Editorial Modellix cover reading Vidu Q2 API on two bold lines over a dark amber technical architecture with an amber key and cyan rim light

The short answer: the Vidu Q2 API is real, credit-metered, and asynchronous

The Vidu Q2 API is ShengShu’s official programmatic interface to its Q2 video-generation family — text-to-video, image-to-video, start-end (first/last frame), and reference-to-video, billed per second of output through a credit system. As of August 10, 2026, it is fully operational on platform.vidu.com, requires a Token-style API key, and every generation is an async task: you submit, poll, and download.

This guide covers the official contract first — model names, endpoints, auth, code that runs, and the exact credit-per-second pricing — then compares it with aggregator routes, including Modellix’s. Modellix is an aggregator and has a commercial interest in this comparison: we resell Vidu Q2 routes on our own per-second pricing. None of the numbers below support a blanket “one route is cheapest” conclusion; the winner changes by model, resolution, and duration, and the pricing structures are not directly comparable until you normalize them.

Which Vidu Q2 model are you actually calling?

“Vidu Q2” is a family, not one model ID. On the official API, the model parameter accepts five Q2-era names across the video endpoints, and they are priced differently:

Model name Where it appears Positioning (official docs) Resolution options
viduq2 text2video, reference2video The base Q2 model 540p / 720p / 1080p
viduq2-pro img2video, start-end2video, reference2video “New model with excellent performance” — quality tier 540p / 720p / 1080p
viduq2-turbo img2video, start-end2video “New model with good effect, fast generation” — speed tier 540p / 720p / 1080p
viduq2-pro-fast img2video, start-end2video “Fast speed, low price, stable quality” — cost tier 720p / 1080p
viduq2-image reference-to-image Image generation, up to 4K —

The pragmatic split: viduq2-pro is the quality pick, viduq2-turbo is the speed pick, and viduq2-pro-fast is the budget pick for image-to-video. Text-to-video only exposes the base viduq2. Vidu Q1 and Q2 both remain live, and the official docs have moved their examples to the newer Q3 family — which matters for one decision only: if you are starting a fresh integration today and have no dependency on Q2 output characteristics, the Q3 routes are where the vendor’s own attention (and its newest pricing tiers) now sits. Q2 remains fully documented, cheaper per second at several cells, and is the safer choice when you need the Q2 look or the viduq2-pro-fast price point. The Q2 family is also where Vidu’s subject-consistency pitch lives: reference-to-video keeps a character or object stable across clips, which is the main reason teams still pick Q2 over the newer generation for character-driven work. For the per-second rates and input rules of one specific variant, the Vidu Q2 Pro Multi Frame model page shows the same family under an aggregator’s model ID.

All Q2 video models generate 1–10 second clips (default 5s) on text2video and img2video, and 1–8 seconds on start-end2video. Resolution is 540p/720p/1080p (720p/1080p for viduq2-pro-fast), and 3:4 and 4:3 aspect ratios are supported only on Q2 and Q3 text-to-video.

Getting a key and credits on the official Vidu API platform

Registration is Google-based: sign in at platform.vidu.com, then open the API Keys tab to create a key (each account can hold up to 10 active keys). The API does not bill per request — it bills credits, and every call consumes credits at the rate in the next section.

Credits are purchased in the Billing tab: the standard rate is $0.005 per credit, single top-ups run from $10 to $10,000, and higher-volume packages lower the effective rate (the first-time pack works out to $0.003/credit). There is no free tier; the cheapest way to test is a $10 top-up, which buys roughly 2,000 credits — enough for over a hundred 5-second 540p clips.

Two platform limits matter before you build: concurrency is capped at 5 concurrent tasks on every standard plan (excess tasks queue in submission order), and per-organization rate limits apply regardless of how many keys you create. Usage and credit consumption are visible under the Usage tab, recorded in UTC+0.

The endpoints and the async lifecycle

All Vidu API calls share one host, one auth header, and one async pattern. The base URL is https://api.vidu.com/ent/v2/, and every request carries:

1
2
Content-Type: application/json
Authorization: Token {your_api_key}
Capability Endpoint
Image-to-video POST /ent/v2/img2video
Text-to-video POST /ent/v2/text2video
Start-end (first/last frame) POST /ent/v2/start-end2video
Reference-to-video POST /ent/v2/reference2video
Query a task GET /ent/v2/tasks/{id}/creations

A generation is a task with a lifecycle: created → queueing → processing → success or failed. The submit call returns a task_id immediately; you poll the task endpoint until the state is terminal, then download the video from the returned URL. Optionally, pass callback_url and Vidu will POST status updates to you (retrying up to three times, with a signature header you can verify — documented in the callback signature reference). There is no synchronous mode. If the submit-and-poll pattern is new to your team, our Image to Video API guide walks through the same async lifecycle with generic examples.

Quick start: generate your first Vidu Q2 video

The shortest path is one POST and a polling loop. This curl submits a 5-second 720p image-to-video job with viduq2-pro-fast (the cheapest Q2 video model):

1
2
3
4
5
6
7
8
9
10
curl -X POST "https://api.vidu.com/ent/v2/img2video" \
  -H "Content-Type: application/json" \
  -H "Authorization: Token YOUR_VIDU_API_KEY" \
  -d '{
    "model": "viduq2-pro-fast",
    "images": ["https://your-cdn.example.com/frame-1.jpg"],
    "prompt": "A drone slowly orbits a modern glass building at dusk, warm light",
    "duration": 5,
    "resolution": "720p"
  }'

The response returns task_id and state: created (plus the credits the job will consume). Poll the task endpoint until it reaches success:

1
2
curl -X GET "https://api.vidu.com/ent/v2/tasks/{task_id}/creations" \
  -H "Authorization: Token YOUR_VIDU_API_KEY"

The same flow in Python, using only the standard library plus requests:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
import time
import requests

API = "https://api.vidu.com/ent/v2"
HEADERS = {"Authorization": "Token YOUR_VIDU_API_KEY", "Content-Type": "application/json"}

resp = requests.post(
    f"{API}/img2video",
    headers=HEADERS,
    json={
        "model": "viduq2-pro-fast",
        "images": ["https://your-cdn.example.com/frame-1.jpg"],
        "prompt": "A drone slowly orbits a modern glass building at dusk, warm light",
        "duration": 5,
        "resolution": "720p",
        "seed": 42,  # fixed seed = reproducible frames across retries
    },
)
resp.raise_for_status()
task_id = resp.json()["task_id"]
print("submitted:", task_id)

# Poll with backoff: first check at 10s, then 5s up to 30s, max 15 attempts
attempt, wait = 0, 10
while attempt < 15:
    time.sleep(wait)
    data = requests.get(f"{API}/tasks/{task_id}/creations", headers=HEADERS).json()
    state = data["state"]
    print(f"attempt {attempt + 1}: {state}")
    if state in ("success", "failed"):
        break
    attempt += 1
    wait = min(30, 5 * 2 ** attempt)

if state == "success":
    # The video URL lives in the task result; save it immediately —
    # generated results are not retained indefinitely on the platform.
    print("video ready:", data.get("result_url"))
else:
    raise RuntimeError(f"Vidu Q2 job ended in state: {state}")

Image constraints are worth preflighting before you send anything: one image per job (two for start-end), max 50MB each, PNG/JPEG/WEBP only, aspect ratio must be under 1:4 or 4:1, and the HTTP body — including Base64 images — must stay under 20MB. Upload files to your own CDN and pass URLs where possible; it keeps the request body small and avoids hitting the 20MB ceiling. Two parameters the docs list are worth knowing even though they do little on Q2: seed (pass a fixed value to make retries reproducible — the example above uses one) and movement_amplitude (documented, but the official reference notes it does not take effect on Q2 and Q3 models).

Vidu Q2 API pricing: per-second credits, resolution multipliers, and off-peak

Vidu publishes credit consumption per second of output, at $0.005/credit. The table below normalizes the official credit schedule to USD per second (start cost + per-second increment, as published on the official pricing page on August 10, 2026):

Model 540p 720p 1080p
viduq2-pro-fast (I2V / start-end) — $0.04 + $0.01/s $0.08 + $0.02/s
viduq2-turbo (I2V / start-end) $0.03 + $0.01/s $0.04 + $0.05/s (2s: $0.05) $0.175 + $0.05/s
viduq2-pro (I2V / start-end) $0.04 + $0.025/s $0.075 + $0.05/s $0.275 + $0.075/s
viduq2 (text-to-video) $0.05 + $0.01/s $0.075 + $0.025/s $0.10 + $0.05/s
viduq2 (reference-to-video) $0.075 + $0.025/s $0.125 + $0.025/s $0.375 + $0.05/s
viduq2-pro (reference-to-video) $0.10 + $0.025/s $0.15 + $0.025/s $0.425 + $0.05/s

Two adders change the bill: audio on image-to-video or reference-to-video adds 15 credits per task (voice_id + audio), and the recommended-prompt feature (is_rec) adds 10 credits per task. Off-peak mode — jobs that may complete within 48 hours instead of immediately — runs at half price on the standard tiers. The Q2 audio splitter (audio_type) is a Q2/Q1/2.0 feature; the Q3 family handles audio differently.

A useful rule of thumb: a 5-second 720p viduq2-pro-fast clip costs $0.04 + 5 × $0.01 = $0.09; the same clip on viduq2-turbo costs $0.04 + 5 × $0.05 = $0.29; on viduq2-pro it is $0.075 + 5 × $0.05 = $0.325. The gap between the budget and quality tiers is real and mostly visible in motion fidelity, not resolution.

Vidu official Q2 credit pricing table on platform.vidu.com showing per-second credit costs by model and resolution

Vidu’s official Q2 credit consumption table, captured from platform.vidu.com/docs/pricing on August 10, 2026. Prices are quoted in credits; this guide converts them at the published $0.005/credit rate.

What a batch costs: a worked example

To make the unit comparison concrete, here is one workload — 1,000 five-second 720p image-to-video clips, no audio — priced through three routes with verified rates (all read on August 10, 2026):

Route Unit Price per clip 1,000 clips
Official Vidu, viduq2-pro-fast per second (credits) $0.04 + 5 × $0.01 = $0.09 $90
Official Vidu, viduq2-pro per second (credits) $0.075 + 5 × $0.05 = $0.325 $325
Modellix, vidu/viduq2-turbo-multi-frame 720p per second $0.0322/s × 5 = $0.161 $161
Modellix, vidu/viduq2-pro-multi-frame 720p per second $0.0690/s × 5 = $0.345 $345

Do not read this as a ranking — read it as a demonstration that the same “5s 720p video” spans $0.09 to $0.345 depending on model tier and route. The honest comparison for your budget is: pick the model tier your quality bar needs, then compare the same cell across routes. The official viduq2-pro-fast cell is hard to beat on price; the multi-frame routes on aggregators are priced for a different capability (keyframe-driven animation), not as a generic I2V substitute. Our pricing page publishes the full per-second schedule per model if you want to re-run this arithmetic for your own volumes.

Official Vidu API vs aggregator routes: how to choose

The decision is not “official vs aggregators” as a single axis — it is which contract matches your integration. The table below compares the same Vidu Q2 capability family across the routes that actually rank for this query (each row’s source verified on August 10, 2026):

Route Model IDs Billing unit Auth Best when
Vidu official (platform.vidu.com) viduq2, -pro, -turbo, -pro-fast credits ($0.005 each), per second Token header You want the vendor’s own contract, the cheapest pro-fast cell, and off-peak half-price mode
Pollo AI (Vidu Q2 Pro docs) viduq2-pro, viduq2-turbo, … per video (package credits) API key You want a consumer-grade dashboard plus an API, and are already on Pollo’s ecosystem
Runware (Vidu catalog) vidu/* routes per second API key You want Runware’s queue tooling and storage around Vidu
AtlasCloud (Vidu Q2 catalog) vidu/q2/* routes per video / per second API key You want an OpenAI-style SDK across many models with per-route pricing pages
Modellix (Vidu provider page) vidu/viduq2-pro-multi-frame, -turbo-multi-frame, -pro-digital-human, -turbo-digital-human per second Bearer key You want one key and one submit-poll contract across Vidu, Kling, Wan, Hailuo, and the rest, with per-job cost logs

Aggregators earn their place on the workflow axis, not on a universal price advantage: one key, one billing surface, and consistent async patterns across many models. Modellix’s argument is exactly that — one API key for 12 providers, parameter-level transparent pricing, and per-call logs — not that every Vidu cell is the cheapest. If your only requirement is a single Vidu Q2 price sheet, the official platform is the simplest source of truth; the moment your product spans several video and image models, the per-vendor keys, schemas, and polling loops multiply, and that workflow cost is what an aggregator removes.

Model page for vidu/viduq2-pro-multi-frame showing per-second pricing by resolution

Modellix model page for vidu/viduq2-pro-multi-frame, captured from modellix.ai on August 10, 2026. Aggregator routes price Vidu Q2 per second by resolution, with per-call cost logging on every job.

Errors, limits, and the failure modes nobody documents

The Vidu API returns structured error codes, not HTTP status soup. The ones you will actually meet, per the official error-code reference:

  • FieldLacking / FieldUnwanted — a required field is missing or an unknown field was sent. The message names the exact field; fix locally, do not retry as-is.
  • ImageDownloadFailure — your image URL is not reachable or not valid. Preflight image URLs with a HEAD request before submitting.
  • FieldItemCountOutOfRange / PageSizeOutOfRange — wrong array length (one image for I2V, two for start-end) or oversized pagination.
  • OperationInProcess — a conflicting operation is already running; back off and retry.
  • ImageFormatInvalid — format or size outside PNG/JPEG/WEBP and 50MB.

Beyond errors, three constraints shape production designs: concurrency is 5 on all standard plans (queue instead of failing, but budget queue latency), off-peak jobs may take up to 48 hours (fine for batch, wrong for interactive), and generated results are not guaranteed to persist indefinitely — the platform documentation tells you to retrieve outputs promptly, so download and store every result at success time. If you are polling, use exponential backoff with a cap (the quick-start loop above implements one), log the task_id and state transitions, and pin the model name in your config so a future model update cannot silently change your cost slope.

FAQ

Is the Vidu Q2 API still available in 2026?

Yes. As of August 10, 2026, viduq2, viduq2-pro, viduq2-turbo, and viduq2-pro-fast are all live on the official platform and priced on the official pricing page. The docs now lead with the Q3 family, but Q2 is documented, priced, and served — it is the previous generation, not a retired one.

How much does one Vidu Q2 video cost?

At the official $0.005/credit rate: a 5-second 720p viduq2-pro-fast image-to-video clip is $0.09; viduq2-turbo is $0.29; viduq2-pro is $0.325. Text-to-video on base viduq2 is $0.20 for 5s at 720p. Off-peak mode halves the credit cost. Audio adds 15 credits per image-to-video task.

Is there a free way to try the Vidu Q2 API?

No free tier is published. The minimum top-up is $10 (~2,000 credits at the standard rate, or 5,000 credits in the first-time pack at $0.003/credit), which covers over a hundred short test clips. Some aggregators pair new accounts with starter credits — treat any credit as a test budget, not a production baseline.

Does the Vidu Q2 API support image-to-video?

Yes — img2video takes one start-frame image plus a prompt, and start-end2video takes a first and last frame. viduq2-pro-fast is the cheapest image-to-video tier; viduq2-pro is the quality tier.

What is the difference between Vidu Q2 and Vidu Q3?

Q3 is the newer family with audio-video synchronization and shot segmentation, and it is now the vendor’s headline. Q2 is cheaper at several resolution cells (notably viduq2-pro-fast at 720p/1080p) and remains the better pick when you need Q2 output characteristics or the budget tier. The official pricing page lists both; compare the same model, resolution, and duration before choosing.

Which is cheaper: Vidu official or an aggregator?

There is no blanket answer in the current public data. On the 5s/720p I2V cell, official viduq2-pro-fast ($0.09) is the lowest of the verified figures; Modellix’s viduq2-turbo-multi-frame ($0.161 for 5s) is priced for a different capability — keyframe-driven multi-frame animation — and its digital-human routes serve a different use case. Compare the exact cell for your workload; an aggregator’s argument is workflow (one key across 12 providers, per-call logs), not a universal price floor.

Where can I find Vidu Q2 API code examples?

The official docs include curl, Python, TypeScript, Go, Java, and PHP examples on every endpoint page, and Vidu maintains the viduhq/api-docs GitHub repository for the full reference. The quick-start section of this guide is a complete submit-and-poll implementation against the documented contract.

Is “Vidu Q2 API download” a thing?

The Q2 generation API has no downloadable SDK you must install — it is a REST API you call with any HTTP client, and the “download” intent is usually one of two things: retrieving your generated video URL (poll until success, then download), or cloning the official docs repository from GitHub. If you were looking for a consumer app, that is vidu.com, not the API platform.

Vidu API details and pricing reflect public information as of August 10, 2026 and change frequently; Vidu’s own pricing page is the current reference. Validate against each route’s live pricing before committing. Access image and video models — including Vidu, Kling, Wan, and Hailuo — through a single API key at www.modellix.ai.

Cover image: illustrative Modellix artwork; it is not a Vidu product screenshot or source evidence.