🤯 Consistency HACK: From 1 portrait to 5 perfect scenes using “Rule Prompts”

🔥

What’s the biggest headache with AI? Getting it to keep the character’s face consistent across different images without having to train a custom model.

Yesterday I tested a different workflow that worked incredibly well: 𝗥𝘂𝗹𝗲-𝗯𝗮𝘀𝗲𝗱 𝗜𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲.

Instead of writing 5 huge prompts describing the character over and over again, I did this:

1️⃣ 𝗧𝗵𝗲 𝗩𝗶𝘀𝘂𝗮𝗹 𝗔𝗻𝗰𝗵𝗼𝗿 (𝗧𝗵𝗲 𝗥𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲)

I uploaded a clear image of the character. This is the “absolute truth” the AI cannot modify.

2️⃣ 𝗧𝗵𝗲 𝗥𝘂𝗹𝗲 “𝗠𝗲𝘁𝗮-𝗣𝗿𝗼𝗺𝗽𝘁” (𝗧𝗵𝗲 𝘀𝗲𝗰𝗿𝗲𝘁)

Here’s the key. I didn’t just say “put him in a gym.”

I pasted the 𝗹𝗶𝘁𝗲𝗿𝗮𝗹 𝘁𝗲𝘅𝘁 𝗼𝗳 𝘁𝗵𝗲 𝗰𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 𝗿𝘂𝗹𝗲𝘀 𝘄𝗲 𝘄𝗲𝗿𝗲 𝗴𝗶𝘃𝗲𝗻.

By giving explicit instructions like:

“Same exact person. Not ‘similar.’ COOL scenarios: neon, rooftops, boxing…”

you create 𝘀𝘁𝗿𝗼𝗻𝗴 𝘀𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗰𝗼𝗻𝘀𝘁𝗿𝗮𝗶𝗻𝘁𝘀.

The AI understands that priority #1 is obeying the identity rule from the reference image, and priority #2 is being creative with the environments listed in the text.

3️⃣ 𝗧𝗼𝗽-𝗗𝗼𝘄𝗻 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻

The AI generated all 5 scenes automatically based on those general rules.

Once the model “understands” the scene and the character, it becomes super easy to ask for variations (like converting them to vertical format) without losing consistency.

💡 𝗖𝗼𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻:

Sometimes, to achieve consistency, you don’t need longer prompts — you need better rules of the game.

Let the AI fill in the creative details (environment, lighting) while you tightly control identity with a clear instruction text.

👇 𝗜’𝗺 𝗹𝗲𝗮𝘃𝗶𝗻𝗴 𝘁𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 👇

7 comments