headshot 📸

Turn a selfie into studio headshots — on your own GPU, for free.

People pay $40+ to apps that do this in the cloud with your face. headshot does it locally: it generates a professional portrait and swaps your real face onto it (via ReActor), so the result actually looks like you — not a stranger who kinda resembles you.

headshot me.jpg                      # corporate look (default)
headshot me.jpg --style linkedin     # pick a vibe
headshot me.jpg --style outdoor --count 4
headshot --list-styles               # see them all

Output lands in ./headshots/.

---

Styles

Style	Look
corporate	Dark suit, grey studio backdrop, soft key light
linkedin	Smart-casual blazer, blurred office, friendly + confident
outdoor	Natural daylight, blurred park, golden hour, warm
creative	Moody editorial, dramatic rim light, magazine cover
bw	Classic black & white, Rembrandt lighting, timeless

Requirements

A running ComfyUI with:

• the ReActor face-swap node, and
• the inswapper_128.onnx model (+ a face-restore model like GFPGANv1.4.pth).

No ComfyUI yet? → ai-workstation. ReActor node → ComfyUI-ReActor. headshot itself only needs Python 3 (standard library).

Quick start

git clone https://github.com/CastilloworksAi/headshot.git
cd headshot
./setup.sh                  # link `headshot`, check ComfyUI + ReActor
headshot --check            # confirm everything's wired
headshot me.jpg --style linkedin

How it works

your selfie ─────────────────────────────┐ (source face)
                                          ▼
prompt → checkpoint → KSampler → VAE → [ReActor face swap] → SaveImage → ./headshots/
         (generates a clean studio portrait "base")   ▲
                                          puts your real face on the base

Because the base is freshly generated, you get clean studio lighting and wardrobe; because ReActor swaps your actual face on top, it stays you.

Tips for good results

• Use a clear, front-facing selfie with even lighting — that's the face it copies.
• --count 4 gives you variations (it nudges the seed each time); keep the best.
• Different --style values reuse the same face, so you can build a whole set.

Usage

Flag	Default	Meaning
--style	corporate	one of the styles above
--count	1	how many variations to make
--model	juggernautXL_v9.safetensors	base checkpoint
--steps	28	sampler steps
--seed	random	reproduce a result
--size	832x1216	portrait WxH
--out	./headshots	output folder
--dry-run	—	print the graph, generate nothing
--check	—	verify ComfyUI + ReActor
--list-styles	—	print styles

Development

bash tests/test.sh

No GPU/ComfyUI/model needed: it checks syntax, that the generate→swap graph is valid and wired correctly (selfie → ReActor source, base → ReActor input), that all styles exist, and that --dry-run injects the right style + selfie. CI also runs shellcheck.

The face swap itself runs on your GPU through ComfyUI, so it can't run on a GPU-less CI box. headshot --check then a real selfie confirms it live.

A note on consent

Only swap faces you have the right to use — your own, or someone who agreed. Don't make images of real people without their consent.

License

MIT — see LICENSE.