You will generate a video faster than you can watch it. Pika H3 Flex breaks records. The new AI model needs only 6 steps to create videos with sound. H3 Flex is supposed to be much faster than its predecessor. 02.11.2026 15:42 Generating a 15-second video with sound in just 7 seconds? The company Pika claims that the new H3 Flex model can achieve exactly such a result. Artificial intelligence is supposed to create material faster than a person could watch it, while also allowing free changes to characters, scenery, and camera movements. There are, however, a few catches. The record 7 seconds does not mean the actual waiting time for a finished video, and the promised 4K resolution is not generated natively. Despite these limitations, Pika has shown a technology that could significantly speed up the creation of short video materials. 15 seconds of video in 7 seconds. Pika H3 Flex speeds up video generation. The company Pika presented the H3 Flex model, developed with creators of ads, animations, and short films in mind. The new tool is based on MiniMax H3, but it has been additionally trained and optimized for speed and the ability to edit generated materials. The declared performance is attracting the most interest. According to the manufacturer's official statement, H3 Flex can generate a 15-second video material with synchronized sound in just 7 seconds. The result was achieved during tests at 768p resolution. That is more than twice as fast as the time needed to watch the finished recording. The company emphasizes that in comparable tasks its model is on average about 1.8 times faster than H3Max. At the same time, it is supposed to maintain image quality sufficient for use in professional projects. The secret lies in reducing the number of computational steps needed to generate the video. The base MiniMax H3 uses 50 denoising steps, during which the model gradually transforms the initial noise into image and sound. H3 Flex performs a similar process in just 6 steps. Such a large reduction in the number of operations unfortunately does not translate into an eightfold increase in overall performance. It does, however, significantly shorten the most demanding part of generation, which will matter when repeatedly refining the same materials. 7 seconds does not mean a finished video on screen. Pika clearly states that the advertised 7 seconds applies only to the work of the transformer responsible for video generation. It does not include the entire request handling process, input data preparation, decoding, audio processing, or transferring the finished file to the user. Waiting may therefore take longer than the recording itself. Especially when the user provides reference materials, selects a higher resolution, or uses additional audio editing functions. The manufacturer does not guarantee record results with every prompt. The second limitation concerns video length. H3 Flex allows creating materials lasting from 5 to 15 seconds, at 24 frames per second. It is therefore not a tool that will generate a several-minute film or a full narrative production with a single prompt. The model does offer output resolution up to 4K, but it does not create the image at that quality from the start. It natively supports only 480p and 768p, while 1080p, 2K, and 4K materials are created by upscaling the finished recording. Increasing the number of pixels does not, of course, guarantee recovering details that were not originally generated. Higher resolution can improve the look of the material, but it is not equivalent to fully fledged image generation in native 4K. AI lets you change the actor and scenery without filming from scratch. H3 Flex, however, is not merely a faster version of popular video generators. The creators placed great emphasis on the ability to control finished shots and use reference materials. The model can accept up to 12 files at once, including a maximum of 9 images, 3 videos, and 3 audio materials. The user can provide a product photo, a scene sketch, a character appearance template, a camera movement recording, or a voice sample. The algorithm will then use them to build new material. This makes it possible to prepare an ad in which the product remains recognizable but the surroundings, lighting, and camera movement change. It is also possible to replace a character or prop in an existing recording while preserving the original choreography. Pika additionally allows creating sequences composed of several shots, containing editing cuts, dialogue, music, and sound effects. The finished material can be exported with 5 separate audio tracks covering dialogue, music, special effects, sounds of performed actions, and ambient sounds. This solution is especially useful in editing. Instead of regenerating the entire video, a creator can mute the music, change the dialogue volume balance, or improve selected sound effects in an external program. On Spider's Web, we have already described the possibilities of generating videos with dialogue and music by Google Veo 3. H3 Flex develops a similar concept but focuses more strongly on the speed of successive attempts and the ability to modify specific elements of a recording. The new model has open weights. However, not everything can be run locally. An interesting part of the launch is making the model weights and the code allowing it to run on one's own hardware available. Pika published a repository on GitHub and files on the Hugging Face platform, enabling developers to experiment with the technology. This does not mean, however, that every computer user will be able to reproduce the declared generation speed. The company stipulates that local performance depends on hardware and configuration, and Pika's optimized server infrastructure has not been made fully available. There are also functional limitations. The local version supports native generation at 480p and 768p, but upscaling to 4K, exporting separate audio tracks, and some additional tools remain features of the manufacturer's online service. Licensing terms are also not insignificant here. The code for running the model was made available under the Apache 2.0 license, while its weights are subject to a separate MiniMax H3 community license. Describing the model as completely open without any restrictions would therefore be an oversimplification. How much does it cost to create videos in Pika H3 Flex? The new model was made available through Pika API Club, and its introduction to the main Pika tool was announced for November 2. The manufacturer offers billing based on the length of the generated material, and rates depend on the selected resolution and method of generation. The regular price of 1 second of video at 768p is 0.04 dollars. This means that a 15-second recording costs 0.60 dollars, that is, still without additional options and possible costs of processing reference materials. As part of the launch promotion, the rate was reduced by half. For 4K output video, the regular cost rises to 0.16 dollars per second, or 2.40 dollars for a maximum 15-second material. The model also supports a more expensive mode using reference video recordings. The company emphasizes that H3 Flex is intended primarily for rapid experimentation. A creator is supposed to be able to prepare several versions of a scene, choose the best one, and only later decide on more expensive upscaling and final editing. This is the direction in which competing tools are also developing. Sora from OpenAI has already shown how easily generated videos can go directly to social media, while Adobe is developing its own environment of AI tools for creators. H3 Flex does not, of course, solve all the problems of generative video. According to tests published by the manufacturer, the model still has difficulty credibly reproducing physics, complex scenes, and maintaining the coherence of multi-threaded stories. It does show, however, that another important area of competition is becoming not only image quality but also the time needed to receive the next version of a recording. *Intro graphic source: AI / Pika

Pika drops H3 Flex, an open video model generating audio 1.8x faster. Pika releases an open-weight video model post-trained on MiniMax H3 that generates 15-second clips with audio in 6 denoising steps. · 5 hrs ago Read 5 min * Pika released Pika H3 Flex, an open-weight video+audio model post-trained on MiniMax H3. * Generates 15-second clips with soundtrack in as little as 7 seconds using a custom 6-step sampling schedule. * Ranks #7 on DesignArena t2v with Elo 1293, the top open-weight entry and above H3Max and Seedance 2.5. * Supports up to 12 reference inputs (9 images, 3 videos, 3 audio) plus 5 exportable audio stems via API. * Weights on Hugging Face, inference code on GitHub, with ComfyUI and diffusers support day one. * API pricing starts at $0.01/sec at launch; local inference needs an H100 80GB and ~215GB of downloads. Pika has released Pika H3 Flex, an open-weight video model that generates synchronized video and stereo audio with six denoising steps. The model is post-trained from MiniMax H3, with adapter weights on Hugging Face, inference code in the GitHub repository, and integrations for ComfyUI and diffusers. At the time of Pika's announcement, H3 Flex ranked seventh on DesignArena's text-to-video leaderboard and led its open-weight entries. Pika also reports that its hosted API can produce a 15-second clip with audio in as little as 7 seconds, meaning the output runs longer than the generation process. Two adapters cover five workflows. Pika H3 Flex keeps MiniMax H3 as its base and adds two task-specific adapters. FL2VA handles prompts and endpoint frames, while Ref2VA conditions generation on image, video, and audio references. | Adapter | Mode | Input | | FL2VA | t2va | Text prompt | | FL2VA | i2va | Text and first frame | | FL2VA | l2va | Text and last frame | | FL2VA | fl2va | Text, first frame, and last frame | | Ref2VA | ref2va | Text plus multimodal references | The model accepts up to 12 reference inputs, including sketch, voice sample, image, sound, video, and moodboard. That mix supports character, scene, motion, and soundtrack references within one request. | Specification | Support | | Local resolution | 480p and 768p | | Hosted resolution | 1080p, 2K, and 4K through draft-to-video upscaling | | Clip length | 5 to 15 seconds | | Frame rate | 24 fps | | Audio | 32 kHz stereo | | Aspect ratios | 11 options, from 21:9 to 9:21 | | Prompt length | Up to 8,000 characters | Pika says the hosted upscaling stage preserves the draft's motion and audio while increasing resolution. Developers who need direct local inference are limited to the two native resolutions in the open release. A custom six-step sampler. H3 Flex uses six denoising model evaluations across seven timestep boundaries: 1000/750/500/250/100/40/0. A denoising evaluation is one pass that converts a noisier intermediate representation into a cleaner one, so reducing the pass count directly lowers generation latency. The adapters also use video shift 12, audio shift 3, and fresh noise at intermediate stages. The repository warns that a generic six-step Euler sampler follows a different process, because the released schedule is tied to the adapter training. Pika attributes the faster-than-real-time result to this schedule. Its matched comparison reports roughly 1.8 times faster generation than MiniMax H3Max. Actual local latency will vary with GPU, resolution, clip length, and CPU offloading. Audio arrives ready to mix. Local inference produces a mixed stereo track alongside the video. The hosted API can also return five separate .wav stems: Dialogue, Score, SFX, Foley, and Ambience. Separate stems let developers send generated material into a nonlinear video editor or digital audio workstation without first attempting source separation. Pika charges an additional $0.001 per generated second for stem export. Local inference needs substantial hardware. The README recommends Python 3.12 and an NVIDIA GPU, and Pika tested representative requests on a single H100 with 80 GB of memory. Downloading both tasks requires about 215.5 GB, comprising roughly 210 GB of base files and 5.5 GB of adapters. * Each request runs on one GPU. Single-clip sharding across multiple GPUs is unavailable. * The -cpu-offload option reduces VRAM use by moving data through host memory, increasing RAM demand and transfer time. * Output lengths follow a 17n+5 frame grid at 24 fps. * A nominal 5-second request produces 124 frames, or about 5.17 seconds. * A nominal 15-second request produces 362 frames, or about 15.08 seconds. The repository's command-line interface exposes the task, model revision, resolution, aspect ratio, duration, and seed: pika-h3-pro -task t2va -prompt-file examples/prompts/t2va.txt \ -model-repo "$MODEL_REPO" -revision "$MODEL_REVISION" \ -resolution 480p -aspect 16:9 -duration 5 -seed 42 \ -output outputs/sample.mp4 Hosted pricing trades hardware for resolution. Launch pricing on the Pika API starts at $0.01 per output-video second for 480p and reaches $0.08 per second for 4K. Pika describes those rates as roughly half its regular pricing. | Charge | Launch rate | | 480p output video | $0.01 per second | | 4K output video | $0.08 per second | | Audio stems | Additional $0.001 per second | | Reference video | Input and output seconds billed at the generation rate | | Image or audio references | No reference-input charge | At the disclosed endpoint rates, 15 billed seconds cost $0.15 at 480p or $1.20 at 4K before stems and reference-video charges. Pika also announced a launch promotion of 15 free daily generations for API Club members during the first two weeks. The leaderboard lead is narrow. DesignArena summarizes head-to-head user preferences as Elo ratings. H3 Flex recorded an Elo score of 1293 and ranked seventh overall when Pika announced the release. | Model | Elo | Position relative to H3 Flex | | Gemini Omni 1.1 Flash | 1360 | Ahead | | Wan 3.0 | 1324 | Ahead | | Pika H3 Flex | 1293 | Seventh overall; highest open-weight entry | | MiniMax H3 | 1277 | Below | | H3Max | 1277 | Below | | Seedance 2.5 | 1256 | Below | The 16-point lead over MiniMax H3 indicates that Pika's post-training improved the base model's standing on this evaluation. Elo scores remain relative and change as votes accumulate, so the figures capture the board at announcement time. A reusable adapter strategy. Pika's release shows how a team can specialize an existing video model, outperform its base on a preference leaderboard, and distribute the resulting adapters with inference code. The six-step schedule is especially relevant to developers because it connects post-training choices directly to lower inference latency. The release also offers three deployment paths: local generation at 480p or 768p, hosted generation at higher resolutions, and node-based workflows through the live ComfyUI integration. The adapter weights are available on Hugging Face. The MiniMax H3 Community License governs use of the released adapters, so teams should review its terms before commercial deployment. Local operation also carries a high storage and GPU cost, making the hosted API the more accessible route for developers without server-class NVIDIA hardware.