Pika H3 Flex

Pika H3 Flex is an open-weight video model that Pika released on October 29, 2026. Pika post-trained it on MiniMax H3, so it makes video with synchronized sound in six sampling steps and takes up to 12 references: images, video clips and audio. Pika says it runs faster than real time in most setups and up to 1.8× faster than H3 Max. Pika H3 Flex is coming to GoEnhance soon; until it goes live, the console below runs MiniMax H3 reference-to-video, the model Flex is built on.

Reunion in the Lantern Alley

Corgi on the Cliff Path

Fox in the Snow

Vintage Camera Ad

MiniMax H3

Reference images

Add 1–4 images: a person, a pet, a product or a place. In your prompt, call them @[image1], @[image2]… in this order.

Prompt

Example prompts

By uploading a file, you agree to GoEnhance's Terms of Use and Privacy Policy

What Pika H3 Flex Brings

Six Steps, Faster Than Real Time

Pika H3 Flex samples each clip in six steps. Pika reports it is up to 1.8× faster and up to 80% cheaper than H3 Max on average across text, image and reference tasks, and that a 15-second clip can take as little as 7 seconds. Pika notes these are model (DiT) times, not guaranteed end-to-end waits.

Up to 12 References in One Prompt

The reference mode accepts up to 9 images, 3 video clips and 3 audio clips, 12 files in total. Use them to lock characters, products, places, motion and voice; each clip can be 2–15 seconds.

Five Ways to Start a Shot

Besides references, Flex can work from text alone, from an opening frame, from an ending frame, or from both endpoint frames, which suits multi-shot stories, product films and motion graphics.

Draft at 768p, Finish Up to 4K

Run locally, Flex outputs 480p or 768p at 24 fps in 11 aspect ratios from 21:9 to 9:21, for 5 to 15 seconds. Pika's hosted API can take a draft to 1080p, 2K or 4K.

Sound You Can Take Apart

Every clip comes with a synchronized stereo soundtrack. Through Pika's API you can also export five separate stems, dialogue, score, sound effects, foley and ambience, to fix one part without touching the rest.

Pika H3 Flex vs MiniMax H3 vs FastH3 Omni

	Pika H3 Flex	MiniMax H3	FastH3 Omni
Made by	Pika, post-trained on MiniMax H3	MiniMax	FastVideo (Hao AI Lab), distilled from MiniMax H3
Sampling steps	6	50	8
Ways to start	Text, first frame, last frame, both frames, references	Text, image, references	References
References	Up to 12: 9 images, 3 videos, 3 audio	Up to 9 images, 1 video, 3 audio (on GoEnhance)	Tested up to 9 images plus 3 videos
Resolution	480p / 768p locally; 1080p–4K via Pika's API	720p / 1080p (on GoEnhance)	480p–768p
Audio	Stereo soundtrack; 5 separate stems via Pika's API	Synchronized sound	Synchronized stereo sound
Open weights	Yes, on Hugging Face	Yes	Yes

How to Make Reference Videos Like FastH3 Omni on GoEnhance

Upload Your References

Add up to 9 images for characters, products and places. You can also add one video clip for motion and up to 3 audio clips for voice or sound, each 2–15 seconds.

Tag Them in the Prompt

Refer to each reference by its tag, such as @[image1], and describe the action, the camera and what we hear. Put spoken lines in quotes.

Choose Length, Shape and Quality

Pick 5 to 15 seconds, an aspect ratio from 16:9 to 9:16, and 720p or 1080p.

Generate and Refine

MiniMax H3 renders the clip with native sound. Adjust the prompt or swap a reference and run it again.

Try MiniMax H3 Reference to Video

Pika H3 Flex FAQ

What is Pika H3 Flex?

Pika H3 Flex is a video-and-audio model from Pika, released on October 29, 2026. It is MiniMax H3 post-trained by Pika with fine-tuning and reward-based optimization, shipped as two adapters: one for text and first/last-frame generation and one for multimodal references.

Is Pika H3 Flex open source?

The weights are open: Pika published the adapters and the MiniMax H3 base components on Hugging Face, with inference code on GitHub. Check the model card for the terms before you use it in a product.

How is Pika H3 Flex different from MiniMax H3?

Flex keeps H3's video-with-sound design but samples in six steps instead of 50, accepts up to 12 references including 3 video clips, adds last-frame and first-and-last-frame starts, and through Pika's API adds 4K output and separate audio stems.

How fast is Pika H3 Flex?

Pika says it is faster than real time in almost every configuration, up to 1.8× faster than H3 Max on average, and can make a 15-second clip in as little as 7 seconds. Pika counts model time only, so real waits depend on hardware and queue.

Is Pika H3 Flex better than other video models?

Pika's own tests rank it #1 among open-source text-to-video models and first in its human study for anatomy and artistic style. These are Pika's results; we have not run an independent comparison yet.

What resolutions and lengths does it support?

Locally, 480p or 768p, 24 fps, 5 to 15 seconds, in 11 aspect ratios from 21:9 to 9:21. Pika's hosted API can raise a draft to 1080p, 2K or 4K.

Can I run Pika H3 Flex in ComfyUI?

Yes. ComfyUI has an official Pika H3 Flex tutorial with text-to-video, first/last-frame and reference workflows, and the repository includes BF16 LoRA files for Comfy. Hugging Face Diffusers is supported too. Pika tested 480p and 768p on one H100 80GB; the base components take about 210 GB of disk.

Can I use Pika H3 Flex on GoEnhance?

Not yet; we are adding it now. Until then, the console on this page runs MiniMax H3 reference-to-video, the model Flex is built on, and the four examples above were made with MiniMax H3. We will switch the console to Pika H3 Flex when it goes live.

Who is Pika H3 Flex for?

Pika aims it at creators who iterate a lot: multi-shot stories, product films, motion graphics and reference-guided animation, where you draft fast at low resolution and finish at high resolution.

Make Reference Videos with Sound

Turn your characters, products and places into a clip with native audio. Start with MiniMax H3 today; Pika H3 Flex is coming soon.