Cindy Zhu.
Free guides
Design & visuals Multi-tool

Korean profile photos, made at your desk with Soul ID

Hey, it's Cindy ๐ŸŒฑ This is the whole setup from the reel: connecting Higgsfield to Claude, training a character on your own face, and the exact prompt that shoots the four Seoul studio looks.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
before you start

What this costs and what you need

I would rather you know this up front than find out three steps in.

  • A Claude account. Custom connectors work on Free, Pro, Max, Team and Enterprise. On the free plan you get one custom connector, so this will be it.
  • A Higgsfield account. This is the paid part. Training a character and generating images both spend credits, and Higgsfield runs free trial credits on and off, so check what is on your account before you plan a big shoot.
  • 20 photos of yourself. Not 20 versions of the same selfie. Real variety, and I have put the exact shot list further down.
  • About 25 minutes. Roughly five for the connector, ten for gathering photos and training, and the rest is generating and picking.
You do not need Claude Code for this. The whole thing runs in the normal Claude chat at claude.ai. Higgsfield's own page notes that if you are in Claude Code or Codex you are better off with their CLI, but for this workflow the chat connector is the simpler path.

step 1

Connect Higgsfield to Claude

This is the plumbing. You do it once and it stays connected.

  1. Open Claude and go to Customize. At claude.ai, open the settings area and find Customize in the sidebar. It sits below the main Settings group, alongside Skills and Plugins.
  2. Click Connectors. This is the list of everything Claude can reach outside the chat.
  3. Click the "+", then Add custom connector. Higgsfield is not in the built-in gallery, so it goes in as a custom one.
  4. Paste the server URL. It is https://mcp.higgsfield.ai/mcp. Nothing else needs filling in, and you can ignore the advanced OAuth settings.
  5. Click Add, then authenticate. Claude will send you through a Higgsfield sign-in. Once it comes back, the connector card shows in your list and Claude can generate images and video for you from inside a normal chat.
If the connector will not add. Check you pasted the full URL including https://, and check you are not already at your free-plan limit of one custom connector. If Claude adds the connector but never calls it, say so directly in the chat: "use the Higgsfield connector for this". It sometimes needs to be told the first time.

step 2, the one that matters

Train a Soul ID character on your own face

This is the difference between a photo of someone who vaguely resembles you and a photo people think you actually took. Higgsfield calls it a Soul ID character: you feed it a set of your own photos, it learns your face, and then every image you generate afterwards keeps that face locked.

  1. Go to Higgsfield and open Soul. From higgsfield.ai, open Soul, then Train new character.
  2. Upload 20 or more photos. Higgsfield's own requirement is 20 minimum and up to 80 supported. More variety beats more photos, so 20 good ones beat 60 near-identical ones.
  3. Name it. Give it your own name so you can find it in the Character tab later.
  4. Wait for training. It takes a few minutes. You can close the tab.

The photo set that actually works. Higgsfield's guidance plus what I learned making mine:

Do include Different angles and expressions, clear even lighting, recent photos taken in the last few months, and at least one full-height shot so it understands your proportions.
Do not include Sunglasses, hats, scarves over your face, heavy shadow, cropped faces, group shots, or ten photos from the same afternoon in the same pose.
A quick way to gather them. Open your camera roll, search your own name in the People album, and pull the twenty that look least alike. Fighting the instinct to only pick photos you like is the trick here: the model needs range, not a highlight reel.

step 3

The prompt that shoots the four looks

Back in Claude, in a chat with the Higgsfield connector on. Type or paste this, and Claude will run it through Higgsfield with your character attached.

๐Ÿ“ธ the Seoul studio profile set
Using my Soul ID character, create 4 Korean self-profile studio photos in the style of Seoul profile photo studios. Photorealistic, 4:5 portrait, one image per shot.

Keep my face, features, skin tone and proportions exactly the same as my character. Do not slim, reshape, lighten or smooth my face.

Look: clean seamless studio backdrop, soft diffused front light from a large beauty dish, gentle fill from below, no harsh shadows, a subtle healthy glow on the skin while keeping real skin texture, crisp focus on the eyes, 85mm lens at eye level.

Styling: natural no-makeup makeup with soft coral lips and straight natural brows, neat hair with a few soft face-framing strands.

Shot 1: chest-up, facing the camera, calm closed-mouth smile, soft grey backdrop, black blazer over a white tee.
Shot 2: slight head tilt, gentle smile, warm cream backdrop, crisp white shirt with the collar open.
Shot 3: shoulders angled 30 degrees, looking back at the camera, pale blue backdrop, fine-knit cardigan.
Shot 4: black and white close-up, chin slightly down, looking straight into the lens.

No text, no watermark, no props, no extra people.

Why the prompt is built that way. The three lines doing the heavy lifting are the ones people delete. "Keep my face, features, skin tone and proportions exactly the same" stops the model quietly prettifying you, which is what makes an AI portrait stop looking like its subject. "Keeping real skin texture" stops the plastic finish. And naming a lens and a light source gives it a real photographic reference to copy instead of a vibe.


more looks

Three more sets from the same character

Once the character exists, you never do the setup again. These run off the same character and give you a full year of profile photos in one sitting.

๐Ÿ’ผ the corporate set
Using my Soul ID character, create 3 professional corporate profile photos. Photorealistic, 4:5 portrait, one image per shot, shot on an 85mm lens at eye level.

Keep my face, features, skin tone, skin texture and proportions exactly the same as my character. Do not slim, reshape, lighten or smooth my face.

Look: a modern office interior softly blurred behind me, large window light from camera left, warm neutral tones, a confident relaxed posture with shoulders back.

Shot 1: chest-up in a well-fitted charcoal blazer over a white shirt, small closed-mouth smile, looking straight at the camera.
Shot 2: three-quarter body, arms loosely crossed, standing beside a window, a warmer natural smile.
Shot 3: seated at a desk with a laptop softly out of focus in the foreground, mid-conversation expression, looking slightly off-camera.

No text, no watermark, no logos, no extra people.
๐Ÿ‚ the seasonal outdoor set
Using my Soul ID character, create 3 natural outdoor profile photos that look like a friend took them on a good camera. Photorealistic, 4:5 portrait, one image per shot, 50mm lens, shallow depth of field.

Keep my face, features, skin tone, skin texture and proportions exactly the same as my character. Do not slim, reshape, lighten or smooth my face.

Look: late afternoon golden light, soft natural shadows on the face, a gently blurred background, relaxed candid body language rather than posed.

Shot 1: walking toward the camera on a tree-lined street, mid-stride, laughing slightly, in a cream knit and a long coat.
Shot 2: sitting on stone steps, chin resting on one hand, looking off to the side.
Shot 3: leaning against a plain painted wall, arms relaxed, calm direct eye contact with the camera.

No text, no watermark, no extra people.
๐ŸŽž๏ธ the black and white editorial set
Using my Soul ID character, create 3 black and white editorial portraits in the style of a magazine contributor page. Photorealistic, 4:5 portrait, one image per shot, 85mm lens, fine natural film grain.

Keep my face, features, skin tone, skin texture and proportions exactly the same as my character. Do not slim, reshape, lighten or smooth my face.

Look: a single soft key light from camera right with deep controlled shadow on the opposite side, a plain mid-grey backdrop, strong contrast but detail held in the shadows, crisp focus on the eyes.

Shot 1: tight head and shoulders, chin slightly down, direct eye contact, neutral expression.
Shot 2: profile view looking away from the camera, light catching the edge of the face.
Shot 3: chest-up, one hand touching the collar of a plain black top, a small honest smile.

No text, no watermark, no props, no extra people.

when it goes wrong

The four things people hit

  1. ๐Ÿ˜ It does not look like you. Almost always the training set. Twenty photos from the same event teach it one angle in one light. Retrain with real variety and one full-height shot, and the likeness jumps.
  2. โœจ It looks like you but airbrushed. The model smooths by default. Add "keep visible skin texture, pores and natural under-eye shadow, no beauty retouching" to the look paragraph and run it again.
  3. ๐Ÿ™ˆ Claude generated something without your character. It forgot to attach the character. Say "use my Soul ID character called [name] for this" and rerun. You can also generate directly in Higgsfield's Soul interface and pick the character in the Character tab.
  4. ๐ŸŽฝ The clothing is nearly right but not right. Describe fabric and fit rather than a garment name. "A fine-gauge cream knit with a crew neck, fitted through the shoulders" gets you much closer than "a nice jumper".

the honest bit

Where these belong, and where they do not

These are for LinkedIn, Instagram, a speaker bio, a team page, a dating profile if you are upfront that it is a stylised photo. They are genuinely good for that.

Not for official ID. The US State Department's position on passport photos is that "any use of computer software, phone apps, filters, or AI tools to alter your photo is prohibited", and a flagged photo delays the whole application. Other countries take the same line. Use a real photo for anything official, every time. The reporting on it is here.

The other honest note: these are generated images of you, not photographs of you. I would not post one as a "before and after", and I would not use one where someone is going to meet you in the next hour expecting that exact lighting. Used as what they are, a styled profile photo, nobody has a problem with it.


after the shoot

Getting the most out of one session

Generate more than you need Run each prompt two or three times. The variation between runs is bigger than the variation inside one run, and the keeper is usually in run two.
Keep a prompt file Paste the prompts that worked into a note with the character name. Six months from now you will want the same look with a different outfit, and you will not remember the phrasing.
the waitlist is now open

AI System Club is launching soon

Build AI systems with me through recorded lessons you can follow at your own pace. Join the waitlist for first access and the founding price.

btw this list gets first dibs
the moment doors open :)