Troubleshooting FLUX.2 Klein in ComfyUI Without Guesswork

By Dominic Marrocco, 42 UK Research. Technical claims below are bounded to the linked primary sources.

New Zealand troubleshooting edition · evidence review: 5 August 2026

Diagnose the actual failure before changing settings

The former 2 GB claim on this URL was not supported by the official sources and has been removed. The FLUX.2 klein base 4B card says that model fits in about 13 GB of VRAM. The 9B model card describes a separate release and says it fits in about 29 GB. These are publisher specifications, not universal minima.

Check in this order

1. Model identity: confirm the repository, revision and licence. A 4B workflow and a 9B workflow are not interchangeable merely because the family name matches.
2. Supported environment: compare the host with the current ComfyUI requirements. Resolve unsupported Python, PyTorch, driver or accelerator combinations before tuning memory.
3. Workflow completeness: verify that required files and nodes come from the official model instructions or a workflow with declared provenance.
4. Memory pressure: if the error is genuinely an allocation failure, follow the official ComfyUI troubleshooting guide.

Documented memory-pressure options

The official guide includes --lowvram, lower resolution or batch size, --preview-method none, and reserved-VRAM controls. Apply one controlled change at a time. Note whether the workflow completed and capture peak memory using a named tool. Success at one resolution and precision does not establish a universal minimum.

Escalate with a useful record

Include the error, the ComfyUI version source, model revision, platform details, workflow and changed flags. Store proven configurations in the installation matrix, certified assets in the workflow library, and measured memory results in the low-VRAM benchmark hub.