Skip to content

Aether

Aether is a canvas that acts: an AI-native creative system for turning one intent into multi-format, multimodal social campaigns.

Core trick

Turns references and a brief into linked key visuals and format variants, with canvas edits propagated across all variants.

Tags

Description

Aether is a canvas that acts: an AI-native creative system for turning one intent into multi-format, multimodal social campaigns. Creative directors and creators today jump between multiple tools, losing context at every handoff. Even after a key visual lands, they still pay the “format tax”: resizing, rewriting, recropping, localising and rebuilding the same idea across placements and languages. Aether rethinks that workflow for what’s next. Drop in references, prompts, URLs or files, and Opus 4.7 turns them into meaningful creative components: brand context, research signals, visual clusters, copy, safe zones, key visuals and variants. The canvas stays editable, with deterministic controls like alignment guides, segmentation, transparent elements, smart text placement and global/local edit propagation. Claude can plan, use tools and coordinate agents, while the creator remains in control. The goal is not “Canva with AI.” It is a responsive creative substrate for social production.
erniesg/aether★ 00 forksTypeScriptCreated Apr 21, 2026 · last push Oct 17, 2026

README excerpt

# aether

aether is a creator-first canvas for turning references, a brief, and product constraints into editable key visuals and linked multiformat variants.

The canvas is the product surface. Rails feed material into it; the bottom composer acts on an explicit scope; generated artifacts return to the canvas for review, editing, propagation, and export. Research, automation, and provenance support that loop rather than becoming separate dashboards.

## Creator loop

1. Gather references, brand material, product facts, a brief, and output targets in the left rail.
2. Compose a multimodal input set from the selected material and constraints.
3. Generate through the prompt composer at the bottom of the canvas.
4. Promote a result into a key visual and make precise canvas edits.
5. Pin useful Claude-driven actions as reusable capabilities.
6. Fan the key visual out to linked format variants; global edits propagate while local overrides remain scoped.
7. Approve and export the pack with provenance.

See [AGENTS.md](./AGENTS.md) for the product and interaction contract.

## Primary surfaces

- `/workspace/<wsId>` — the single synthesis-shell workspace and primary creator surface.
- `/auto-mode` — a supporting automation entry point, not the product shell.
- `/inspect/<campaignId>` — a disclosed run/provenance view for diagnosis and review.

Raw identifiers, payloads, traces, and health information belong in diagnostic surfaces or `?debug=1`, not in the default creator experience.

Team