Haiper

What is Haiper?

Haiper was a London-built generative video platform that turned text prompts and still images into short clips inside a browser app. It used its own video models rather than licensing another lab’s system.

You typed a scene description, uploaded a picture or chose a template, and the service rendered a few seconds of footage. Early clips lasted two to four seconds. Haiper 1.5, announced in July 2024, raised that to eight seconds and added a one-click upscaler to 1080p, according to VentureBeat. Haiper 2.0 followed on 28 October 2024, combining transformer and diffusion techniques for more lifelike motion and faster generation.

The company was founded in late 2021 by Yishu Miao and Ziyu Wang, Oxford machine learning PhDs who had both worked at DeepMind. It left stealth in March 2024 and raised a $13.8 million seed round led by Octopus Ventures. Sifted reported in March 2025 that Microsoft had hired both founders, and in June 2025 that NetMind had bought the video model. Guides say the consumer app went offline in early 2025, and in October 2026 haiper.ai showed only a title and no working product.

Key Features

Text-to-video and image-to-video generation drove Haiper’s appeal, backed by templates, an HD upscaler and clip extension that made short experiments quick to repeat.

• Text to video: Written prompts became animated clips describing a subject, action and setting. Haiper 1.5 produced eight-second results in SD or HD quality, double the two to four seconds of the first release.
• Image to video: Uploaded stills could be brought to life with motion and camera movement. Artists used this to animate illustrations, while marketers tested how a product photo might look as a short video.
• Video repainting: Parts of an existing clip could be restyled by prompt, such as changing colours or textures. This let users adjust a promising result instead of generating a new clip from the beginning.
• Templates: Haiper 2.0 introduced ready-made workflows for logo and product animations, plus playful effects such as character morphs and dance clips. Users uploaded an image and the template handled the prompt design.
• Upscaling and extension: A built-in upscaler raised clips to 1080p with one click, including videos uploaded from elsewhere. Earlier two- and four-second generations could also be extended to eight seconds, VentureBeat reported.
• Text to image and keyframes: A text-to-image model let users create a still before animating it, keeping the whole workflow in one place. The 2.0 release also listed keyframe conditioning for more control over how scenes began and ended.

Pros

Low-cost access to in-house video models was Haiper’s main draw, with a free tier for experiments and paid plans that started at roughly $10 a month.

• Generous free tier: Casual users could generate watermarked clips without paying anything. That made it easy to try AI video before committing to the paid tiers of Runway or similar tools.
• Research pedigree: Both founders came from DeepMind, and the team trained its own diffusion transformer models. That depth helped Haiper ship versions 1.0, 1.5 and 2.0 within about seven months.
• Simple interface: Prompting, image upload and templates suited first-time users well. Most people could produce a usable clip in a few minutes without learning camera terms or editing software.
• Fast iteration: Short clips rendered quickly, and version 2.0 promised faster generation than before. Creators could test several variations of an idea in one session instead of waiting for each render.
• Rapid user growth: The October 2024 launch release claimed 4.5 million users in under 200 days. A large community meant plenty of shared prompts and examples for newcomers to learn from.

Cons

Closure outweighs every other drawback, because the consumer service has gone offline and none of its features, plans or saved clips can be reached.

• Service closed: The consumer app stopped working in early 2025, according to several guides. Former users must move to active tools such as Runway, Kling AI or Luma Dream Machine.
• Short clips: Even after the 1.5 update, results topped out at about eight seconds. Longer clips of up to 30 seconds were announced as in development but never confirmed as released.
• Watermarks: Free and Explorer videos reportedly carried a Haiper watermark. Anyone wanting clean footage for client work had to pay for a higher tier or crop the output.
• Uncertain data fate: No public notice explained what happened to uploads and generated videos after the closure. Users who stored work only in the cloud may have lost it without warning.
• Paywalled quality tools: At the 1.5 launch, eight-second clips and the upscaler were limited to paid users. Free users therefore saw the shortest, lowest-resolution results, which undersold the model.

Who is Using Haiper?

No one can use Haiper now, but its millions of users were mostly social creators, marketers and artists testing short visual ideas cheaply.

• Social creators: People making short posts for TikTok or Instagram used it for eye-catching clips. Templates for dances and morphs were aimed squarely at this group.
• Marketers: Teams tested visual concepts before paying for a shoot. Logo and product animation templates gave them quick drafts for campaign pitches.
• Artists: Illustrators animated still artwork through the image-to-video mode. Repainting let them restyle a finished clip without starting the whole generation again.
• Hobbyists: Curious users explored AI video for fun on the free tier. The community gallery gave them prompts and examples to copy and adapt.

Pricing

Haiper sells no plans today, and the table lists tiers recorded by pricing trackers and press coverage (historical prices, checked October 2026).

VentureBeat quoted Pro at $24 a month billed annually, which explains why trackers disagree, and one tracker lists a $59.99 Unlimited tier.

Disclaimer: Treat these as historical figures only, since no Haiper plan can be purchased now.

What Makes Haiper Unique?

A small London team built its own diffusion transformer video models rather than wrapping another lab’s system, shipping three major versions in roughly seven months. Few start-ups of that size trained foundation video models at all.

Its ending was unusual too. Instead of a public shutdown, the founders joined Microsoft AI in 2025, and the model itself was reportedly sold to NetMind, leaving the consumer app to fade without a formal notice.

Compatibilities and Integrations

Haiper ran as a web app in desktop and mobile browsers, with MP4 downloads and a reported enterprise API but no consumer integrations.

• Platforms: A browser-based app that worked on desktop and mobile, with no native apps confirmed.
• Community: A Discord server carried release news, feedback and shared prompts.
• Export: Clips downloaded as MP4 files for editing in other software.
• API: Only enterprise API access was reported, and no public documentation survives.

Haiper Tutorials

Former Haiper users can rebuild the same prompt-to-clip workflow in an active video generator by following these steps.

1. Gather any Haiper clips you saved locally as MP4 files.
2. Sign up for a current tool such as Runway or Kling AI.
3. Write a prompt describing the subject, motion and camera angle.
4. Upload a reference image if you want image-to-video results.
5. Generate several takes and extend the strongest one.
6. Upscale the final clip before exporting it for social media.

How We Rated It

Haiper earns 3.0 out of 5, because strong ease of use and quick renders are cancelled out by the closed service and ended support.

Summary

Haiper suited creators who wanted quick, low-cost AI video experiments built on its own research-grade models. Its limitation is simple and final, because the consumer service has closed and its founders have moved to Microsoft. The eight-second clips, templates and 1080p upscaler are now matched or exceeded by active rivals. We see it today as a reference point in AI video history rather than a tool to choose. If you relied on it, test the same prompt in two or three active generators and keep the one that handles motion best.