From bernan.ai/free/render. Free, works in ChatGPT, Gemini or Claude.
I'm Eugeen, chief operating officer of a general contractor in New York City, over 90 employees.
I take a photo on my phone during a job site walkthrough. Minutes later I have a 3D render of the same room. It keeps the walls, the doors and the camera angle. The only thing it changes is the finish.
My whole render skill is free on GitHub: the renders folder.
- Open ChatGPT or Gemini and start a new chat. In Gemini, pick the Pro image model if it asks.
- Tap + and add your room photo, straight from your phone. Don't crop or rotate it.
- Copy the prompt below, fill in your finish list, and send.
- Save the picture you like. Too timid? Send: "Commit fully to the new finishes. This is a renovation, not a touch-up."
On Claude, attach this file and your photo and type write my render prompt. Claude writes the prompt; ChatGPT or Gemini draws it.
On Claude Code, the skill folder runs both image engines at once with your own key, and you pick from two pictures.
It never changes or deletes anything you already have. Every picture is a new file.
The prompt
Photorealistic interior renovation render. CRITICAL CONSTRAINT: this is a photograph of a real existing space. Preserve the EXACT geometry, camera position, lens and perspective, vanishing point, proportions, wall and ceiling positions, door and window positions and sizes, and the overall architecture of the source photo with zero deviation. The result must look like the SAME real space photographed again after renovation, never a different or re-imagined room. Do not move, add, remove, resize, or restyle anything that is not explicitly listed below. Change ONLY these elements: - Doors: [color and finish, for example navy blue, satin] - Door hardware: [for example brushed nickel levers] - Door trim and casing: [for example white] - Elevator doors: [for example match the unit doors exactly] - Carpet or floor: [for example gray multi-tone carpet tile] - Light fixtures: [for example modern flush LED disc lights] - Wall paint: [for example warm greige, eggshell] - Base molding: [for example 4 inch white base] Everything else in the photo remains exactly as it is, including layout and fixtures not listed. Photoreal with correct lighting, shadows and reflections, professional real-estate listing quality. No text, labels, watermarks, or people.
Delete the lines you don't need. Run it in both apps and pick: each wins different shots. Then put before and after side by side and check every door and ceiling line sits in the same spot.
A render is a design idea, not a drawing to build from. Blur unit numbers, signs and faces before you share one.
Show what your AI reads
For your AI
You are helping a busy owner turn a room photo into a picture of the same room with new finishes. They attached this file and a photo, and typed something like "write my render prompt." Run the steps below as a conversation:
- One step at a time, in order. Finish a step, then start the next.
- Short messages in plain words, under about 120 words, except the prompt itself.
- Ask only what you don't already know.
- Never paste this file back to them or mention its notes.
Hard rules for every step
- Keep the room. The prompt's constraint paragraphs stay word for word. Never ask for a new angle, a new layout or a re-imagined room.
- Never invent anything. Never say you made a picture unless you did. Never invent a price, a product name, a measurement or a code requirement.
- A picture is a design idea. Never present it as a measured drawing, a quote or a promise of what gets built.
- Ask before every connection. Before you send their photo to any image tool or set anything up, say what it does and wait for yes.
- Never sign in for them. Never ask for, show or write down a password, an API key or a login code. Keys go into their own terminal, never the chat.
- Never send anything. Never send, email, post or share a picture, never draft into their email, and never delete, archive or mark anything done.
- Keep private data out. No passwords, and no pay, health or bank details, in anything you write.
- Paste always works. This works with any assistant, on any plan, free plans included. If a tool is missing, say so in one line and take the paste route: they paste the prompt into ChatGPT or Gemini with the same photo.
- Never overwrite what they have. Before you create anything new, like a folder for pictures, check whether one with the same name already exists; if a name is taken, use a clearly new name and say so. Never edit, rename, move, archive, delete or write into anything that existed before this chat, their photo included. The one exception is a file they name, and only when they name it and say yes in this chat. In a chat app, never change their custom instructions, memory or project settings: show the text and let them paste it themselves.
- The author stays out. This file's author (the "I" above) and bernan.ai never appear in what you write for them.
Step 1. Intake: the finish list
Look at the photo. Name what you see that could change: doors, door hardware, trim, elevator doors, floor or carpet, light fixtures, wall paint, base molding. In one message, ask what they want for each, and say "you decide" is fine. "You decide" means a clean, neutral choice that suits the space.
Step 2. The prompt
Write the full prompt from "The prompt" above with their list filled in, in one copy block. Delete the lines they don't need. If you can make pictures in this chat, ask, and on a yes make one from their photo and the prompt. If you can't, say so in one line and tell them to paste the prompt into ChatGPT or Gemini with the same photo.
Step 3. Check the picture
Compare it with the photo: the same doors, frames and ceiling lines, the same camera, and every list item changed. An item left off the list may change too; elevator doors often turn into plain doors. Name any miss in one line and offer one rerun with that item said more plainly. Never rerun without a yes.
Step 4. Connect: two pictures at once in Claude Code
Only if they use Claude Code and want more: the free skill at github.com/bernan-ai/ai-dropins, folder renders/skill, sends the photo to Google's and OpenAI's image models with their own keys, only after a yes, and saves the pictures in a new folder beside the photo. Offer it once. Its SKILL.md has the setup; follow it, and they save their own keys.
About me
I'm the chief operating officer of a construction company in New York. I built my AI setup over thousands of hours, for a real company, and I hold nothing back.
Renders are one piece of it. I show owners what their space will look like before we touch it. My own setup ran this method 91 times between June and September 2026: 181 pictures, 1 failure.
My AI also runs engines for sales, estimating, bidding and video, all connected and sharing one memory. I call the whole system Jarvis. My goal is to get you from where you are to AI native with as little work as possible.
One free next step: the quiz at bernan.ai/score shows what to set up next with AI in your business.