FigCraftFigCraft

Everything in Craft.

One canvas, one local agent. Images, copy, video and drawings — all finished right here.

Live demo · the agent builds, wires and generates nodes on the canvas itself (sample data)
Partners

End to end. One task, done.

Local agent

Breaks it down, routes, executes, delivers — on its own.

It reads the assets and guidelines on your disk, breaks the job down, picks the models, executes, checks its own work and writes the results back to your folder. All on your machine — your files never leave it.

How it works
Local workspace · file opsclaude-opus

The whole canvas, laid out by the agent.

Infinite canvas

Nodes, connections, generation — fully automatic.

Say what you need and it lays out nodes, wires them together and generates them one by one. Every step stays on the board — change the one you dislike and keep the rest.

Download free

It actually understands your image.

Visual understanding

Subject, light, composition, style — read one by one.

It reads your reference before it touches anything. So an edit changes the part you pointed at, not the whole image — and a restyle changes the style while the subject stays put.

How it works
Reading image…

Many agents, working at once.

Multi-agent

Read, search, edit in parallel — results merged for you.

Several agents each take a slice and nobody waits on anybody. The results are merged into a single hand-off — the only thing you need to read.

Download free
figcraft — multi-agent3 agents

What one run hands you

Pick a track to see its format sheet: which aspect ratios come out, and how many of each. All built against one set of brand rules.

1:1
Hero
×4
4:5
In-scene
×3
1:1
Close-up
×4
3:8
Detail page
×1
One set of brand rules, one run12pieces

Anyone can wire up a model. Finishing the job is the hard part

Nano Banana, Seedream and Veo are open to everyone — plugging them in is not a moat. Turning one sentence into a finished set, without drifting or stalling halfway, is the work our own agent does.

It plans first, you approve, then it runs

Anything multi-step starts as a plan: how many pieces, which models, which brand rules. Nothing is spent until you approve it.

Plan, approve, execute — three separate steps

Long jobs do not lose the thread

Dozens of steps in, it still holds the brand rules you set at the start. Oversized intermediate results are compressed in place, and the whole run is summarized as it nears the limit.

100k-token context budget per job, auto-compressed and retried when it overflows

It routes around a flaky model

When an upstream model times out, rate-limits or drops out, it switches within the same turn and keeps going instead of handing you an error.

Falls back to a backup model once retries run out, inside the same turn

Every step is in front of you

Which file it read, which model it called, what it produced, where it got stuck — all listed as it happens. Stop it mid-run if it is going the wrong way.

Tool calls, arguments and results all visible

A loop, not a pipeline

It looks at what it just produced and redoes the parts that are wrong, instead of carrying a bad result all the way to delivery.

Up to 200 self-correcting turns per job

It runs on your own machine

Reads project assets and brand rules straight from disk, and writes finals back to the folder you choose. No upload-then-download round trip.

Reads and writes local disk; finals land in ~/Downloads or a folder you pick

Common questions

If yours is not here, check the docs or get in touch.

Read the docs

macOS 12 or later (separate builds for Apple silicon and Intel), Windows x64, and Linux. Same features everywhere, one account across all three.

Twenty-nine today: 9 for images (Nano Banana Pro, Seedream 5.0, GPT-5.4 Image 2, Wanxiang 2.6, Grok Imagine and others) and 20 for video (Veo 3.1, Seedance 2.5, Kling v3, MiniMax H3 and others). Let the agent choose, or name one yourself.

Project files, layers and local assets stay on your machine. When you run a generation, the prompt and any reference images it needs go to the model provider you picked — no online model can skip that step. The privacy policy spells out exactly what is collected and shared.

By model and output resolution. The cost of a step is shown before it runs, and the full table is on the pricing page.

Yes. Everything that lands on the canvas is a layered project, so you can regenerate a single layer or repaint a selected area instead of starting over. Layers, selections, transforms and shortcuts follow Photoshop conventions.

Yes. Open a canvas room and teammates edit the same board with you, cursors and changes syncing live. How many can be in at once depends on your plan.

Hand your whole visual workflow to an agent that does the work

On your own machine, turn a single line into brand visuals, e-commerce shots, posters and short videos — every one fully editable.

macOS · Windows · Local-first, private by design
Generative-AI service registration·Guangdong-FigCraft-20260817S0078