Files

2.5 KiB

Iro — agent notes

Iro (iro, iro-1) is Hanzo's image generation and editing SKU. This repo is its model card, API contract, and deployment spec; inference serving runs in hanzo-inc/zen. Read this before changing anything that serves iro.

Where Iro lives across the stack

Layer Repo / file What it owns
The SKU, its price, its rungs hanzo/universe → charts/app/files/zen/catalog.yaml the iro entry (mode: image, zen-image: *iro, zen3-image: *iro, alias iro-1: iro)
The serving plane hanzo-inc/zen → media.go mediaProxy (gate, meter, relay), nativeImage (request shape), fitImage/imageSize (size fitting), openAIImage (answer shape)
Cloud gateway & Studio hanzo-inc/cloud → apps/ai/media_rpc.go, apps/studio/exec.go /v1/images/generations and /v1/images/edits at api.hanzo.ai; Studio step defaults image.generate and image.edit to iro
Studio UI & wire client hanzo-inc/studio → src/kinds/image.tsx, src/wire.ts iro is the default top image model; decodes data[0].b64_json into canvas asset
1-click K8s Stack hanzo-inc/platform → src/lib/deploy/stacks.ts iro stack template (ghcr.io/hanzoai/iro:latest)
Product page hanzo-inc/app → lib/creative-models.ts /iro (iro-1, open foundation zen3-image / zenlm/zen-artist)

The translations zen performs for iro

  1. Request (nativeImage): When a rung carries path: (e.g. /v1/genai/black-forest-labs/flux.2-klein-4b), model, n, response_format, and cfg_scale are stripped from the JSON body and size (WxH) is fitted to a width/height pair on the 16-pixel grid [512, 1568] within 1,062,400 total pixels (fitImage).
  2. Answer (openAIImage): Upstream {"artifacts":[{"base64":"..."}]} is translated into OpenAI-compatible {"data":[{"b64_json":"..."}]}.

Verifying a deploy

# 1. Verify iro is in the live model catalog
curl -s https://api.hanzo.ai/v1/models | jq '.data[] | select(.id=="iro")'

# 2. Generate an image with iro
curl -s https://api.hanzo.ai/v1/images/generations \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"iro","prompt":"a red cube on a white table","size":"1792x1024"}' \
  | jq -r '.data[0].b64_json' | base64 -d > /tmp/iro.jpg

file /tmp/iro.jpg   # JPEG image data