On October 29 Pika released Pika H3 Flex, a joint audio-video generator post-trained on MiniMax H3, with weights on Hugging Face, code on GitHub, and ComfyUI and Diffusers support. Pika says it distilled sampling from 50 denoising steps to six, averages about 1.8x faster DiT denoising than Fal H3 Max at 768p, and can produce a 15-second clip with soundtrack in as little as about 7 seconds; the hosted API adds 1080p to 4K upscaling and five exportable audio stems (dialogue, score, SFX, foley, ambience). All speed and quality figures are vendor-reported.
Key Takeaways
- ✓Pricing (vendor figures): text/image-to-video from $0.01 per output second, reference-to-video from $0.03/s; Pika's comparison table lists MiniMax H3 at $0.08/s and Seedance 2.5 at $0.103/s
- ✓Speed (vendor figures): six-step sampling, about 1.8x faster DiT denoising than Fal H3 Max at 768p on average and about 2x for 10–15 s text/image-to-video; 1.60x faster end-to-end 5 s 768p generation on eight B300 GPUs
- ✓Ranking: Pika's product page says it scores 1293 and ranks #7 on DesignArena's text-to-video board, the highest open-weight model there
- ✓Release form: the Hugging Face repo is tagged as an adapter on MiniMax-H3 under an 'other' license; local inference natively supports 480p/768p with mixed stereo audio, while 4K upscaling and audio stems are hosted-API only
- ✓Accepts up to 12 reference inputs (image, audio, video, sketch and more) and supports style transfer, motion transfer, character swap and targeted edits
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
AI video is usable in 2026, but for professional creators the bottleneck is iteration speed and cost: a 15-second clip with sound from a leading closed model often takes longer than the clip itself and is billed per second. High-quality open-weight audio-video models are also rare, so teams struggle to put generation inside their own pipelines. Pika did not train a new base; it post-trained MiniMax H3 and focused on speed and control.
Architecture and mechanism
Per Pika's technical blog, training had three stages: supervised fine-tuning on an internally curated audio-video set, distribution matching distillation from 50 sampling steps down to six, and post-distillation reinforcement learning plus preference optimization for aesthetics and instruction following. The hosted API adds a 1B-parameter, single-step pixel-space upsampler (1080p/2K/4K) and a prompted audio separator that exports five stems: dialogue, score, SFX, foley and ambience. The model takes up to 12 reference inputs.
Benchmarks
Only vendor numbers exist so far. In Pika's automated evaluation it beat MiniMax H3, Fal H3 Max and HeyGen Video on most prompts across five visual and audio quality metrics, scoring above H3 Max on video aesthetics for 91% of prompts; in blind human preference tests it led on anatomy, complex composition and audio sync. The product page says it scores 1293 and ranks #7 on DesignArena's text-to-video board, the top open-weight entry, still below closed models such as Gemini Omni 1.1 Flash (1360). No independent third-party evaluation yet.
Getting started
Pull the weights from Hugging Face and follow the GitHub README to run 480p/768p locally via Diffusers or ComfyUI. Hugging Face tags it as an adapter on MiniMax-H3 with an 'other' license, so read the terms and confirm base-model access before commercial use. For 4K or audio stems use the Pika API (dev.pika.art), from $0.01 per second. Pika's suggested workflow is to draft at 768p and upscale only the final take.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.