Skip to content

Claude planning + QA playbook

Use Claude for the brief and the truth check.

Claude does not currently render ordinary image outputs. Use its image understanding to plan the edit and audit the result, then use Gemini, ChatGPT, or Grok for the actual staging render.

Practical guideReviewed August 20, 2026
Important: this controlled fictional demonstration shows the room brief and evidence pair Claude can help inspect. It is not a claim that Claude directly produces the staged image.

Treat this as one connected operating system: capture an honest source with the source-photography guide, build the property locks and client direction in the prompt library, run the edit in the model playbook you choose, compare both files with the quality-control gate, then package the accepted draft with the disclosure and record system. FSBO sellers can continue into the owner publishing workflow. Agents can turn the verified evidence pair into a source-led content bundle and a citable SEO/GEO page.

The two-model workflow.

  1. 01

    Give Claude the original

    Ask it to identify permanent features, ambiguities, likely room use, and visual risks without guessing.

  2. 02

    Build the edit brief

    Have Claude turn its inspection into a short furniture plan and a tool-ready image-editing prompt.

  3. 03

    Render in an image tool

    Take the brief to Gemini, ChatGPT, Grok, or another image editor that can produce the actual edit.

  4. 04

    Return both images to Claude

    Upload original and staged versions together and ask for evidence-backed comparison, not a general opinion.

  5. 05

    Correct failed checks

    Send exact failed items back to the rendering tool. Keep all already-correct areas locked.

  6. 06

    Run the human gate

    Claude can miss visual changes. A person still verifies property facts, platform rules, and disclosure.

Customize for the client

Brief the listing before you brief Claude.

The client should be able to choose the presentation direction without giving the model permission to redesign the home. Record these six fields, get the direction approved, then insert the exact answers into the prompt.

01

Room use

Living room, primary bedroom, office, dining area, or another specific function.

02

Listing position

Entry, mid-market, premium, rental, or FSBO context. This controls restraint and finish, not property facts.

03

Style direction

Name one useful direction such as warm contemporary or soft modern-traditional.

04

Furniture density

Minimal, balanced, or fuller. Set the walking path and the short essential-furniture list.

05

Must-keep facts

Camera, crop, geometry, openings, fixtures, materials, view, light direction, and visible condition.

06

Do-not-infer rule

Use property and design context. Do not target or assume protected classes, family status, disability, religion, race, or another protected trait.

Treat this as one connected operating system: capture an honest source with the source-photography guide, build the property locks and client direction in the prompt library, run the edit in the model playbook you choose, compare both files with the quality-control gate, then package the accepted draft with the disclosure and record system. FSBO sellers can continue into the owner publishing workflow. Agents can turn the verified evidence pair into a source-led content bundle and a citable SEO/GEO page.

Give Claude evidence, not vibes.

Upload the original and staged files together. Ask for a check table that cites visible evidence. “Looks accurate” isn’t a useful answer.

Original and staged primary-bedroom comparison: Before — original empty room

Before — original empty room

Original and staged primary-bedroom comparison: After — virtually staged draft

After — virtually staged draft

What this example teaches: Claude’s job is to inspect the evidence pair and identify changed geometry, permanent features, scale, or uncertainty. A human still owns the final decision. Controlled fictional example created for this guide. It is not a real listing, endorsement, or comparison of named models.

Claude briefing prompt.

Photo-to-edit-brief prompt

Inspect the uploaded room photo and help me prepare a virtual-staging brief. Do not claim to render the image.

Return:
1. probable room type and current visual constraints
2. permanent features the image editor must preserve
3. visible condition or ambiguity that staging must not conceal
4. a minimal furniture plan with scale and circulation notes
5. light direction, likely shadow behavior, and reflection risks
6. a concise image-editing prompt for [TARGET IMAGE TOOL]
7. a 12-point side-by-side verification checklist specific to this photo

If a dimension, material, view, or feature is unclear, label it uncertain rather than guessing.

Claude verification prompt.

Original-vs-staged QA prompt

Compare the original room photo and the virtually staged version I uploaded.

Create a table with: check, original evidence, staged evidence, pass/fail/uncertain, and required correction.

Check camera and crop; perspective; walls; windows; doors; trim; ceiling; flooring; built-ins; fixtures; exterior view; material colors; visible condition; furniture scale; circulation; contact shadows; reflections; and disclosure readiness.

Do not judge style preference until property accuracy is complete. If you cannot verify a detail from the images, mark it uncertain.

Research sources

Check the current rules, capabilities, and source basis.

  1. 01Anthropic: Can Claude produce images? ↗

    Anthropic’s current first-party explanation of Claude’s image-output limitation and visual-analysis capability.

  2. 02Anthropic: Uploading files to Claude ↗

    Current first-party upload guidance; supported inputs and limits can change.

  3. 03NIST Generative AI Profile ↗

    Supports layered human review, tracking, and documentation.

  4. 04FTC: Advertising and Marketing Basics ↗

    The external listing image still has to be truthful and non-deceptive.

Product access, limits, local listing rules, platform policies, and laws can change. Re-check the linked authority before important or public work.

Keep building

Put AI to work in your real estate business.

AI Acceleration: Real Estate is a private, self-directed curriculum for practical AI work across content, communication, lead follow-up, CRM, client research, operations, and relationship management. Build it. Review it. Keep what works.

Explore the Skool community ↗

Follow Ryan + AI Acceleration

No ranking, revenue, closing, or time-saving promise. The classes teach the work and the review systems around it.