producing asset variations benefit from the streamlined pipeline, which replaces tangled node clusters with one configurable block.
Behind the conditioning logic
The release consolidates complex routing arrangements into a direct connection placed between text encoders and sampling guides. Early testing focused strictly on shorter generation steps, and the creator notes that smaller model variants might process combined grids differently. Users should verify compatibility when switching between checkpoint versions, as data handling can shift across architecture sizes.
"Providing reference latents this way is essentially equivalent to using a mini-LoRA without requiring any training,"
said the developer in a community post. Integrating multi-image references into a stable generation routine becomes significantly more straightforward with this dedicated workflow block.