Pollo AI's parallel tasks, read against the register's job signals

Pollo AI's own page describes its Pollo models plus video and image models from other makers, with parallel tasks sold by plan: one to six, from the Lite tier up to Ultra. The register reads the other half of that arrangement as a hook — how a pipeline learns a job is done. From 8 source pages, read 2026-09-25 to 2026-10-25; no maker pays to be listed.

1What the register already records

Eight of the sixteen webhook rows the register keeps, then the two readings that show the same arrangement from other angles — an API that batches, and a file that simply arrives.

• Features
  Batch APIs of up to 100 requests per call, with webhooksHeyGen, developer site / recorded 2026-09-25
• Page header
  A video API that orchestrates avatars, voices, editing and musicHeyGen, developer site / recorded 2026-09-25
• Inference methods
  Queued requests that are polled or answered by webhookfal, documentation: Model APIs / recorded 2026-09-25
• Welcome
  Notification through a callback when a video is readyColossyan, API documentation / recorded 2026-09-25
• Welcome
  A REST API for generating avatar videos at scaleColossyan, API documentation / recorded 2026-09-25
• Callback protocol
  A callback URL notified when a task's status changes, with signed requestsKling AI, API documentation: callback protocol / recorded 2026-09-25
• Getting started
  Webhooks that report when async jobs completeKrea, API reference: introduction / recorded 2026-09-25
• Request parameters
  A callback URL set when the video task is createdBytePlus ModelArk, API reference / recorded 2026-09-25
• Overview
  A webhook that reports each change in a task's statusPixVerse, platform documentation: webhook callbacks / recorded 2026-10-25
• Guide
  A full-quality download and a share link; the browser plays a lighter copySceneMixer, guide to compositing and sharing an episode / recorded 2026-09-25

2What Pollo AI's own pages state

Pollo AI's page describes its own Pollo models alongside video and image models from other makers, and its plans list a parallel-task count that rises with the tier: one to six jobs at once, from the entry plan up to Ultra. The count is a queue depth, sold as a feature.

For a pipeline, that number is the first half of a question. Starting several jobs is easy to describe; what a pipeline needs next is a way to know that job number four of six has finished, and which file came back. The register files that as a hook of its own, and the rows below are the tools that answer it.

3Starting several jobs and hearing each one end

Sixteen of the twenty-eight tools document a webhook. HeyGen pairs its callbacks with the largest batch described anywhere in the register: batch APIs of up to a hundred requests per call, with webhooks reporting as they finish. fal queues requests and lets them be polled or answered by webhook, with the storage time of the results set per request. Kling notifies a callback URL whenever a task's status changes, with signed requests. Seedance takes a callback URL at the moment the video task is created, and PixVerse's webhook reports each change in a task's status.

Colossyan's reading is the shortest statement of the same design: a REST API for generating avatar videos at scale, with a callback when a video is ready. Krea reports async jobs through webhooks. Nobody in this row had to be called to be read — the rows are quotes from documentation.

4The other way round: a pipeline that asks

Seven tools document polling instead, and some document it beside the webhook. Runway's task polling is built into its SDKs rather than offered as an endpoint to configure. LTX asks to be polled at least five seconds apart; Veo, MiniMax, D-ID and Firefly are polled until a job succeeds or fails, and Wan's task ID stays valid for a day.

Neither pattern is better in the abstract. A webhook needs somewhere to receive a call and a way to trust it; polling needs a clock, a budget for repeated requests and a decision about when to give up. What the register adds is which tools force which choice, in their own words — which is the thing a pipeline lead is actually short of.

5What a queue depth does not settle

The community material keeps a record for one family in particular: Kling holds 15.4% of the posts read in one subreddit, 2,786 of them, with 8,089 posts and 1,318 comment mentions in the file overall. A family that size is what a six-wide queue is usually pointed at, and the material says nothing about how well it queues.

Parallel tasks also do not answer the file question. SceneMixer's own row beside the others is about waiting rather than concurrency: several minutes for a clip of a couple of minutes, an estimate on the progress page, and the option to walk away. A pipeline that starts six jobs and cannot say when the first file lands has bought queue depth and nothing else.

6What each maker's own documentation states

Inclusion rule. Rows are the makers this page is about, each read on its own public documentation, plus the drama pipeline whose hooks this register keeps. A line a maker's documentation does not carry is written here as not carried on the pages read, which describes the reading rather than the product; nothing was called or installed to take it. Order. Alphabetical by maker name.

Inclusion rule. Answers read from each maker's own documentation as this register recorded it — the vendor pages on 2026-10-13, the register rows on 2026-09-25. A line the documentation does not carry is written here as not documented in the docs read, which describes the reading rather than the platform. Order. As the lines are listed above.

7What the collection counted

One family below, with two readings of the material: its share of the posts in one subreddit, and the counts kept for it overall. Neither is a queue measurement.

Inclusion rule. Families the collection counted on this page's subject, from one subreddit. A thread can be counted on both sides. Order. Descending by posts.

Inclusion rule. Model families the collection put a share against, counted from the posts it read. Order. Descending by share.

8When the maker's own way fits the pipeline better

• One account, several makers' models — Its own Pollo models sit beside video and image models from other makers, which suits a pipeline that wants breadth without a key per supplier.
• The queue is sold by size — Parallel tasks rise with the plan, from one to six, so the concurrency a pipeline needs can be bought as a step rather than as a custom arrangement.
• The entry tier still queues — The parallel count is stated on every plan including the entry tier, which lets a small pipeline plan its throughput before paying for the top one.

The words behind this page are "pollo ai video"; the counts below belong to one model family the community material keeps a record for. Beside it: a model maker's own developer platform and batch mode against a pipeline.