Agnes Video Generator

Jimeng workflow FAQ (including how to handle Agnes 503 / queue is full)

What do you do when Agnes returns HTTP 503 or video queue is full? Should you wait? Can you take another route? Plus the questions people ask most while using the Jimeng workflow.

This page first answers the step where most people actually get stuck — upstream queueing — and then the details of the Jimeng workflow itself. Every item is something you can act on directly.

The order to handle a 503 / queue is full

Try them in this order — in the vast majority of cases you will have a clip within three steps:

  1. 1

    Identify which kind of rejection it is

    The panel should show a clear cause: video_queue_full (queue full), a bare 503 (upstream overload) or model_not_found (no available channel for that model). The first two just need patience; the last requires switching models.

  2. 2

    Retry off-peak

    The free queue is busiest during peak hours. Let the app queue and retry on its own (about 15 minutes by default), or go do something else and click “Retry task” during an idle window to resume from the failed step.

  3. 3

    Lower the per-run load

    Fewer scenes, shorter per-clip durations and a smaller resolution all reduce the per-run footprint and the wait; run one task at a time so several tasks do not fight over the submission quota.

  4. 4

    Switch video model

    Queues are pooled per model, so switching to another free model often gets you a slot right away; note that older models removed by the gateway reliably return model_not_found.

  5. 5

    Take another route: the Jimeng workflow

    If you do not want to wait, have Jimeng generate the picture while this project handles only the storyboard, subtitles and stitching. See the Jimeng workflow guide linked on this page.

Read the Jimeng workflow guide →

The short answer first: 503 is not your fault

Why do I get HTTP 503 or “video queue is full” when generating a video?
The Agnes upstream free video queue is saturated (queues are pooled per model), and the request is rejected by the gateway before it is queued. It is **not** a problem with your prompt, settings or Key. A rejection creates no task and consumes no quota, so you can retry without worry. During peak hours this lasts ten minutes or more, which is why it keeps happening. One more note from the provider: the Agnes China site enforces real-name verification from 2026-10-26 12:00 (UTC+8) — unverified accounts get a lower API call rate — and upstream expects 503 / queue is full busy errors to improve significantly after that point.
Agnes keeps returning 503 — does the app retry automatically, and for how long?
Yes. Every 503 (including a bare 503 with no error code) is normalized to a “busy” signal and retried on a separate busy-track: the first retry is immediate, after that it keeps queueing at a fixed interval, up to 15 times by default (adjustable via the environment variable AGNES_VIDEO_BUSY_RETRY_ATTEMPTS). The progress panel shows in real time “Agnes video queue is full (HTTP 503 · video_queue_full), queueing to retry (attempt N / waited X minutes)” and keeps the upstream’s original error.
Retries still do not get me a slot — do I have to keep waiting?
No. A saturated queue is a matter of timing, so retry off-peak or later; in the meantime you can lower the per-run load (fewer scenes, shorter durations) or switch video model (queues are pooled per model). If you simply want a clip right now, switch to the Jimeng workflow: Jimeng generates the picture and audio, and this project handles the storyboard, subtitles and stitching.
If a retry fails or is interrupted, does the whole run go to waste?
No. Clicking “Retry task” resumes from the step that failed: storyboards and reference images already generated are kept, not resubmitted, and not charged twice. In External Asset mode the imported video clips are kept too, so you only fill in the missing segments.
Are 503 and 401 the same kind of problem?
No. 503 is an upstream capacity problem, so wait or take another route; 401 means the Key does not match the domain you are connecting to (the request carries a Key, but that site does not accept it), so you need a different Key or to re-bind the domain. Both have their own documentation, so do not troubleshoot them as one.
Does the Jimeng workflow cost extra?
Jimeng itself grants daily credits, so use the free credits first; whether to top up when they run out is up to you. This project remains MIT open source and free to self-host, and switching to Jimeng only changes the source of the “video generation” step, which adds no cost to this project.
Why does the Jimeng workflow ask me to paste twice? How do the “storyboard” and “video clips” prompts relate?
The prompt at the storyboard checkpoint is a “creative brief” — it tells Jimeng (or any LLM) how to refine the storyboard under picture + audio + duration constraints; the prompt at the video clips checkpoint is simply this project’s finalized storyboard script, numbered segment by segment as [Scene i/N · duration], so you can copy and paste it to generate. The former is for polishing, the latter for execution.
Why can’t I see the “External assets” option in the web Studio?
Right. External Asset mode is currently only available in the self-hosted desktop app (v7.6+); the web Studio and the online Demo still use automatic mode. If you want the web version to support external assets too, vote for it in an Issue.

Your options when you hit 503

There is no single right answer — pick according to what you need right now:

StepAutomatic / manual modeExternal Asset mode (Jimeng)
Storyboard and promptsGenerated by this projectGenerated by this project, plus a per-scene prompt set you can paste straight into Jimeng
Picture and audioAgnes video models (the API frequently returns 503; only the voiceover comes from this project’s edge_tts)Generated by Jimeng (with native audio), exported as one mp4 per scene and imported in a batch
SubtitlesAligned to word-level timestamps from the voiceover (more accurate)Estimated from each storyboard’s real duration (may drift, check it at the subtitle checkpoint)
Stitching and final cutProject stitchesThis project normalizes + stitches while keeping the original audio track + burns in subtitles

Are there other free video generation channels?

It would be no surprise, and we are happy to add them. This project’s tool list is an extensible registry: adding a tool only means adding a set of prompt templates and fill-back instructions, and neither the workflow nor the compositing layer has to change. It is even easier to integrate if your channel has native audio, controllable native duration, or can export subtitles. Open an Issue on GitHub with the tool name, free quota, duration tiers and whether audio is supported, and we will evaluate adding support.

GitHub Issues →

Related Pages

Jimeng (Dreamina) Workflow Guide (Beta Feature)

Hand only the picture and native audio to Jimeng, while storyboards, dialogue, subtitles and stitching stay in this project — an alternative path when the Agnes API keeps returning 503.

Agnes Video Generation Errors: HTTP 503 Queue Full, 401 Token & Retry Fixes

How to fix Agnes Video Generator errors: HTTP 503 video queue is full, 401 invalid token, failed or interrupted generations with checkpoint resume, plus system requirements and API Key basics.

Manual Mode Guide — Auto vs Manual: Vibe Generating or Precise Output | Agnes Video

How to use Manual Mode in Agnes Video Generator: pause at checkpoints (storyboard, references, video clips, narration, subtitles, final) to review or refine each artifact before continuing, optionally pairing with stronger external AI or tools for a precise, predictable result. Auto Mode fits vibe video generating from a rough idea; Manual Mode fits clearly specified outputs with fine-tuned artifacts.

Is Agnes Video Generator Free? Free Quota & Commercial Licensing Explained

Why Agnes Video Generator is completely free, how the free quota works, whether you can use it commercially, and how it compares to Runway, Pika and other free AI video tools.

Error & Retry Basis

This page's attribution of HTTP 503 / video queue is full, its retry strategy, and off-peak advice are based on the following upstream behaviour records and implementation source.

  1. Agnes AI Open API — 视频 / 图片生成接口Agnes AI · apihub.agnes-ai.com (2026)

    HTTP 503 and the body error codes (video_queue_full, etc.) are returned by the Agnes Open API gateway; rejection happens before queueing and consumes no render quota.

  2. agnes-video-generator — 开源 AI 视频生成项目lcy362 · GitHub (2026)

    Busy-track classification and attempt caps (all 503s treated as busy, 15 attempts by default, immediate first retry) plus resume logic live in core/api/retry_policy.py and core/api/agnes_video.py.

  3. 即梦 AI (Dreamina)：AI 图片与视频创作平台ByteDance · 即梦官网 (2025)

    Jimeng (Dreamina) as the alternative path: its daily credits and video model tiers ground this page's judgement about switching tools.

Why you can trust this

Written by a practicing developer and cross-checked against authoritative primary sources such as arXiv papers, vendor technical reports, and the Stanford AI Index.

S

SandGrid@lcy362

Author of Agnes Video Generator · Full-stack Developer

Independent developer and author of Agnes Video Generator, an open-source (MIT) AI video generation tool built on Agnes AI's free video models. Has helped 1,000+ creators produce AI videos at zero cost. Focused on making video generation models accessible and production-ready; content is based on hands-on practice and primary research.

View on GitHub

Last updated：2026-08-21

Ready to Start Creating?

"Making world-class AI belong to everyone." — Bruce Yang. It's completely free, no credit card, and you won't need a high-end GPU. Your first AI video starts at zero cost. Want to use Agnes AI's free video models? This is the easiest way in.

Try Online DemoGet Started Free

Clone the GitHub repo and launch in 2 minutes