Six new Flow tools, and no credit costs or plan terms in Google's launch post. A secondary write-up flags two checks worth knowing: caption accuracy, and text inside generated covers.
What to check first ↓
Kling 4.0 doubles a single AI video generation from 15 seconds to 30.
Kuaishou's video unit announced the model on September 28: more than a dozen reference inputs in text, image and video, and up to 10 keyframes to fix how a segment looks. A lite version is open to annual…
ElevenLabs put a performer inside the voice model.
v4 reads a script like an actor: stacked direction tags, scene-context dialogue, a voice cloned from ten seconds, 90+ languages up from 70, and Turbo answering in about 150 milliseconds.
The catch, and the full breakdown here:
ElevenLabs put a performer inside the voice model.
Dialogue was the last unperformed layer of a generated scene. You could generate the shot, the face, the light and the score, and then the room spoke like a…
https://t.co/ko8kJNLanr
#ElevenLabs#AIvideo#VoiceAI#TTS#AI
ElevenLabs put a performer inside the voice model.
v4 reads a script like an actor: stacked direction tags, scene-context dialogue, a voice cloned from ten seconds, 90+ languages up from 70, and Turbo answering in about 150 milliseconds.
The catch, and the full breakdown here:
ElevenLabs' v4 clones a voice from 10 seconds of audio and stacks direction tags into a scene; Turbo answers in ~150ms. The company says it ranks #1 on Artificial Analysis's voice arena. We haven't tested it.
Youku commissioned a face to put a banned hit back online.
2016's Love O2O streams again as A Smile Is All It Takes - the original lead's face swapped out in every scene. The new lead, Wang Zhuyue, says she authorised it herself.
How it works ↓
Two of the most capable AI models on sale can't make a single frame of video.
Claude Fable 5.1 finishes things. GPT-6 Astra sees space. Their benchmarks split where the split-brain research drew its line.
Why a 2026 video pipeline needs both halves ↓
Two of the most capable AI models on sale can't make a single frame of video.
Claude Fable 5.1 finishes things. GPT-6 Astra sees space. Their benchmarks split where the split-brain research drew its line.
Why a 2026 video pipeline needs both halves ↓
A $34 robe. An $8 garage-sale puppet. One cola can, shot thirty-one times.
Based Boomerstein wrote down how Ep.69 was made, scene by scene, and the budget is the argument.
The diary ↓