Use the workflow attached! It was made with this in mind, to reduce broken outputs.

One photo in, sheet out.

Drop in one photo of someone and you get back a 3x2 grid of their face on grey: neutral, soft smile, big laugh, angry, surprised, and a wink with tongue out. This is an edit LoRA, so you give it a photo, not a text-to-image prompt.

Already run Qwen Image 2.1 in ComfyUI? Then the only new file is this LoRA: expressions_qwen_edit.safetensors → ComfyUI/models/loras/. Missing a base file? Open the workflow and ComfyUI offers to download it.

How to use it

1. Download Expressions_Sheet.json from the files on this page and drag it into ComfyUI.
2. Drop your photo in the load image box.
3. Press Run. Don't like it? Press Run again for a new try.

The prompt and LoRA strength are preset in the purple box. Change them if you like; the original prompt is saved in the top note. No extra node packs needed.

Photo tip: Close selfie where the face fills the frame works best. A far-away photo means the face is tiny and it has to guess. The expressions are acted out, not her real ones. The third gallery board shows a good vs bad photo.

Settings (preset): the standard Qwen Image 2.1 template files, not Turbo. LoRA strength 1.0, 25 steps, CFG 1, euler / simple. Output is 1344x768 (wide). Any photo shape works, it gets padded onto grey, nothing cut off.

Prompt (built into the workflow, here if you use your own setup)

turn this photo into a facial expression sheet: a 3x2 grid of head-and-shoulders close-ups of her on a plain grey background, showing a neutral face, a soft smile, a big open-mouth laugh, an angry frown, a surprised open mouth, and a wink with her tongue out

See also, same set:

• Three Faces Sheet – Qwen Image 2.1 Edit LoRA
• Angles Sheet – Qwen Image 2.1 Edit LoRA
• Clothing Sheet – Qwen Image 2.1 Edit LoRA
• Character Sheet Edit LoRA (the all-in-one original)

Follows the Qwen Research License (non-commercial).