FLUX 3 Klein in ComfyUI: setup and image editing

This tutorial builds a unified ComfyUI workflow for FLUX 3 Klein text-to-image generation, reference editing and bounding-box placement. Setup covers the generation checkpoint, Qwen3-VL 4B text encoder, Qwen 3.5 9B prompt upsampler, FLUX 2 VAE and Vantage-NodesV2. The presenter instructs viewers to update ComfyUI for FLUX 3 support, install missing nodes and restart.

The model switch selects matching conditioning and sampler presets: 16 steps with CFG 1 for Distilled, or 25 steps with CFG 3.5 for Base. The presenter distinguishes that shorter Base preset from the published 50-step reference setting. Reference order matters, and the workflow allows ten images; an active box-editor background occupies the first slot.

The main tests use Distilled INT8 ConvRot at 4K with Prompt Enhancer enabled. Examples include portraits, a letterpress poster and clothing transfer. The presenter reports readable poster wording but missed small caps, plus changes to sleeves, facial details and the background during clothing editing.

Bounding boxes provide spatial guidance rather than strict masks, so surrounding details may change. A final thumbnail example uses one megapixel and disables Prompt Enhancer because it can strip structured object data from the layout prompt node. This explains a practical tradeoff between prompt expansion and placement control.