Back
~8 min read 23 Mar 2026

CLAUDE + OPENART WORLDS. The Virtual Production Pipeline

AI Summary

Text prompts alone cannot handle complex visual storytelling for commercial campaigns. A spatial workflow using Google Nano Banana, OpenArt Worlds, and OpenArt Characters gives designers precise control over environments, characters, camera angles, and poses. A token-based prompting system replaces freeform text with structured variables for camera framing, body language, and facial expressions, enabling agency-grade output in 10 to 15 seconds per shot.

Key takeaways

1

Build 3D navigable environments by generating a 2D scene in Nano Banana, then uploading it to OpenArt Worlds. Keep the space clean with no characters or branding to avoid distortions.

2

Create consistent custom characters by generating 4 views (full body, bust, profile, back) in Nano Banana, then training OpenArt Characters on that set. Tag the character with @ in any prompt or World.

3

Use a token matrix instead of descriptive paragraphs. Swap structured tokens into a base prompt to control framing, spatial rotation, kinetics, micro-expressions, and camera angle with precision.

Original post

The Framework to build agency-grade design pipelines.

INTRO*

*This article is sponsored by@openart_ai. Thanks to the OpenArt team for supporting this post and my work for you.

For 10+ years, I’ve been working for big & small brands. From Warner Music & Kimberly Clark, to local coffee shops. Do you know what sets them apart?

Attention to details and control.

Now, we’re AI’ing everything - but what I hate about AI is the ILLUSION of control. 1000’s tools, daily overpromises... and 90% of them are useless for any commercial work.

Text prompts can’t handle complex visual storytelling (aka “campaigns”). You need more manageable tools.

LET’S BUILD A PIPELINE TO FIX THIS

Video thumbnail

In 2026, we need to leverage spatial workflows - for controlled visual production. Interiors, characters & cameras - must follow your ideas.

I will show you how to build strong workflows. Plus, you’re getting my Tokens Matrix (Notion Doc):a free system for total control over your AI generations.


The stack I use:

  • Claude AI: For brainstorming & prompting
  • Google Nano Banana: For scenes and objects.
  • OpenArt Worlds: For building 3D environments from those scenes.
  • OpenArt Characters: For integrating custom avatars into your worlds.

Save this article. Come back when you need these tips & assets.

Article image

TABLE OF CONTENTS:

  1. The Stage: setting up your virtual studio
  2. The Actor:building a pro-grade character
  3. Token Matrix: become a pro-prompter with these tokens
  4. The Busines:turn your raw generations into commercial assets
Article image

1. THE STAGE: Preparing your location

Let’s prepare a base for the 3D world you will walk through.

I created 1000’s of visuals for companies, so let me share some knowledge:

MY WORKFLOW:

REFERENCE RESEARCH:Find a reference image of your desired environment 🔎 WHERE TO SEARCH: Pinterest, Savee.it, shotdeck.com, frameset.app, archdaily.com

🔎 WORDS TO SEARCH:

- Infinity cove; - Colored cyclorama studio; - Soundstage set design; - Limbo background lighting; - LED volume stage.

Find 30+ editorial requests in my Notion Doc: Link - at the end of the article👇

LLM PROMPTING: Deconstruct the image using LLMs to build a proper prompt: ▶️ Step 1: Upload your image to Claude ▶️ Step 2: Paste this prompt & run the model:

This section contains content that couldn't be loaded. View on X

▶️ Step 3: You’ll get back the detailed prompt to generate your scene. Copy it & go to: https://amirmushich.link/OA-NB2
GENERATION: Paste your prompt & generate the environment using Nano Banana 2.
WORLD CREATION: Upload your generated image into OpenArt Worlds & hit Create World: https://amirmushich.link/OAWorlds The generation will take up to 5 minutes.

Watch the tutorial:

Video thumbnail
The 2D concrete space generated in Nano Banana being uploaded into OpenArt Worlds, instantly becoming a navigable 3D space.

TIP:

Don’t add characters, text or branding elements here. Keep the space clean to avoid distortions.

🎁BONUS:

I prepared the pack of pro studio interiors + exact prompts. Download them for free at the end of this article (+prompts) 👇

Article image

2. THE ACTOR (creating characters)

For commercial campaigns, you need custom avatars that work consistently across any scene.

Images with people increase business metrics: ⤴️ 3x better ER (engagement rate) in Social Media ⤴️ 60% better eye-tracking ⤴️ 30%+ better click-through rate (CTR) in ads

So, let’s train the AI system with a custom Character Kit. AI can’t guess what your model’s shoes look like from behind. Let’s define it.

Article image

MY WORKFLOW:

Article image

1️⃣ REFERENCE HUNT:

Savee.itor Pinterest: Filter by high-end editorial boards Models.com: For clean casting shots Vogue Runway (www.vogue.com/fashion-shows): For high-end styling
🔎 WORDS TO SEARCH: - high-fashion street casting polaroid; - athletic sportswear editorial model full body; - minimalist activewear turnaround; - cinematic character portrait raw lighting.

2️⃣ LLM PROMPTING:

Let Claude do the heavy lifting & describe the model properly. ▶️ Step 1: Upload your chosen image to Claude ▶️ Step 2: Paste this prompt & run the model:

This section contains content that couldn't be loaded. View on X

▶️ Step 3: You’ll get back the detailed prompt for your image generation. Copy it & go to: https://amirmushich.link/OA-NB2

3️⃣ GENERATION:

Take the prompt and paste it into Google Nano Banana. To build a robust model, generate 4 variations. Replace the[FRAMING & ANGLE]variable in your prompt with these 4 views: 1. Full Body shot, frontal view; 2. Bust Up Close-up portrait; 3. Strict Profile view; 4. From Behind (back view).

4️⃣ CHARACTER CREATION:

Let’s bring the talent to life inside the OpenArt ecosystem: - Go to OpenArt Characters (https://amirmushich.link/OA-Model) - Upload your set of 4 images. - The AI will synthesize them into a character ready to be added into your Environments. - Once created, give your character a name and add into your prompts & Worlds just by tagging it with “@”
Video thumbnail

These 4 camera tokens and 20+ others - is in my Notion Doc, too. Steal them at the end of the article 👇

Article image

3. TOKENS MATRIX

Your control panel to craft commercial shots on the agency level.

Writing paragraphs of descriptive text? Hell no! We’ll use the token-based system.

Plug thesetokensinto yourBase Prompt 01 below to control the:

  • camera,
  • kinetic energy,
  • micro-expressions of your character.

Tutorial first, assets below 👇

Video thumbnail

BASE PROMPT 01 (FOR PRO SCENE GENERATIONS):

How to use: ▶️ Copy this prompt ▶️ Change [variables] with the tokens from my Matrix Doc (blocks 02-06) ▶️ Send the prompt to OpenArt Worlds prompt window (https://amirmushich.link/OAWorlds)

This section contains content that couldn't be loaded. View on X


TOKENS:

Below is a preview of the Tokens Matrix Yeah, the full list of tokens is ALSO in the Notion Doc (I don't want you to miss it, so I have to bump this everywhere. No selling, I swear.🙃)

Article image

SECTION 1: Camera & Space

Control your framing and alignment with total precision.

TOKEN EXAMPLES:

[FRAMING]:Extreme Wide Shot: The widest view of the space. Full Body Shot: Full-length shot. Focus on the silhouette, outfit, and footwear.
[POSITION]Centered Composition: Perfect symmetry, with the subject strictly centered in the frame. Left-Aligned Composition: The subject is shifted to the lef..

60+ tokens - in my free Notion Doc 👇

Article image
Replace the [FRAMING] and [POSITION] variables of the BASE PROMPT 01 with this section’s tokens in my Notion Doc.

SECTION 2: Spatial Rotation

Direct the model's exact angle relative to the lens.

TOKEN EXAMPLES:

[ROTATION]: Facing Camera: Direct frontal view. The face and torso are facing the camera. 3/4 View Left: Classic editorial pose with the torso turned 45 degrees to the left. Strict Profile Right: A sharp profile to the right.

More tokens - in my free Notion Doc 👇

Article image
Replace the [ROTATION] variable of the BASE PROMPT 01 with one of these tokens.

SECTION 3: Kinetics & Body Language

No more default standing poses. Use editorial, dynamic posture tokens.

TOKEN EXAMPLES:

[ROTATION]: Power Stance: Feet shoulder-width apart, hands on hips. A confident, grounded pose. Overhead Extension: Arms extended vertically upward, torso stretched. Starting Block: A deep, athletic squat with weight distributed across the hands and feet.

More tokens - in my free Notion Doc 👇

Article image
Replace the [POSE] variable of the BASE PROMPT 01 with one of these tokens.

SECTION 4: Micro-Expressions

High-end aesthetics demand controlled facial tension.

TOKEN EXAMPLES:

[EMOTION]: Fierce Focus: A narrowed gaze that pierces the lens. A firm, tense jawline. Subtle Arrogance: Chin slightly raised, cold, condescending gaze. Candid Profile: Gaze directed off-camera. The effect of a candid, authentic moment.

More tokens - in my free Notion Doc 👇

Article image
Replace the [EMOTION] variable of the BASE PROMPT 01 with one of these tokens.

SECTION 5: Camera Angle

Control the virtual camera angles easily.

TOKEN EXAMPLES:

[ANGLE]: Bird’s-Eye View: Direct top-down view. Architectural shot of the space. Worm’s-Eye View: Extremely low angle from floor level. The character appears huge. Ground Level: The camera rests on the floor; sharp focus on feet and the floor texture.

More tokens - in my free Notion Doc 👇

Article image
Replace the [ANGLE] variable of the BASE PROMPT 01 with one of these tokens.

4. BUSINESS SHIFT: Why this works?

The final asset. Raw AI renders → final designs:

Article image
The final, polished campaign banners set featuring my Character (Lana) in the generated OpenArt World, overlaid with the custom typography. Want these designs? Grab my Figma template (yep, it’s also in that Notion Doc) - no attribution needed.

👆I’ve included this Figma layout template in the Notion Doc below.

  • download it,
  • drop your OpenArt generations directly into the frames,
  • and ship your beautiful campaign today.

I don’t care about AI models’ benchmarks.

I want solutions for my design work. Just like our clients.

2026 will be the ‘UI Battlefield’

(because AI models are already great enough 🤷‍♂️)

Spatial Calibration

is one of these specific UI-controllers that change the way we work:

  • select the environment,
  • place the character,
  • move the camera,
  • find the angle,
  • take the shot.

All - in ~10-15 seconds.

This level of productivity is impossible via text prompting.

Practical utility & UI/UX dictate who wins now.

Done is better than perfect. With OpenArt’s Worlds you get things done 5-7 times faster.


TAKE ACTION TODAY:

  1. Grab my Notion Doc below.
  2. Use my free assets in your pipeline.
  3. Stand out in the market

GRAB YOUR FREE ASSETS HERE:

https://amirmushich.link/T-Matrix


I SHARE WORKFLOWS I USE MYSELF.

Save this article, come back when you need these tips.

Article image

Follow @AmirMushich for more free articles & workflows

Turn on notifications on my profile and never miss the AI sauce that works🔔

AmirMušić

@AmirMushich

Creative Director. Ex-Warner Music, PepsiCo, Spotify designer. Fusing 10+ yr of brand & ad design with AI. Collabs: amir@amirmushich.com

Follow

Join builders who get the best curated content in their inbox weekly.