Identity-preserving image edit workflow. Feed in one reference photo and a scene prompt ending in "keep her face, identity, and outfit exactly the same" — the model edits the background/pose/outfit while keeping the person recognizable. Uses TextEncodeQwenImageEditPlus, which conditions on the reference image directly (not just a text description of it), so the result actually looks like the same person, not a lookalike.
Nodes: Load Diffusion Model (qwen_image_edit_2509_fp8_e4m3fn), Load CLIP, Load VAE, Load Image, TextEncodeQwenImageEditPlus (positive/negative), ModelSamplingAuraFlow, KSampler (40 steps, CFG 4), VAE Decode, Save Image — flat graph, no subgraph.
Pairs with: Qwen-Image (Workflow 1 https://civitai.com/api/download/attachments/681445) — generate the original reference face there first, then use this one to reuse it forever.