How to Crop an AI Headshot for a Real LinkedIn Photo
A 12-minute image skill to crop an AI headshot into a real LinkedIn photo. Head-and-shoulders. Even light. No cinematic bokeh. Midjourney, Flux, or ChatGPT image, then one human pass against a real grid.
Anyone sitting on a generated portrait can crop an AI headshot for LinkedIn in 12 minutes: head-and-shoulders, even light, no cinematic bokeh. Do it in Midjourney, Flux, or a ChatGPT image edit before a movie still trains recruiters to skip you.
An AI headshot that belongs on LinkedIn is a passport with better hair, not a poster. If the frame still has a rim light and city blur, you prompted a trailer still.
Do not reopen the face. That is keeping a character consistent across prompts. This job is the frame: crop, light, and a wall that could exist in an office. If subject and atmosphere are still one soup string, structure the image prompt in layers and keep the person in the subject block. After the photo is honest, write the LinkedIn post that is not a humblebrag and the one-paragraph bio that is not a resume as separate jobs. No slogan on the portrait.
Crop an AI headshot to look like a real photo
1. Name the sins: width, bokeh, movie light
Look at the still. Write what is wrong on paper. Not "make it more professional."
Typical sins: a 16:9 or full-body crop with empty city to the side; cinematic bokeh, fog, or flare; Rembrandt or teal-orange light with a hard rim; a costume coat.
Banned: cinematic, movie still, Hollywood, hero lighting, dreamy, bokeh as a compliment, editorial, magazine cover.
2. Instruct head-and-shoulders, even light
Stay on the same face. Do not reroll the person.
Instruct Midjourney, Flux, or ChatGPT image: "Same person. Head-and-shoulders crop. Eyes in the upper third. Both shoulders in frame. Even, soft daylight, no rim, no colored gel, no fog. Neutral office or plain wall. Shallow depth is fine only if the wall is a wall, not a city. No cinematic bokeh."
If the tool can region-edit, mask the background and the empty sides first. A full regenerate is how a cousin arrives.
3. Crop 1:1, then check the neck
LinkedIn's circle will eat corners. Crop square, or 4:5 if you must, then preview in a circle.
Eyes sit in the upper third. Chin not clipped. Crown not clipped. Both shoulders visible. A crop that starts at the lips is a selfie. A crop that starts at the belt is a lookbook.
If you only have a wide still, crop in any editor first, then send the tight crop back as the reference. Do not ask the model to "zoom in" with no crop. Zoom-in prompts invent a new face.
4. Kill glow, grit, and costume
Instruct: "Remove film grain, bloom, and color grade. Skin is skin. Shirt is a shirt I would wear to a standup. No jewelry the original face did not have. No extra hair volume. No smoke."
If the coat is a costume, that is a wardrobe problem, not a crop problem. Change the garment in a region edit. Do not prompt "more LinkedIn" as a vibe. LinkedIn is not a filter.
5. One human pass against a real grid
Open five real coworker photos. Put yours in the same row.
Check three traps:
- Bokeh that still reads as a music video
- Light that still has a colored rim
- A crop the circle will decapitate
Fix crop in the editor. Send only lighting or background notes back to the model: "Replace only the background and the light. Do not change the face or the crop I attached."
One pass. Export. If the file still looks like a poster, you have a movie. Stop.
Reusable skill card
Trigger: You have an AI portrait that is too wide or too cinematic and you are about to upload it to LinkedIn as-is.
Input: The generated still, a head-and-shoulders crop instruction, even daylight, a plain wall or bland office, plus a ban on cinematic bokeh and costume wardrobe.
Output: One still. Head-and-shoulders. Even light. Face unchanged. No city bokeh. Previewed in a circle.
Stop: After one pass against a real coworker grid. No face reroll. No slogan on the photo. No 16:9 banner as a headshot.
Takeaways
- Head-and-shoulders, even light, no cinematic bokeh. Then export.
- Face lock is identity. This skill is crop and light. Do not mix the jobs.
- Crop square and preview in the circle. Zoom-in prompts invent a cousin.
- A costume at the collar is still a costume. Region-edit the shirt.
Frequently Asked Questions
Can I keep the cinematic bokeh if the face is sharp?
No. Bokeh that reads as a music video is the tell. LinkedIn wants a wall or a bland office, not a city at f/1.2.
Should I regenerate the face to look more professional?
No. Face lock is a different skill. This job is crop and light. A new seed is a cousin in a cousin shirt.
What aspect ratio should I export?
Square 1:1, then preview in LinkedIn's circle. 4:5 is acceptable if the circle still shows both shoulders. 16:9 is a banner. Do not upload a banner as a face.
Midjourney, Flux, or ChatGPT image for the fix?
Any of the three that will take the still as a reference. Flux is strict with a locked face plus a region. Midjourney wants remix from this frame. ChatGPT image is fine if you stay in the same chat. Do not start a new prompt from memory.
Can I wear the generated leather trench if I crop tight?
No. A costume is still a costume at the collar. Region-edit the garment into a shirt you would wear to a standup.