ComfyUI-Klein9B-AsymFlow wires pixel-space FLUX 9B into local workflows
Third-party ComfyUI extension ComfyUI-Klein9B-AsymFlow brings the 9-billion-parameter FLUX Klein pixel-space model to local workflows, requiring 38 GB VRAM and shipping with an example workflow on GitHub.
A third-party ComfyUI developer has released nodes that integrate the FLUX Klein 9B pixel-space model into local image-generation workflows. The extension, ComfyUI-Klein9B-AsymFlow, was published on GitHub this week by developer CanFromEarth and requires at least 38 GB of VRAM to run.
The nodes wrap the AsymFlow architecture from the LakonLab research repository into custom ComfyUI components. The underlying FLUX Klein 9B model operates in pixel space rather than latent space, a design choice that typically trades higher memory overhead for finer spatial detail. Pixel-space models process images at full resolution throughout the generation pipeline, avoiding the compression artifacts that can emerge when working in latent representations. The tradeoff is steep hardware requirements — the 38 GB VRAM floor puts Klein 9B out of reach for consumer cards below the RTX 4090 or professional A5000 tier.
The extension ships with an example workflow JSON that demonstrates the node configuration for practitioners who want to integrate Klein 9B into existing pipelines. The developer has opened the repository for pull requests and feedback, signaling that the extension remains in active development.
ComfyUI's node-based interface has become the standard for practitioners building complex image-generation workflows that chain multiple models, preprocessors, and post-processing steps. Third-party extensions like ComfyUI-Klein9B-AsymFlow expand the platform's capabilities without requiring upstream changes to the core codebase. Practitioners with workstation or data-center GPUs can now test Klein 9B's pixel-space approach against latent-space alternatives in side-by-side workflows.
The extension is available at github.com/CanFromEarth/ComfyUI-Klein9B-AsymFlow, with original AsymFlow documentation in the LakonLab repository.
