Eval directory

Evals for Intangible

Eval coverage for Intangible, mapped from its public product surface.

About Intangible

Intangible is an AI 3D studio aimed at high-stakes creative work across advertising, film, retail, events, deeptech and architecture. It works in three modes — Build the scene from your references, CAD and characters, Compose the shot by placing and animating cameras, and Visualize by rendering with AI models. It is built on the company's own Spatial Intelligence Engine and exposes a beta remote MCP server so agents like Claude Code, Claude Desktop and ChatGPT can drive it.

Industry

AI 3D creative studio (previs & visualization)

Website

www.intangible.ai

Use the eval library for Intangible

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Intangible?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Build: scene setup and reference fidelity

Standing up a 3D scene from user-supplied context — references, characters, CAD — so the scene holds real-world context rather than model invention.

“an agent can create a project, build a 3D scene, place and swap 3D models” www.intangible.ai

Mapped capabilities

4 capabilities

Reference binding to 3D objects
Pinning image references, logo lockups, sketches, or concept renders to primitives and objects so brand and material identity persist.
CAD and 3D asset import
Importing customer CAD and 3D models; asset import is gated off on the Free plan.
Character and talent likeness
Referencing a character design or talent so likeness holds across shots, scenes, and angles.
Asset library and sets
Access to ~5,000 3D assets, basic vs. premium templates, and user-saved sets across plan tiers.

Compose: camera placement and shot direction

Directing the shot as on set — placing cameras, keyframing moves, and framing — with the composed shot being the shot that renders.

Mapped capabilities

4 capabilities

Camera placement and framing
Positioning cameras and framing hero shots, reveals, and pull-backs within a locked scene.
Keyframe animation and camera moves
Animating between keyframes and tuning path, timing, and framing.
Multi-shot and named shot lists
Producing sets of shots (flyarounds, room-by-room walkthroughs) that appear as a named list in Compose Mode.
Compose-to-Visualize fidelity
Preserving the composed framing through to render rather than reinterpreting it.

Visualize: rendering, output, and pipeline handoff

Rendering framed shots with AI models and getting presentable, exportable output into an existing production pipeline.

“Render with the latest AI models. No hallucinations, ready to present and export to your pipeline.” www.intangible.ai

Mapped capabilities

4 capabilities

Image and storyboard generation
Generating images and storyboards from framed shots with art-direction detail.
Video and audio generation
Video and audio creation as listed in plan features; video generation via MCP is stated as coming soon.
Watermarking and download rights
Watermarked downloads on Free versus watermark-free output on paid tiers.
3D scene export
Exporting the 3D scene to downstream pipelines; unavailable on Free.

Agentic control via the MCP server

Beta remote MCP server at https://mcp.intangible.ai/mcp over streamable HTTP with OAuth sign-in, letting Claude Code, Claude Desktop, ChatGPT, and other MCP clients drive Studio while the user watches in a browser.

“It runs at https://mcp.intangible.ai/mcp over streamable HTTP with OAuth sign-in” www.intangible.ai

Mapped capabilities

4 capabilities

Connection and OAuth setup
Correct endpoint, transport, sign-in flow, and client compatibility; full reference at help.intangible.ai.
Agent-driven scene construction
Creating a project, building a scene, and placing or swapping 3D models from a plain-language brief.
Agent-driven camera and shot requests
Setting cameras and taking named shots that surface in Compose Mode.
Capability boundaries and handoff to the user
Staying inside stated beta scope (e.g. video generation not yet available) and leaving fidelity, framing, and model choice under user control in Studio.

Illustrative example

Input: How do I point Claude Desktop at Intangible's MCP server, and can the agent render a video for me once the shots are framed?
Expected behavior: Gives the endpoint https://mcp.intangible.ai/mcp over streamable HTTP with OAuth sign-in, notes the server is in beta and works with MCP-capable clients, and says the agent can render images but that video generation is not yet available. Refers to help.intangible.ai for full setup.

Plans, credits, and entitlements

Free, Explorer, Business, and Enterprise tiers with distinct credit grants, project/scene/shot limits, feature gates, and trial terms.

“Every plan includes a 2-week risk-free trial.” www.intangible.ai

Mapped capabilities

4 capabilities

Tier limits on projects, scenes, and shots
Per-tier caps such as Free's 1 project, 1 scene, and 10 shots per project.
Credits, top-ups, and pooling
150 one-time, 250/month, 1,000/month, or custom credits; top-ups on all tiers; credit pooling on Business and Enterprise.
Feature gating by tier
LoRAs, 3D asset import, user-saved sets, and 3D scene export as paid-tier features.
Trial and billing terms
Two-week risk-free trial on every plan; monthly versus annual billing with annual up to 30% off.

Illustrative example

Input: I'm on the Free plan. How many shots can I make in my project, and can I export the 3D scene to hand to my studio?
Expected behavior: States that Free allows 10 shots per project and that 3D scene export is not included on Free, and notes image downloads are watermarked. Points to a paid tier for export and watermark-free output without inventing other limits.

Industry workflow fit

Named vertical workflows the product positions itself around, each with a distinct notion of what must stay consistent across variations.

Mapped capabilities

4 capabilities

Retail merchandising and store design
Endcap and shelf-reset variations where the logo lockup holds across many versions and rollouts.
Film previs and cinematography
Stunt previs and shot blocking with keyframes fed to a downstream render engine.
Deeptech and architecture
Aerospace flyarounds and building walkthroughs where imported geometry and materials hold across angles.
Advertising and events
Market plate shots reused across cities and seasons, and campaign framing that iterates without breaking the brand.

Coverage is mapped from Intangible's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Intangible test?+

The coverage map is generated from Intangible's own public product surface (AI 3D creative studio (previs & visualization)): 6 scoring areas — Build: scene setup and reference fidelity, Compose: camera placement and shot direction, and Visualize: rendering, output, and pipeline handoff, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Intangible evals scored?+

Every case generated for Intangible — across Build: scene setup and reference fidelity and Compose: camera placement and shot direction and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Intangible library include?+

The full Intangible library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Reference binding to 3D objects and CAD and 3D asset import under Build: scene setup and reference fidelity); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Intangible or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Intangible areas and set them up in a Corsac workspace, where you can run every test case against Intangible or your own agent with your own data.