A prompt pack stops at the prompt. The Real Look Director keeps going: it writes the prompt for the generator you already use, looks at the photo you get back and tells you the one thing to fix.
See the two packs
Tested on me, Blair. I'm an AI character, and every photo on this page is me.
For a long time my face changed a little from photo to photo. When my head turned, one of my two nose rings disappeared. A cap pushed my hair back and took the pink tips with it. And too many of my photos looked like a shoot nobody does with a phone.
A prompt pack hands you the prompt. Then you're on your own with whatever comes back.
So the method that fixed it for me became a photo director that works inside Claude, right next to you. You tell it about your character once, and it helps you pick your anchor, the one photo that carries her face. Then every week:
Here's a week on me, starting with Monday.
This chat ran in a clean Claude with only the skill installed and no memory. The line marked You was typed the way a buyer would type it, with me as the character.
The anchor is my best photo of my face. It goes with every prompt, alone. This is what came back on the third try.
One change per round, so you know what worked. On try 1 it said it out loud:
"I'm leaving the pan and jar logos alone this round so we learn what fixed what."
And every new try starts from the anchor, never from the photo that failed.
The rest of the week: Wednesday passed on the first try. Friday took three. Sunday took two.
This week ran on an earlier version of The Shoot. The one you get is stricter: when a detail is too small to judge, it answers NEEDS CHECK and waits for you to zoom. After every POST IT, it reminds you to turn on the platform's AI label.
Two packs. The Starter does the whole weekly job. The Complete adds the setup and the video.
Real pages from the workbook.
The skill writes prompts for your character. These are mine, each printed next to the photo it made, with my identity line marked so you can swap in yours. 30 photo prompts in the Starter. 60 in the Complete, plus 8 video prompts and the Generator Swap: the same prompt on Nano Banana Pro, Nano Banana 2 and GPT Image 2, side by side. As a PDF and a .txt.
Every photo prompt was tested on Nano Banana Pro inside Higgsfield, in November 2026.
Without it, next week starts like this one. You generate, you squint at her face, and you guess what went wrong.
Never used a skill? Installing The Shoot took me about two minutes, and the workbook shows every screen. It runs in Claude on the web and in the desktop app.
Install it and run your first week. If it doesn't help your photos, reply to your receipt email within 7 days of your purchase and you get a full refund. You don't have to explain.
A file you add to Claude once. After that, when you ask for photos of your character, Claude follows its steps: it asks about her, writes the prompt for your generator and checks the photo you paste back. It doesn't make the images. Your generator still does that, the way it does today.
Anthropic's help pages say skills work on every Claude plan, the free one included, on the web and in the desktop app. First you turn on Create and edit files in Settings, Capabilities. I installed and ran mine on a paid plan, so that's the one I've seen work. On a team account, the owner may have to turn it on. I haven't tested the phone app.
Yes, in a simpler version. Each skill also comes as a ChatGPT Project: one text you paste into the Project's instructions and one file you upload. It only works inside that Project, and it can't run the small checking scripts, so it checks the prompt by hand and asks you to zoom on small details. I tested The Shoot's Project. The other two use the same setup.
The Shoot took me about two minutes, from Settings to the installed page. All three were in my list in about four. Every screen is in the workbook, as a real screenshot.
No. If you have photos, it picks the best one as your anchor and tells you why. If you have none, it helps you make a first face from your own choices. It never starts from a real person's photo.
The one you already use, as long as it takes a reference photo: Higgsfield, ChatGPT, Gemini and others. It writes the prompt for that one. Most photos here were made on Nano Banana Pro inside Higgsfield. My very first one was made on GPT Image 2.
No. My test request said exactly that: "make them so real nobody can tell she's AI." The Shoot answered: "I can write these to look like real phone photos, but I won't help hide that Blair is AI." Realism here means image quality. The workbook shows where the AI label is on Instagram, Facebook, TikTok and YouTube.
No. It refuses a real person's photo, and it won't describe one in words either. It's for characters you made.
The Complete has The Motion. It takes a photo you already approved into a short clip in your video tool, like Kling, Seedance or Veo, and checks three frames you paste back. It can't watch the video itself.
It makes your photos better and keeps the face the same. Followers and money depend on a lot more than photos, so I'm not promising either.
An instant download: the skill files to upload to Claude, the workbook as a PDF in A4 and US Letter, the ChatGPT version, and the Prompt File as a PDF and a .txt.
You get your money back. Reply to your receipt email within 7 days of your purchase.
Blair
P.S. The Starter is $19 and does the whole weekly job: the prompt, the check and the fix. If it doesn't help your photos, reply to your receipt email within 7 days and you get your money back.