A learning environment for creative intelligence.

A learning environment for creative intelligence.

Expert creative data, captured from real production workflows, teaching AI how humans actually create masterpieces.

How the data is made.

How the data is made.

How the data is made.

Three things make expert creative data possible. Real production workflows where decisions carry weight. A vetted network of creators whose work stands up to professional scrutiny. A platform that captures the process at its source, as it happens.

1

"iFood Benefícios — AI-animated brand campaign produced in Studio."

Real production workflows

Every data point comes from actual production work — content being made to be published, delivered, or broadcast. Projects range from commercial brand work to independent creator productions, all of them real, all of them consequential.

2

.

Vetted creator network

Clonex works with a curated community of creators whose output is verified against production-grade standards. Only creators meeting quality benchmarks contribute to the dataset. Craft is the filter.

3

.

Structured capture at the source

Studio is the platform where creators produce their work. As they produce, the system captures decisions, alternatives, rejections, tool choices, and outcomes — structured, timestamped, machine-readable. No retrofitting. No reconstruction. The data comes out of the act of creation itself.

What you get.

What you get.

What you get.

Four types of training data, compatible with standard post-training pipelines. All captured from real production workflows. All verifiable against the original output. Available with multi-creator diversity the same brief solved by multiple experts.

Creative Planning Traces

(SFT + CoT)

Full plan graphs from raw client brief to final asset, intent, constraints, decomposition, tool strategy, evaluation. Rejected options tagged at every decision, with explicit rationale for both chosen and rejected paths. Structured for supervised fine-tuning and chain-of-thought reasoning.

Multimodal Preference Pairs (RLHF)

Chosen-versus-rejected pairs across text-to-image, image-to-video, and voice generation, with expert justification for every choice and every rejection. Captured from real production decisions, not synthetic curation. Compatible with standard reward model training formats.

Creative Planning Traces

(SFT + CoT)

Full plan graphs from raw client brief to final asset, intent, constraints, decomposition, tool strategy, evaluation. Rejected options tagged at every decision, with explicit rationale for both chosen and rejected paths. Structured for supervised fine-tuning and chain-of-thought reasoning.

Multimodal Preference Pairs (RLHF)

Chosen-versus-rejected pairs across text-to-image, image-to-video, and voice generation, with expert justification for every choice and every rejection. Captured from real production decisions, not synthetic curation. Compatible with standard reward model training formats.

Creative Execution Trajectories

Complete tool-use traces across multi-model pipelines, every prompt, iteration, parameter, and output. Includes model routing decisions and creator-agent interactions: how expert creators correct, refine, and collaborate with AI assistants inside the workflow. The data creative agents orchestrating multiple tools need to learn.

Outcome-Linked Creative Data

Creative decisions connected to post-publication performance metrics at 24h, 7d, and 30d. The missing layer in creative training data today — process grounded in measurable real-world results.

Creative Execution Trajectories

Complete tool-use traces across multi-model pipelines, every prompt, iteration, parameter, and output. Includes model routing decisions and creator-agent interactions: how expert creators correct, refine, and collaborate with AI assistants inside the workflow. The data creative agents orchestrating multiple tools need to learn.

Outcome-Linked Creative Data

Creative decisions connected to post-publication performance metrics at 24h, 7d, and 30d. The missing layer in creative training data today — process grounded in measurable real-world results.

Talk to us about a pilot.

We work with frontier labs and research teams to design custom data programs for the creative domain. First conversations are scoping, not pitch.

Talk to us about a pilot.

We work with frontier labs and research teams to design custom data programs for the creative domain. First conversations are scoping, not pitch.