Visual control
CyaniModel
Control subject identity, spatial structure, image edits and visual delivery constraints.
Explore CyaniModel
LlamaGen Research · Technical definitions
CyaniModel Technical Library
A source-backed guide to controllable visual generation and the systems that make creative work coherent over time.
LlamaGen's working definition
Creative Intelligence
Creative Intelligence is a system-level capability in which persistent project context, controllable multimodal generation, continuous task execution and feedback-aware review work together to produce coherent creative outputs over time.
This is LlamaGen's working definition, not an established academic benchmark or standardized industry term.
Four connected technical layers
Each layer has a different responsibility. Together they describe a path from visual control to context-aware creative execution.
Persistent context
DeepStory
Preserve source-backed story facts, character states, project decisions and reusable playbooks.
Read the Creative Context definitionContinuous execution
Agent Loop
Plan, act, inspect, revise, checkpoint and resume multi-step creative tasks.
Read the Agent Loop definitionEvidence-aware improvement
Feedback Learning / RSI
Use explicit corrections and accepted outcomes as contextual signals, not unverified global preferences.
Read the feedback-learning definitionThe technical series
Definitions, mechanisms and evaluation methods
CyaniModel · Core definition
What Is Controllable AI Image Generation? A Practical Framework for Creative Control
Learn what controllable AI image generation means, how references, structure, masks and editing instructions guide outputs, and where CyaniModel fits.
Core topic: controllable AI image generation9 min read
CyaniModel · Character continuity
How AI Character Consistency Works Across Comic Panels
A technical guide to character identity, changing appearance states, reference selection, multi-character control and evaluating continuity in AI comics.
Core topic: AI character consistency10 min read
CyaniModel · Visual understanding
What AI Needs to Understand in a Comic Page: Panels, Characters and Story Structure
Explore how AI can understand comic pages through panel segmentation, reading order, OCR, character roles, scene graphs and cross-panel story context.
Core topic: AI comic understanding9 min read
Creative Intelligence · DeepStory
What Is Creative Context? Persistent Memory for AI Storytelling
Learn how persistent creative context connects source evidence, canonical story facts, character memory, production playbooks and feedback across AI agents.
Core topic: AI agent memory for creative projects10 min read
Creative Intelligence · Agent Loop
What Is an AI Agent Loop? Continuous Execution and Feedback for Creative Work
Understand the AI agent loop for creative production: plan, execute, inspect, revise, checkpoint, resume and learn from accepted outcomes.
Core topic: AI agent loop for creative workflows10 min read
Personal Media Agent research
From personal context to measurable creative performance
These companion papers examine what a personal agent must remember, how its memory can be corrected, and how creators should evaluate continuous execution. They describe research principles rather than an autonomous Jev service already released.
What Is a Personal AI Agent?
An agent compared with a chatbot: permissioned context, tools and creative outcomes.
Personal Agent Memory
Evidence, approved canon, provenance, scope, correction and reusable playbooks.
From Context to Performance
Accepted results, bounded execution and metrics for a creative feedback loop.
See how this research informs Introducing Jev..
Editorial and evidence policy
Research papers are cited as research context, not as endorsements or proof of LlamaGen implementation. Product capabilities, experiments and proposed architecture are labeled separately. Generated illustrations are disclosed as illustrative concepts, not benchmark results or customer outcomes.
Reviewed November 6, 2026 · English canonical edition · Search-volume and ranking claims are not made unless supported by an inspectable source.