Keep the worker. Skip another startup.
A live handoff worked. The shared eight-second generated section was byte-identical; waiting in line was still real.
Same source, reused worker
The original sound was preserved in the test file. This public preview is silent; the generated tail adds no sound.
Inspect the join below: the synchronized comparison covers the shared first 16 seconds.
Inspect the join · first 16s comparison
Compare Same source, reused worker
Inspect movement closely
Both videos stay on the same timeline at every speed; this does not retime the character’s moves. Frame buttons seek by 41.7 ms on the output grid; browser decoding and a different source frame rate can affect the displayed frame.
Loading the comparison videos… Open the result file
Silent publication previews. This pair compares the shared interval; full-length files are below. Audio findings refer to preserved original test files.
What to look for Both include the original eight seconds. Compare seconds 8–16, the shared generated section. The longer cold result has a second section. Use the full-file players below to see each ending.
Original video, full results & downloads
AI-generated outputs are identified in each comparison. Public previews omit source audio where present; visual streams are unchanged. Original test files and audio measurements remain preserved. No performer or photographer endorses ModelPilot.
The finding
Can queued videos share a ready worker without changing the recipe?
Yes. A compatible queued job can inherit a ready worker and its remaining time allowance. That avoids allocating another machine. It does not reduce sampling steps or select a cheaper-quality recipe. The public previews are silent: original-audio retention describes the preserved test files and measurements, not a soundtrack distributed on this page.
One cold/warm Extend handoff plus a later two-job replacement handoff. Not a controlled speed benchmark.
What we would choose
Use worker reuse automatically when a compatible ready worker exists. Show queue wait separately from processing time, and keep the original worker deadline.
Where it still falls short
- The cold case made 16 new seconds; the warm case made eight. Their total times and charges cannot establish a controlled percentage saving.
- The warm job waited 31m 32s for the worker. Reuse can save another startup without giving the customer an earlier result.
- Byte-identical footage is established for one shared generated section, not a promise of identical GPU outputs generally.
What changed
| Choice | Baseline / first option | Candidate / second option |
|---|---|---|
| Video recipe | Same extension recipe | Same extension recipe |
| Worker state | Cold allocation | Compatible ready worker |
| New footage requested | 16 seconds | 8 seconds |
| Worker deadline | Set at allocation | Inherited; not reset |
Time & cost, without the guesswork.
| Saved run | Recorded processing | Recorded amount | Scope |
|---|---|---|---|
| Cold Extend · +16s | 30m 55s | $3.4775 | Queue: 1m 25s |
| Reused worker · +8s | 7m 57s | $1.3813 | Queue: 31m 32s |
| Later cold replacement · 4s | 7m 51s | $0.9760 | Queue: 16s |
| Next replacement · 8s | 3m 31s | $0.6100 | Queue: 6m 39s |
Recorded Studio wallet debits. Different output lengths and cold/warm states prevent a recipe-speed comparison. Provider usage is another view of the same GPU work, not an extra cost to add.
NVIDIA B200 in the saved Extend worker detail; the warm case inherited the same physical worker. The later two replacement cases also shared a B200.
Inspect the setup
- Shared generated section SHA-256
- ccdeedf4104b13ac12e13134b351409a7695ab7e35b11dfe3b3216c783f93d62
- Later replacement handoff
- Two completed jobs, one physical worker; no second replacement allocation
- Compatibility matters
- Model, runtime, hardware and available lease time must match. This is not an unconditional promise of instant capacity.
- Quality observation
- The identical shared section isolates the handoff from a recipe change; unrelated quality issues remain visible.
- Hardware and timing limits
- NVIDIA B200 in the saved Extend worker detail; the warm case inherited the same physical worker. The later two replacement cases also shared a B200.
Study settings summary; not an importable ComfyUI workflow. Requires the private ModelPilot runtime/nodes and separately licensed model weights.
Download study filesIncludes recorded settings, media hashes, credits and license notices. A complete executable graph is not included for this study.
Sources, attribution & publication status
Source credits, original licenses and adaptation notices are recorded below. Results are AI-generated research demonstrations, not endorsements. Model use and output redistribution remain subject to the MiniMax H3 license and any required separate authorization.
All videos on this page are saved outputs, not new generations. Settings and conclusions were transcribed from the original experiment receipts. No customer accounts, private batch identifiers or credentials are included in the downloadable summary.
- Patricia Stephan, TAF German Bellydance National Championship 2024 — Dance — Shark1989z. Excerpt from original 25s onward; resampled/prepared for the saved test; audio omitted from public previews. AI-generated extensions are adaptations, not new footage of Patricia Stephan. Copyright permission record: Wikimedia VRT ticket 2024112010013196.. License recorded at acquisition.
Source credits and changes accompany each download. Read the media notices. Public previews omit source audio; the original test files remain preserved. ModelPilot has received authorization for its H3 use and works. This does not grant rights to third-party source material.