One fixed Prompt, run on both models. Clips and measurements are published once the runs are complete.
Summary and how to choose
Duration: 4–15 seconds for MiniMax H3, versus 5 seconds for Wan 2.2. Resolution: 768P / 2K for MiniMax H3, versus 480P / 720P for Wan 2.2. Controls: T2V · first/last-frame I2V · R2V for MiniMax H3, versus T2V · I2V · TI2V for Wan 2.2. Audio: Native 32 kHz stereo audio; accepts reference audio in reference-to-video requests for MiniMax H3, versus not confirmed for Wan 2.2.
MiniMax H3
Strengths: a documented output envelope (4–15 seconds · 768P / 2K), documented generation controls (T2V · first/last-frame I2V · R2V), documented audio support (Native 32 kHz stereo audio; accepts reference audio in reference-to-video requests), published exact per-second API rates, and published input/token pricing rules.
Wan 2.2
Strengths: a documented output envelope (5 seconds · 480P / 720P), documented generation controls (T2V · I2V · TI2V), and a checked plan, credit or no-rate path.
Trade-offs: audio support is not confirmed and pricing is not normalized to a per-second rate.
Choose MiniMax H3 when transparent exact API pricing, documented generation controls, documented audio output, and its documented duration and resolution envelope matter most. Choose Wan 2.2 when documented generation controls, a checked plan, credit or no-rate path, and its documented duration and resolution envelope matter most.
No overall output-quality verdict is available for this pairing yet. This summary compares confirmed documentation and pricing only; it does not infer visual quality or rank models.
Prompt 1 of 1
Durations differ. MiniMax H3 is produced at 10s and Wan 2.2 at 5s. These are recorded deviations rather than identical conditions — see the conditions under the Prompt for why.
Not run yet
Planned · 5s · 720P · 16:9
No clip is published for this model on this Prompt yet.
Prompt
A single continuous shot, 16:9, no cuts. A young ceramicist sits at a wooden workbench beside a large window, shaping a wet clay bowl with both hands. Her fingers press and smooth the clay as the bowl turns; the clay visibly deforms under her touch and slips slightly between her fingers. Steam rises slowly from a stoneware mug beside the wheel. She lifts the bowl a little and turns it toward the camera. The camera performs one slow, smooth dolly-in from a medium shot to a close-up, with no other camera movement. Warm morning window light comes from camera-left, with soft shadows, shallow depth of field, and fine dust visible in the air. Photorealistic live-action footage, natural and unhurried. The wet clay reads matte and slightly glistening, the wood grain of the bench stays legible, and the glaze on the mug catches soft specular highlights. Her face, hands, clothing, and the bowl stay consistent throughout, with anatomically correct hands and fingers. Ambient studio sound only. No cuts, no camera shake, no extra people, no text, no logos, no subtitles, no watermark.
One continuous shot · 16:9 · no reference image · duration per model: MiniMax H3 and Seedance 2.5 at 10s, MiniMax Hailuo 2.3 at 6s (Hailuo 2.3 supports only 6s or 10s, and 10s would drop it to 768P), and Wan 2.2 at 5s (its reference model card documents 5-second 720P generation) · highest available resolution per model · actual parameters and any deviations recorded per result
List-price signal
Checked 2026-09-19
Exact and estimate rows are USD per output second. Rows marked Exact are published provider rates; rows marked Estimate are official examples calculated from token consumption, not fixed per-second plan prices. Plan, credit and token paths are grouped separately. A dash means no comparable checked price.
| Basis | MiniMax H3 | Wan 2.2 |
|---|---|---|
| USD / output second | ||
| 768Pexact | $0.08 | No comparable checked price |
| 2Kexact | $0.13 | No comparable checked price |
| Other checked billing paths | ||
| Model fee | No comparable checked price | $0 |
| CPU Basic | No comparable checked price | Free |
| Nvidia T4 small | No comparable checked price | $0.40 / hour |
| 1x Nvidia L4 | No comparable checked price | $0.80 / hour |
| 1x Nvidia A10G small | No comparable checked price | $1.00 / hour |
| Wan 2.2: Compute examples, not a generation price; actual cost depends on VRAM, throughput and hosting path. | ||
This is a price signal, not a quality ranking. Resolution, capabilities, input materials, failed generations, taxes, and regional account terms are not normalized.
- MiniMax pay-as-you-go pricing · checked 2026-09-19
- Hugging Face pricing · checked 2026-09-19
API comparison
6/7 and 5/7 fields confirmed from official sources
Version-level API facts, taken from the provider documentation cited below. Unconfirmed means we have not verified it from an official source — it does not mean the feature is unsupported.
| Field | MiniMax H3 | Wan 2.2 |
|---|---|---|
| Official model ID | MiniMax-H3 | Unconfirmed |
| Duration | 4–15 seconds | 5 seconds |
| Resolution | 768P / 2K | 480P / 720P |
| Modes | T2V · first/last-frame I2V · R2V | T2V · I2V · TI2V |
| Aspect ratios | Unconfirmed | TI2V-5B: 1280×704 / 704×1280 |
| Native audio |
Native output statement is scoped to MiniMax H3; reference-audio input alone does not establish native output for H3 Max. | Unconfirmed |
| Weights license | Community license; EU, UK, South Korea and US excluded | Apache-2.0 |
Sources · checked 2026-09-19
- MiniMax V2 video generation API · checked 2026-09-19
- Wan2.2-TI2V-5B model card · checked 2026-09-27
- Wan2.2-I2V-A14B model card · checked 2026-09-27
- MiniMax H3 model card · checked 2026-09-19
- MiniMax H3 Community License Agreement · checked 2026-09-06
Video comparison from other creators
A third-party creator ran both models and published the walkthrough below. It is embedded from YouTube: vidirect.org did not produce it, hosts none of it, and does not endorse its conclusions.
Same prompt run across MiniMax H3, Wan 2.2 and Wan 2.1. Watch on YouTube ↗ · Learn Digital Art with William Jiamin
Where to run these models
This site does not generate video and does not sell generation credits, so it has no reason to prefer either model. To run them, use the provider’s own API or app.
Frequently asked questions
Did these two models run the same Prompts?
Not yet. This pairing is a planned test: the Prompt is fixed and clips are published once both models have been run.
Can I generate these videos on this site?
No. This site does not generate video and does not sell generation credits; it compares models and links out to each provider. Running these models means using the provider's own API or app.