FLUX 3 Klein in ComfyUI: setup and 19 image workflows

Learn to generate, edit and combine images with 19 ComfyUI workflows, including a Z-Image Turbo realism pass and edits confined to a box.

Player not loading? Watch on YouTube

This tutorial covers 19 FLUX 3 Klein workflows in ComfyUI for image generation, editing and combining reference pictures. Setup requires ComfyUI 0.41 or newer and Pixaroma 1.4.223 or newer. The speaker explains Hugging Face access approval and where to place the model, text encoder and VAE files, recommending the model without "base" in its name for the faster version.

Generation examples cover detailed prompts, product-label text and layouts controlled with boxes. A prompt upsampler expands short ideas, while a separate workflow uses Z-Image Turbo at twice the image size. The speaker finds that a light realism pass improves some portraits, but stronger settings can change the face. For an RTX 30 card CUDA error during prompt expansion, the suggested workaround is to restart ComfyUI and disable "faster writing."

Editing examples include background replacement, boxed inpainting, restyling, photo restoration and outpainting. Reference-image workflows demonstrate outfit changes, logos on products and group compositions. The speaker reports persistent problems with hands, small faces and handwritten lettering. Whole-image edits can shift details; boxed edits and stitched outpainting preserve the original outside the changed area. Restored faces are reconstructions rather than reliable likenesses. Reported runtimes describe the speaker's card, whose model is not specified.