Topaz upscaling

GPUniq serves Topaz Labs production restoration engines — the same models behind Gigapixel AI and Video AI — on a dedicated job-based surface. There is no prompt: you send a source image or video, pick a model, and receive the enhanced result. One API key, the same USD balance, the same job semantics as the rest of the media API.

Topaz is a restoration / upscaling API, not a generator, so it is billed differently from the generative image and video models. Instead of a flat per-image or per-second price, Topaz bills in credits and a job's credit cost scales with output size. GPUniq meters the exact credits each job consumes and charges:

cost_usd = credits_consumed × $0.14 per credit

You are billed only when a job completes — a failed or cancelled job costs nothing. As a rule of thumb, 1 credit covers up to ~24 MP of output for the precision (Gigapixel-class) image models, so a typical 4K upscale is 1 credit ($0.14); generative models (Wonder / Redefine) and video cost proportionally more. The kickoff response returns an estimated_cost_usd, and the completion response returns the exact credits and cost_usd charged.

Image enhancement & upscaling

Two calls: kick off a job, then poll it.

• POST /v1/llm/topaz/image/jobs → returns a job_id in under a second.
• GET /v1/llm/topaz/image/jobs/{job_id} → poll every 2-3 s until the status is completed or failed.

Request

import time, requests

BASE = "https://api.gpuniq.com/v1/llm"
HEADERS = {"X-API-Key": "gpuniq_your_key"}

# 1. Kickoff
start = requests.post(
    f"{BASE}/topaz/image/jobs",
    headers=HEADERS,
    json={
        "model": "topaz-enhance-standard",
        "image": "https://example.com/old_photo.jpg",
        "output_width": 4096,
        "output_height": 4096,
        "output_format": "jpg",
        "face_recovery": True,
    },
).json()["data"]
job_id = start["job_id"]
print("estimated:", start["estimated_cost_usd"])

# 2. Poll
while True:
    time.sleep(2.5)
    d = requests.get(f"{BASE}/topaz/image/jobs/{job_id}", headers=HEADERS).json()["data"]
    if d["status"] == "completed":
        image_b64 = d["image"]["b64_json"]          # inline base64
        download_url = d["image"]["url"]            # presigned URL (valid ~7 days)
        print(f"credits: {d['credits']}  cost: ${d['cost_usd']}  balance: ${d['balance_usd']}")
        break
    if d["status"] == "failed":
        print("failed:", d.get("error"))
        break

The completed response carries both the inline b64_json and a short-lived presigned url for the enhanced image.

Image models

Grouped by task. All slugs are stable GPUniq identifiers; call GET /v1/llm/topaz/models for the authoritative live list.

Video upscaling

Same kickoff-then-poll shape. Because Topaz needs the source clip's metadata up-front to quote the job, pass the source dimensions, duration and frame rate along with the URL.

• POST /v1/llm/topaz/video/jobs
• GET /v1/llm/topaz/video/jobs/{job_id}

Request

import time, requests

BASE = "https://api.gpuniq.com/v1/llm"
HEADERS = {"X-API-Key": "gpuniq_your_key"}

start = requests.post(
    f"{BASE}/topaz/video/jobs",
    headers=HEADERS,
    json={
        "model": "topaz-video-proteus",
        "video_url": "https://example.com/clip_720p.mp4",
        "source_width": 1280, "source_height": 720,
        "source_duration": 8, "source_frame_rate": 30,
        "output_width": 3840, "output_height": 2160,   # upscale to 4K
    },
).json()["data"]
job_id = start["job_id"]
print("estimated:", start["estimated_cost_usd"])

while True:
    time.sleep(3)
    d = requests.get(f"{BASE}/topaz/video/jobs/{job_id}", headers=HEADERS).json()["data"]
    if d["status"] == "completed":
        print("video:", d["video"]["url"])
        print(f"credits: {d['credits']}  cost: ${d['cost_usd']}")
        break
    if d["status"] == "failed":
        print("failed:", d.get("error"))
        break

Video models

Pricing

All Topaz jobs are metered on the exact credits consumed and billed at $0.14 per credit. There is no fixed per-job price — cost scales with output size and model tier:

The kickoff estimated_cost_usd is computed from a free upstream estimate; the completion response returns the exact credits and cost_usd charged. Query GET /v1/llm/topaz/models for the live usd_per_credit rate.

Errors

Topaz requests return the same stable error envelope as the rest of the media API — see the error reference. The codes you are most likely to meet: