Hanzo Dev
47286f0b70
Koma is Hanzo's image model, served at api.hanzo.ai under the id `koma` on the OpenAI images schema. This repository is its card: what it renders, what it prices at, the shapes it answers, and what is deliberately absent (edits, video, n > 1). The card states the sizes callers actually ask for against the shapes the render ladder can draw, because they are not the same set — 1792x1024 and 1024x1792 are the two most requested shapes in the API and both are past the pixel budget, so Koma fits rather than refuses. LLM.md records the fitting rule and how to verify a deploy. The upstreams that render Koma are not named anywhere in this repository: Koma is sold as Koma. hanzo.yml declares no image and no test gate, because nothing runs from here — the SKU lives in hanzo/universe's zen catalog and is served by hanzo-inc/zen.
3.5 KiB
koma
Hanzo's image model
Koma
Koma is Hanzo's image model, served from Hanzo's own inference fabric at
api.hanzo.ai under the single id koma. It renders one image per request
from a text prompt, at any aspect ratio from a 9:21 portrait to a 21:9 banner.
curl https://api.hanzo.ai/v1/images/generations \
-H "Authorization: Bearer $HANZO_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"koma","prompt":"a red cube on a white table","size":"1792x1024"}'
{"data":[{"b64_json":"/9j/4AAQSkZJRgABAQAAAQABAAD…"}]}
That is the whole API: the OpenAI images schema, one model id, no vendor
vocabulary to learn. Koma answers b64_json whatever response_format asks for.
What it renders
| Modes | text → image |
| Output | JPEG, base64 in the OpenAI data[].b64_json envelope |
| Images per request | 1 |
| Aspect ratios | 9:21 5:11 1:2 7:13 3:5 2:3 3:4 6:7 1:1 7:6 4:3 3:2 5:3 13:7 2:1 11:5 21:9 |
| Sizes | any size you send; Koma fits it to the nearest ratio it can draw |
| Latency | 1.6 s to 4.4 s per render, measured on production hardware |
Koma is a single image per call, and says so rather than quietly serving one
for an n of four: n: 4 is refused with a message that asks for one call per
image. The per-image price is then right for what was delivered.
Sizes
You may ask for the size you actually want — 1792x1024, 1024x1792,
1536x1024, 256x256. Koma fits the request to the nearest renderable shape
instead of refusing it, because the raw render ladder is narrower than the sizes
people ask for: each edge is on a 16-pixel grid between 512 and 1568, and the
whole image is at most 1,062,400 pixels. 1792x1024 becomes 1344x768 — the
same 7:4 ratio, the largest render that fits.
aspect_ratio wins when you send it, and drops size/width/height. Stating
both is a contradiction, not a preference.
| you ask | you get | ratio |
|---|---|---|
1024x1024 |
1024x1024 |
1:1 |
1792x1024 |
1344x768 |
7:4 |
1024x1792 |
768x1344 |
3:7 |
1536x1024 |
1248x832 |
3:2 |
512x512 |
512x512 |
1:1 |
256x256 |
512x512 |
lifted to the floor |
Pricing
$0.16 per image, flat. No charge for a refused request, and no charge for an image that failed to render. Koma is a paid model: free accounts reach it on the free allowance (50 requests a day), a paid plan reaches it on its own included usage.
| Model id | koma |
| Price | $0.16 / image |
| Endpoint | POST /v1/images/generations |
| Minimum tier | paid |
Not Koma
- Image editing is not available on Koma yet.
POST /v1/images/editsand animagefield are refused rather than silently ignored. - Video is not Koma. There is no video model on the platform today.
Surfaces
Koma is a product SKU, not a repository you run. The inference lives in
hanzo-inc/zen; the catalog that names it
(price, rungs, retail) is declared in hanzo/universe
(charts/app/files/zen/catalog.yaml) and reconciled by Hanzo CD.
- API and docs: https://api.hanzo.ai/v1, https://docs.hanzo.ai
- Dashboard: https://platform.hanzo.ai
- Agent notes for this repo: LLM.md
License
The contents of this repository are licensed under Apache-2.0. The model Koma serves is proprietary and no weights are distributed — Koma is available only through the Hanzo API under the Hanzo Terms.