Full-size reference
MiniMax H3 timeline$3.85for this run
16.2 seconds of video
- Start to download
- 26 min 08 s
- Output size
- 756 × 1344
- Machine
- B200 secure
- Recorded
- 2026-10-14
Video cost guide
What does a finished video cost?
Look at the whole clip: what you get, how long you wait, and what the run costs. A low GPU rate alone doesn’t answer those questions.
Sign in to review a current Studio quote. These historical H3 runs are not quotes or examples of the separate Wan workflow.
What these runs actually cost
Full-run charges, including startup. Different inputs and settings can change the result.
Records updated 2026-10-14
Smaller reference
MiniMax H3 timeline$1.73for this run
18.5 seconds of video
- Start to download
- 11 min 51 s
- Output size
- 736 × 1394
- Machine
- B200 secure
- Recorded
- 2026-10-14
The smaller-reference run cost less and finished sooner. These clips also differ in duration and output size, so the pair does not isolate the cause or prove equal visual quality. Inspect an output before choosing a setup on price alone. See the separate matched motion-input study and full videos.
Before you start a video
Choose the result first. Then check the cost and wait for your own inputs.
1. Add your inputs in Studio
Choose an available action. Animate an image with a motion prompt, or supply a performance video and character picture.
2. Save a draft and review the estimate
Check the duration, settings, expected cost and spending limit before starting generation. Preparing your draft is free.
3. Return to your saved result
Automatic runs continue in the background. Come back to Saved work for progress and completed downloads. A completed run may still have visual defects.
Why “generation cost” can look much lower
Warm compute measures a model that is already loaded. For a single clip, getting that machine ready can cost more than generating the frames.
LTX-Video 0.9.8 2B · 3.9-second clip
$0.052Recorded pod total for one clip, including startup
$0.0004Warm compute per video-second; excludes startup
Quality tradeoff in this test: missed object counts, camera direction and on-screen text. End-to-end delivery time was not recorded. Measured 2026-08-17.
Run details and measurement notes
These are individual experiments, not a controlled model ranking. Full-run charges, warm-only compute and older notes are kept separate. No current price or visual-quality guarantee is inferred from them.
| Run | Wait | Warm seconds / video-second | Cost / video-second |
|---|---|---|---|
| MiniMax H3 timelinefull reference | 26 min 08 s | — | $0.2376 |
| MiniMax H3 timeline384x736 reference | 11 min 51 s | — | $0.0935 |
| LTX-Video 0.9.8 2B8 steps, euler | — | 2.2 s | $0.0004 warm only |
Older timing notes
The following entries have not had their transaction receipts re-verified and are not used as price estimates.
- MiniMax H3: 8.7 s of warm compute per video-second · A6000 class · 2026-08-17.
- Wan Animate: 35–60 s of warm compute per video-second · A6000 class · 2026-08-17.