What is P-Video-2-Pro?
P-Video-2-Pro is a video generation model from Pruna AI, the Munich model-compression company, released in September 2026. Pruna's documentation describes it as built on MiniMax H3 and positions it as the quality tier of its P-Video family. One endpoint takes a text prompt, a first-frame image, or a first and last frame, and returns a 5 to 15 second clip at 24 fps with generated dialogue, music and sound effects. You cannot supply your own soundtrack.
It generates at 480p or 768p; 1080p, 2K and 4K are the 768p output upscaled by a separate Pruna model, at an extra $0.02, $0.04 or $0.08 a second. A mode setting chooses between cost, speed (the default) and quality recipes, and an optional prompt upsampler rewrites prompts at no extra charge. Seven aspect ratios are offered, from 16:9 to 1:1.
The safety filter is off by default, though sexual content involving minors and self-harm instructions are always blocked. The API allows 250 requests a minute and bills on the finished clip's length.
Is P-Video-2-Pro worth it?
P-Video-2-Pro is the most transparent of the MiniMax H3 derivatives and among the least distinctive. Pruna publishes a full rate card, a 250-requests-a-minute limit and exact billing rules, which is more than most rivals manage. On Artificial Analysis it ranks 13th for text-to-video at 1128 Elo and 10th for image-to-video at 1119, both below H3 itself.
The pricing is flexible. At 768p a second costs $0.025 in cost mode, $0.035 in speed mode and $0.075 in quality mode, and Artificial Analysis's figure of $5.70 a minute matches the 1080p quality rate of $0.095 a second. That is above HeyGen Video 1's $2.40 a minute, which scores 19 points higher on text-to-video.
The catch: anything above 768p is an upscale, not a native render, and you pay $0.02 to $0.08 a second extra for it. The safety filter is off unless you turn it on, which is convenient for a studio and a liability for a consumer app that has not read the docs. You also cannot feed it your own audio; for that Pruna points to its separate P-Video 2.
Our take: P-Video-2-Pro is the H3 to put in a pipeline when you want to tune each request for price and know exactly what the bill will be. On quality alone, HeyGen Video 1 and MiniMax's own H3 are ahead for less.
Where it earns its place
A backend engineer at a stock-footage marketplace
Generate thousands of short preview clips overnight at the lowest cost
Batch requests in cost mode at 480p, $0.01 a second, within the 250-a-minute limit, then rerun the clips buyers pick in quality mode at 768p for $0.075 a second.
A two-person games studio making trailer drafts
Animate key art from a first and last frame, with sound
Send the start and end frames with a prompt describing action and audio; a 10-second 768p clip in speed mode costs $0.35. Upscale only the final cut to 1080p.
Before you commit
- Setup
- Create a Pruna API key, upload frames with POST /v1/files, then POST /v1/predictions with the p-video-2-pro model; add Try-Sync: true for a synchronous call.
- Integrations
- Pruna REST API; resold on Runware, Segmind, Wiro, Fliki, each::labs and Scenario; p-video-upscale for 1080p to 4K.
- Data and privacy
- Pruna's retention and training terms for prompts and frames were not stated on the model documentation we read.
- Cost at scale
- 100 minutes a month at 1080p in quality mode, $0.095 a second, is $570; the same in cost mode at 768p is $150.
- Limits
- 5 to 15 seconds; native 480p and 768p only; 24 fps; 250 requests a minute; no audio input; safety filter off by default.
What’s good
- Full public rate card with three cost modes per resolution
- 250 requests a minute and billing on finished clip length
- Widely resold, from Runware to Scenario
What’s not
- Below MiniMax H3 and HeyGen Video 1 on Artificial Analysis
- 1080p, 2K and 4K are upscales of 768p output
- Safety filter off by default; no audio input
Best for
Developers who want an H3-based video API with predictable per-request pricing and published limits.
P-Video-2-Pro alternatives
P-Video-2-Pro: common questions
- Is P-Video-2-Pro worth it?
- P-Video-2-Pro is a sensible, well-documented H3 for pipelines, and nothing about its output makes it the one to beat. We rate it 3.8 out of 5.
- How much does P-Video-2-Pro cost?
- Pruna API per output second (cost / speed / quality mode): 480p $0.01 / $0.02 / $0.04; 768p $0.025 / $0.035 / $0.075; 1080p $0.045 / $0.055 / $0.095; 2K $0.065 / $0.075 / $0.115; 4K $0.105 / $0.115 / $0.155.
- What are the downsides of P-Video-2-Pro?
- Below MiniMax H3 and HeyGen Video 1 on Artificial Analysis; 1080p, 2K and 4K are upscales of 768p output; safety filter off by default; no audio input.
- Who is P-Video-2-Pro best for?
- Developers who want an H3-based video API with predictable per-request pricing and published limits.
- How much does P-Video-2-Pro cost for a team or in production?
- 100 minutes a month at 1080p in quality mode, $0.095 a second, is $570; the same in cost mode at 768p is $150.
- Is P-Video-2-Pro safe for company data?
- Pruna's retention and training terms for prompts and frames were not stated on the model documentation we read.
- What are the best P-Video-2-Pro alternatives?
- The closest alternatives we have reviewed are HeyGen Video 1, MiniMax H3 (Hailuo 3.0), Pika H3 Flex, Boreal-H3 and LTX-2.5.
Answers checked Nov 21, 2026.