**Cost**: Model-priced per successful call. See the Credit Matrix for `image_upscale` model and resolution prices.

Image Upscale supports two input modes:

1. **JSON URL mode**: send `application/json` with the existing URL field.
2. **Direct upload mode**: send `multipart/form-data` with a local image file in the `image` field. The compatible field name `file` is also accepted for the primary image. Codia checks credits before reading and uploading the file, so insufficient-credit requests return `402` without uploading anything.

Direct upload example:

```bash
curl 'https://openapi.codia.ai/v2/open/image/upscale' \
  -H 'Authorization: Bearer {codia_api_key}' \
  -F 'image=@./image.png'
```

Enlarge an image to a higher resolution while reconstructing detail (not just bilinear interpolation). Model-driven — different models trade off sharpness, smoothness, and processing time. Synchronous.

## What this API is for

Recovering print-quality assets from low-resolution sources, preparing low-res user uploads for downstream tasks (e.g. background removal that needs more pixels), or salvaging legacy thumbnails. Use it before any pipeline step that benefits from more pixels — don't chain it after a generative step (those models already output at their target resolution).

## Request

| Field | Required | Notes |
|---|---|---|
| `image_url` | yes | Public URL of the image to upscale. |
| `model` | no | Model ID. Defaults to `codia_image_v2`. See the model price table for available models. |
| `target_size` | no | Target output dimensions, such as `3840x2160`. |

Use `target_size` to request output dimensions where supported by the model. For GPT Image 2 and 2.5, omitted or `auto` target size uses a 1K output adapted to the input aspect ratio and supported size constraints. Other models use their own defaults when target size is omitted.

For nano_banana_2 and nano_banana_pro upscale, explicit 1K/2K/4K values select that output and billing tier. For pixel dimensions, the closest supported aspect ratio is selected, then the lowest tier whose supported width and height cover the target, capped at 4K. For example, 1280x720 uses 1K, 2560x1440 uses 2K, and 3840x2160 uses 4K. Billing follows the selected tier. Omitted or auto target size uses 1K. Model size limits still apply.

## Response

`ImageUpscaleSuccess` shape — `data.image_urls` is an array of CDN URLs (length 1 for this endpoint). Use the first entry.

```json
{ "code": 0, "message": "ok", "data": { "image_urls": ["https://cdn.codia.ai/upscale/xyz.png"], "credits_charged": 16 } }
```

## Billing & limits

- Credits follow the `image_upscale` model/resolution table in the global Credit Matrix.
- The response includes `image_urls` and `credits_charged`, which reports the final charge.
- Use `POST /v2/open/estimate` with the same model and target size to estimate credits.
- Synchronous. Latency scales with input size; very large inputs may approach minute-scale.

## Errors

Standard set — `400` bad input, `401`/`403` auth, `402` credits, `429` rate limit, `500` internal.

> **Sync vs. async**
> This synchronous endpoint returns the result inline — the simplest path. For production pipelines that need webhook delivery, automatic retry with idempotency, or explicit task cancellation, call `POST /v2/open/tasks` with `operation=image_upscale` instead. Inputs and credit cost are identical.

Kunci opsional dari pengguna (maksimal 255 karakter). Jika diisi, respons sukses (2xx) di-cache selama 7 hari. Retry dengan kunci yang sama + body permintaan yang identik mengembalikan respons dari cache (tanpa potongan kredit tambahan); retry dengan kunci yang sama tetapi body berbeda mengembalikan 409. Gunakan agar POST aman di-retry saat terjadi gangguan jaringan. Untuk unggahan multipart/form-data, permintaan dibandingkan berdasarkan field formulir dan isi file, sehingga unggahan yang di-retry tetap diputar ulang meskipun klien memilih multipart boundary baru.

Idempotency-Key

image_url

Model yang menjalankan upscale. Default-nya `codia_image_v2`. Kredit mengikuti tabel harga `image_upscale`.

| Model | Default | 1K | 2K | 4K |
| --- | ---: | ---: | ---: | ---: |
| default / `codia_image_v2` | 16 | - | - | - |
| `ideogram_v3` | 16 | 16 | 16 | - |
| `nano_banana_2` | 18 | 18 | 27 | 41 |
| `nano_banana_pro` | 36 | 36 | 36 | 65 |
| `gpt_image` | 57 | 57 | 115 | 190 |
| `gpt_image_2_5` | 57 | 57 | 115 | 190 |

model

Dimensi output target opsional dalam format WIDTHxHEIGHT. Ukuran output yang didukung bergantung pada model yang dipilih. Harga mengikuti tingkat resolusi output yang diminta. Untuk GPT Image 2 dan 2.5, target size yang tidak diisi atau auto memakai output 1K yang disesuaikan dengan rasio aspek input dan batasan ukuran yang didukung. Untuk upscale nano_banana_2 dan nano_banana_pro, nilai 1K/2K/4K eksplisit memilih tingkat output dan penagihan tersebut. Untuk dimensi piksel, rasio aspek terdekat yang didukung akan dipilih, lalu tingkat terendah yang lebar dan tingginya mencakup target, dengan batas atas 4K. Contohnya, 1280x720 memakai 1K, 2560x1440 memakai 2K, dan 3840x2160 memakai 4K. Penagihan mengikuti tingkat yang dipilih. Target size yang tidak diisi atau auto memakai 1K. Batas ukuran model tetap berlaku.

target_size

image

Field file gambar sumber untuk kompatibilitas (usang: gunakan `image`).

file

URL gambar sumber jika tidak mengunggah gambar

Gambar hasil upscale dan rincian penagihan akhir

code

message

image_urls

Kredit akhir yang dipotong untuk hasil ini.

credits_charged

data

Body permintaan tidak valid atau parameter wajib tidak ada

API key tidak ada, tidak valid, atau sudah dicabut

Saldo akun tidak mencukupi atau langganan telah berakhir

Akun tidak memiliki akses ke endpoint ini

Idempotency-Key dipakai ulang dengan body permintaan yang berbeda.

Terkena rate limit. Tunggu selama jumlah detik di header `Retry-After` sebelum mencoba lagi.

Memperbesar gambar dan merekonstruksi detail, bukan sekadar meregangkan piksel.

Upscale Gambar

Input

Output

API terkait