Model

Pika's Enact lets creators direct AI video mid-scene, with answers in under a second

Enact, posted November 18, lets a creator drag an AI actor mid-scene, talk to the cast by name and redo the camera after locking a performance. Pika open-sourced the model underneath and kept the directing layer closed, a bet that value in AI video is moving from model quality to workflow.

Richard TangRichard Tangthree-minute read
Share
Pika co-founders Chenlin Meng and Demi Guo, photographed in a pale studio against a curtain
Pika co-founders Chenlin Meng and Demi Guo. The company's Enact preview, posted November 18, lets a creator direct actors and camera while a scene is still playing.

Pika~$825M — Pika, private, latest valuation ~$825M posted Enact on November 18: a research preview where you direct a scene instead of describing it and waiting for a render 12. Drag a character with the mouse and the scene keeps playing around the move; give a spoken note to one actor by name; lock the performance, then redo the camera 23. Pika, an AI video company backed by the venture firm Lightspeed 4, published the underlying model's weights for anyone to download in late October and is keeping the directing layer closed.

Streaming speed is what makes direction possible

Enact runs on H3 Flex, Pika's post-trained version of MiniMax's H3 video model, a base built elsewhere and then tuned by Pika's researchers for speed and control 14. They cut generation to six denoising steps, the passes a model runs to turn noise into a picture, from the original's 50 14. A model that finishes a clip in seconds can absorb a note while the scene is still playing; one that takes minutes cannot. That speed is the whole trick.

Enact answers input in under a second 2. It streams continuously rather than delivering finished chunks, and it was trained on synthetic scenes that a slower, higher-quality model filmed photoreal 23. An agent layer keeps each note on the right character, so a direction to Sarah changes Sarah, not Darren 3. Chief executive Demi Guo names the cost honestly: the responsive model is slightly lower in image quality than the offline one 3.

Frame from Pika's Enact preview: a close-up of a woman in glasses, subtitled "Yeah, we're ready" over the name tag "ALICE"
A frame from Pika's Enact preview, posted November 18: the actor's name tag and a spoken line sit over the generated scene, the layer a creator talks to while it plays.Pika via X

The model is free, the workflow is the product

One layer up sits the money. Pika published H3 Flex's−8.03% — Flex, down 8.03 percent today weights on Hugging Face and GitHub when the model launched on October 29 4, and its platform already sells rivals' models, Kling 4.0 and Seedance 2.5 among them, charged by the credit 15. The open release stops short of Pika's hosted upsampler and serving stack 5, and nothing in the Enact announcement says the directing layer is open either 6.

Quality, meanwhile, is bunching. On the leaderboard Pika itself publishes, H3 Flex sits two places above the MiniMax base it came from, and it is the highest-ranked model with open weights on that board 5.

The video quality board is bunched, and Pika sits seventh

Quality no longer separates the leaders, so Pika's edge has to be the workflow

05001,0001,500Gemini Omni 1.1 Flash1,360Gemini Omni Flash1,343HeyGen Video 1.0 (1080p)1,334Wan 3.01,324P-Video-2-Pro1,310FLUX 3 Max1,297Pika H3 Flex1,293open weights; its base MiniMax H3 sits two places belowFLUX 3 Video1,286MiniMax H31,277Seedance 2.51,256
Data
Value
Gemini Omni 1.1 Flash1,360
Gemini Omni Flash1,343
HeyGen Video 1.0 (1080p)1,334
Wan 3.01,324
P-Video-2-Pro1,310
FLUX 3 Max1,297
Pika H3 Flex1,293
FLUX 3 Video1,286
MiniMax H31,277
Seedance 2.51,256
DesignArena text-to-video quality scores as published on Pika's own H3 Flex page, November 2026; higher is better.5

The rivals hold the adjacent ground. Runway, another AI video company, sells editing first: Aleph 2.0 in Edit Studio previews a change as an image before committing a generation 7. Luma, a third, pitches Ray 3.2 as directing video the way you would direct a shoot 8. Enact goes further, because the direction lands while the scene is still unfolding 13.

Shipping, not the demo, decides what Pika is worth

Today Enact is a demo on request, with no pricing, release date or public access 6. Pika is private, so what moves here is private-market positioning rather than a listed price.

The preview still has to prove latency on real inputs, how long a scene can run and how many characters can take a note at once 6. If the directing layer ships into the paid platform, Pika would be the only one of the three taking direction mid-performance rather than between renders 178.

The model layer is now free to download. The directing layer is not, and turning it from a demo into a tool working directors finish scenes in is the work that decides what Pika is worth.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading FLEX

Reader comments

0 comments

    Model

    HeyGen sells Avatar 6 while Tavus's Griffin scores 3.83 against a human's 3.92 and still has no price

    HeyGen put Avatar 6 on general availability on November 17 with a public API and first place in New Measure's blind Avatar Arena; Tavus's Griffin, which Tavus says is the highest score ever recorded for an AI on NVIDIA's face-to-face benchmark, is still a preview with no price. Revenue accrues to HeyGen today because Avatar 6 is the model on sale; Tavus's best model is not.

    Model

    Anthropic falls 11% toward its $83 IPO price as Microsoft undercuts Claude

    Anthropic fell about 11% Friday morning, back toward the $83 IPO price, as Microsoft's price attack on Claude met the open market. Microsoft holds Anthropic's compute contract, its shares and a cheaper rival product; a leveraged income ETF on the two-day-old stock listed the same morning.

    Model

    Ramp's invoices show Anthropic held 58.7% of US business AI spend vs OpenAI's 37.2% before its record IPO

    Ramp's index of 85,000 US business bill-payers is the first independent check on the record $81.3 billion IPO: the enterprise lead over OpenAI is real, and so is the falling price of what Anthropic sells.

    Model

    Google's Gemini 4 Neon took the coding lead the day before Anthropic's $1.5 trillion debut

    Google's mid-tier model debuted at No. 1 on Artificial Analysis's coding-agent index at $6.86 a task against Claude Opus 5.5's $12.61, a day before Anthropic's Nasdaq debut. A $1.5 trillion valuation built on coding-led growth now meets a measured, cheaper rival, and the exposure sits with whoever paid $83.

    Model

    ElevenLabs, valued at $22 billion, puts its voice agents inside NICE, Genesys, Amazon Connect and Five9

    Salesforce and Five9 have put more than $3 billion into buying customer-service AI this month. ElevenLabs, valued at $22 billion after a September share sale, embedded its agents natively into Five9's platform and three rivals instead, and the evidence points to pricing power sitting with the channel, not the model maker.

    Model

    OpenAI's agent bill now runs at twice its researchers' cash pay

    OpenAI's own report prices agent use at twice its researchers' cash pay while internal deployment is 2 percent of research compute, and the number lands the week Anthropic lists and OpenAI settles for half its raise. The bill stacks inference on the chip budget, and holders pay it.

    Subscribe to TheInference for free