FLUX.2 Klein 4B and LanPaint for Swap Clothes

Flux.2 Klein 4B is a fast, unified generate‑and‑edit model that can swap clothes on a subject using simple text instructions and optional reference garment images.

What it is

• A 4B‑parameter FLUX.2 variant designed for interactive image editing and multi‑reference conditioning (e.g., “put this dress from image B onto the person in image A”).

• Very suited to ComfyUI clothes‑swap workflows where you keep the person’s face, pose, and background but replace only the outfit.

Key features for clothes swap

• Text‑guided editing: you describe the target outfit in natural language instead of doing manual inpainting prompts like classic diffusion.

• Multi‑image references: load a target portrait and one or more clothing reference images; Klein uses them to match style, color, and structure.

• Unified generate + edit: same model handles base generation and local edits, so identity, lighting, and background stay consistent.

• Fast, lightweight: distilled for low latency and runs on consumer GPUs, ideal for iterative tweaking of outfits.​

• Plays nicely with masking tools (SAM2/3, Kontext, manual masks) so you limit the edit area just to the clothing region for cleaner anatomy.

When it’s the right choice

• You want high‑quality, high‑res outfit swaps where the face, pose, and background must stay nearly unchanged.

• You need speed and lots of iterations (trying multiple outfits, brands, or colors) on local hardware.​

• You’re swapping outfits using one or several reference photos (catalog shots, lookbook images) instead of only text.

• You already use ComfyUI and want a workflow that avoids heavy manual inpainting, using text+mask+references instead.