v0.5 — image-only
One call, N² coherent tiles
USDC · Base
Image API
for AI agents.
One call → an N×N grid → N² tiles · fal + xai + openai · pay per call in USDC on Base
§ I
One prompt. Sixteen tiles.
Send a prompt, receive a grid. Expand draws an N×N composition in a single provider call and slices it into N² independent images. Pick your provider, your grid, your latency budget — or let “auto” choose.
16
Tiles per single 4×4 call
3×
Providers, one endpoint
~6s
Typical xai-routed latency
POST https://glyphapi.dev/v1/expand X-Agent-Address: 0x... X-Glyph-Latency-Budget-Ms: 8000 { "prompt": "cyberpunk samurai sprite sheet", "grid": "4x4" } // → { images: [16], prompt_hash, charge: "$0.20", next_actions }
§ II
A grid that calls back.
Every expand response carries a prompt_hash
and a next_actions block. Re-fetch the grid
hours later (free), re-roll the same intent for a fresh sample
(paid like a new call), or promote one tile to a full-resolution
asset. The agent doesn’t need to remember the URL shape; the
response tells it what to do next.
GET
Re-read tiles, free, 24h cache
POST
Re-roll same intent, new sample
$0.02
Refine one tile to full res
GET https://glyphapi.dev/v1/grids/<prompt_hash> POST https://glyphapi.dev/v1/grids/<prompt_hash>/variations POST https://glyphapi.dev/v1/refine { "prompt_hash": "<hash>", "tile_index": 5, "mode": "upscale" } // → { image, charge: "$0.02" }
Rates
Paid per call.
In USDC on Base.
/v1/expand · /v1/grids/.../variations
Grid 2×2 (4 tiles)$0.10
Grid 3×3 (9 tiles)$0.15
Grid 4×4 (16 tiles)$0.20
GET /v1/grids/<hash>free
/v1/refine
Upscale 4×$0.02
Regenerate full-res$0.10
Cache TTL24h
SettlementUSDC / Base
Settles on-chain · no subscription · no keys to rotate