QINAXIS
All products
Desktop · AI Agent

FigCraft

The coding agent for images.

Overview

A locally-run image & video creation agent. State your intent in one sentence; it breaks the work into sub-tasks, calls tools, generates in parallel, and finishes the whole set — like a coding agent that actually does the work.

80+
Tools
16
Top models
200
Reasoning turns
100%
Runs locally
  • 01Conversational image generation across leading models
  • 02Photoshop-grade editing — layers, selection, masks
  • 03AI video generation built into the same workflow
  • 04Agentic workflows that carry a task to a finished result
How it works

Not a prompt-level model — a task-level agent. You describe the need; it decomposes, schedules, executes and delivers.

1

State intent

One sentence; it reads your project assets locally.

2

Plan

Decomposes into sub-tasks and a visible plan — credits spend only after you approve.

3

Generate in parallel

Picks the best model per sub-task and runs many in parallel.

4

Write back locally

Outputs land in your local folders by naming convention — nothing goes to the cloud.

In action
Home · intent in one sentence

Home · intent in one sentence

Pick a model and generate; common creative scenes built in.

Creation workspace

Creation workspace

Layered canvas + agent task flow, parallel generation visible throughout.

Video line · multi-shot timeline

Video line · multi-shot timeline

Orchestrate shots, scrub frames, preview the cut.

Agent architecture

Fully self-developed, from the reasoning engine to 80+ tools — the group's core technical asset.

01

Two-layer context compression

Short-term token budget + auto-summary, with long-term persistent memory — it still remembers early assets dozens of turns in.

02

Cross-vendor fallback chain

When the primary model fails it auto-switches to a backup — any model (even Qwen) can reliably deliver a full set of visuals.

03

Three-layer consistency anchor

Upload / hero / chained anchoring keeps a subject strictly consistent across images and shots.

04

Structured video line

Per-shot first/last frame, reference clip and voice are controllable — a set of stills becomes a brand short film.

Toolbox & governance

80+ self-built tools, each chosen per turn and graded into three risk tiers — so brands can safely hand execution to the AI.

Read-only tools

Analyze image / search references / capture canvas

Parallel, fastest

Destructive tools

Generate / edit / compose / export

Serial, conflict-safe

High-sensitivity tools

Shell commands / overwrite / bulk delete

Confirm every time, no bypass

AI video generation

Not just single clips — the agent storyboards like a director and keeps consistency across shots.

01

Video line · structured production memory

A video splits into shots, each a structured contract (prompt / frames / refs / audio / params) the agent freely orchestrates.

02

First-frame provenance · anti-drift

Each shot declares where its first frame comes from, keeping characters and scenes strictly consistent across shots.

03

Dynamic asset library

Characters / scenes / audio / text blocks are managed and reused for consistency and efficiency.

04

Timeline · continuous preview

Clips play continuously, drag to reorder, scrub frames — production stays visible and adjustable.

Made with FigCraft

Every image here was produced in FigCraft — from commercial product shots to campaign posters.