RU

Seedance 2.5 from ByteDance: 30-second videos and a filter against real faces

Published: 2026-11-11 · Author: AI Release · @ai_release1
Seedance 2.5 from ByteDance: 30-second videos and a filter against real faces

⚡ The gist in 5 seconds

🔍 What happened

ByteDance released Seedance 2.5 — a video generator that, as of July 31, 2026, can produce 30 seconds in a single pass. You can feed it up to 30 images, 10 videos and 10 audio files. A detailed walkthrough with prompts and step-by-step instructions was published by the SpeShu.AI blog on Habr.

The release's headline feature isn't the video length, but the wall at the input. Upload an ordinary photo and you get: "The input image may contain a real person". Via the API it's the same, just more technical — InputImageSensitiveContentDetected.PrivacyInformation. Rewriting the prompt is useless: the filter looks at the image itself, not the text.

The wall isn't only in the model itself. The same restrictions await in Dreamina, Adobe Firefly, Higgsfield, fal and Replicate — anywhere Seedance is connected via API. According to the BytePlus documentation, which LaoZhang breaks down, the model "does not support direct upload of images and videos with real faces".

The reason is the February scandal around Seedance 2.0. On February 13, 2026, a video spread showing Brad Pitt fighting Tom Cruise: both faces were generated without the actors' consent. The same day, the Motion Picture Association spoke out, Disney demanded the infringement stop, and on February 15 Paramount filed its claim. On February 16, ByteDance promised to strengthen protections, and in March US senators demanded the service be shut down.

Let's count the money using TeamDay's figures: a second in Seedance 2.5 via the API costs about $0.23 in 720p and $0.57 in 1080p. Seedance 2.0 was $0.15 in 720p. A five-second test is roughly a dollar. One more nuance: a VPN doesn't help — it doesn't process payment with a Russian card and doesn't touch the filter, which runs on the platform's side.

🧠 Our take

ByteDance locked down faces not out of a love of purity, but under pressure. The timeline is brutal: a viral video on February 13, letters from Disney and Paramount, a promise to "strengthen protections" three days later, senators' demands in March. Six months later — the filter is still there. It looks like, for ByteDance, this is cheaper than a second round of letters from Hollywood lawyers.

What does this change for the user? The model is no longer universal. Product videos, landscapes, illustrations, digital characters — yes, 30 seconds per pass is genuinely convenient here. A family photo, a portrait, a recognizable face — no, and no masks or noise will help. Well, they'll help you break the rules and Article 152.1 of the Russian Civil Code, but why bother.

As for alternatives, here's the picture. Kling 4.0 takes real faces and delivers 3–30 seconds, doesn't publish pricing, the site opens, but payment with a Russian card doesn't go through. MiniMax H3 — up to 15 seconds at ~$0.08 in 768p, also with a payment barrier. Gemini Omni Flash — 10 seconds extendable to 40 at ~$0.10 in 720p, but blocked by IP. Our assessment: the winners are those selling access and intermediation; the losers are those who need exactly one tool for everything.

⚠️ What we don't know

The prices come from TeamDay's retelling; we haven't seen an official ByteDance price list. The write-up was authored by the blog of a company that sells services around these models — keep that in mind. The dates of the scandal and the claims are taken from the article; we haven't verified the primary sources. The cost and availability of alternatives change faster than tables get updated: check payment and geo-restrictions yourself before buying.

💬 What to argue about

  • Is the real-face filter honest protection against deepfakes, or a convenient way to push competitors out of the video personalization market?
  • Is it worth paying $0.23 per second in 720p if the model rejects half of your input frames anyway?
  • Where's the boundary: the person on camera gave consent, but the platform still won't let them through. Who should decide here — the video's creator or ByteDance?

Prediction: by the end of 2026, ByteDance will add an official path to Seedance for your own face with consent verification — modeled on identity verification. Easy to check: watch the BytePlus release notes.

Disclaimer: we haven't seen an official ByteDance press release about the filter; the details come from the company's blog and a retelling of the BytePlus documentation.

Source: Habr, SpeShu.AI blog

🤖 AI summary
  • Date: 2026-11-11
  • Source: habr.com
  • Tags: Seedance, ByteDance, Generation, дипфейки, API, AI, видеогенерация, neural networks
  • Key facts:
    • Seedance 2.5 launched on July 31, 2026: 30 seconds of video per pass, with up to 30 images, 10 videos and 10 audio files as input.
    • The model won't accept real faces: on the web you get the rejection "The input image may contain a real person", via the API — the InputImageSensitiveContentDetected.PrivacyInformation error.
    • Via the API, a second costs about $0.23 in 720p and $0.57 in 1080p, so a five-second test is roughly a dollar.
    • ByteDance released Seedance 2.5 — a video generator that, as of July 31, 2026, can produce 30 seconds in a single pass.
    • You can feed it up to 30 images, 10 videos and 10 audio files.
    • A detailed walkthrough with prompts and step-by-step instructions was published by the SpeShu.AI blog on Habr .
    • The release's headline feature isn't the video length, but the wall at the input.
    • Upload an ordinary photo and you get: "The input image may contain a real person".
#Seedance#ByteDance#генерация#дипфейки#API#AI#видеогенерация#нейросети
← PreviousSynaptics Blocks DisplayLink Driver for Linux with Patent Claims

Source: habr.com · post in Telegram