
ByteDance's World Model Re-Shoots Movies, Sora Returns? & Krea 3 Is Here!
Theoretically Media15 October 2026Watch on YouTube
Topics
Read next
Chinees AI-bedrijf Kimi brengt concurrerende open model K3 uit
Kimi lanceert K3, een open-weight model met 2,8 triljoen parameters dat zich volgens benchmarks voordoet als competitief met Anthropic Fable 5 en OpenAI's GPT 5.6 Sol, tegen hogere prijsstelling dan voorgangers.
Chinees world model Orca evenaart roboticasystemen zonder labeldata
Het Beijing Academy of Artificial Intelligence heeft Orca gelanceerd, een world model getraind op 125.000 uur video zonder actionlabels, dat gelijk presteert aan gespecialiseerde roboticasystemen.
ByteDance traint massief AI-model ter concurrering van Anthropic
TikTok-eigenaar ByteDance traint een model met 10 biljoen parameters.
Moonshot bereidt Kimi 3 voor met miljarden parameters
Moonshot werkt aan Kimi K3, naar verwachting het grootste open-source AI-model uit China met 2-3 biljoen parameters.
Kling AI brengt versie 4.0 uit met stereo-audio en meertalige lipsync
Kling AI's nieuwe videogenerator ondersteunt clips tot 30 seconden, tien keyframe-afbeeldingen en prompts tot 8.000 tokens, maar onderscheidt zich nauwelijks van concurrenten.
AI-videomarkt herstelt zich na falstart van Sora
AI-productiebedrijven vestigen zich rond Hollywood en gebruiken real-time achtergronden om filmkosten te verlagen; Netflix gebruikt AI in 300 van zijn 1.000 titels.
Description from the channel
ByteDance's new world model, Sora returns?, Krea 3 is here, OmniHuman 2.0 is back, Pico Space Pro, World Gacha, and LTX-2.5's new post tools. ByteDance's unreleased world model took a scene from Stephen Chow's King of Comedy and re-shot it from new camera angles off one drawn camera path, at Sunday's Douyin gala. Then: OmniHuman 2.0 (API only, not in the US yet), Krea 3's beta, a very fast AI image editor, the report that Sora is coming back via a Jeffrey Katzenberg and Bill Peebles startup, and a free browser tool that builds cities you can wander and export to Blender. Plus LTX-2.5's new open-source post tools. WHAT'S IN THIS ONE - OmniHuman 2.0: ByteDance's image-plus-audio avatar model, out today through BytePlus. API only, and not available in the US, hence my region lock. Clips up to 5 minutes, about 40% cheaper, 720p. A little south of Seedance 2.5; handy for patchwork lines. - The world model: shown Sunday at Douyin's Wonderful Night gala. No name, no date, phone footage of an LED wall. Mouse-and-keyboard control, prompt an object in and the character uses it, and the King of Comedy demo: a dozen camera positions from one drawn path, with Stephen Chow's permission. Bloomberg says ByteDance is building a real-time world model for Pico and Douyin. - Pico Space Pro: ByteDance's Vision Pro rival, delayed to Q4 2026. No official price; leaks say roughly $2,400–2,600 vs the Vision Pro's $3,699. - Krea 3 (beta today, final release next week): editor first, generator second. Most 2K edits in under 8 seconds. Draw a box, say what changes, the rest stays put. Tested: GPT Image grain removal, Flamethrower Girl's outfit change (no lecture), Business Shark, and a first/last-frame diner shot. - Sora is coming back? The Information (Oct 12): OpenAI is in talks to give a stealth startup deep access to the shelved Sora model for equity. Co-founders Jeffrey Katzenberg, Bill Peebles and Sujay Jaswa (ex-Dropbox CFO). Aimed at film studios. Nothing signed, no timeline. - The mic: an OpenAI engineer was arrested Oct 9 for allegedly hiding a recorder in an executive conference room. Per the complaint, the how-to prompts came from ChatGPT. Allegedly. - World Gacha: a free browser tool by Sharp (@sharp_maker) that generates fictional towns from random numbers and an erosion sim, with street level and 3D export. Free for commercial use; English toggle added after I recorded. SPONSOR This video is sponsored by LTX. LTX-2.5's new post tools: Alpha Gen (beta) hands you a frame-accurate alpha matte with your shot; SDR to HDR expands 8-bit footage to 16-bit EXR scene-linear; Native HDR keeps 16-bit footage intact while you edit; Native Resolution plus Refine renders at 4K or 8K and rebuilds the detail. Open weights, ComfyUI workflows, or LTX Desktop with no Comfy setup, on your machine. ► LTX-2.5 weights: https://huggingface.co/Lightricks/LTX-2.5 ► Post LoRAs: https://huggingface.co/collections/Lightricks/ltx-25-creative-lab ► ComfyUI workflows: https://github.com/Lightricks/ComfyUI-LTXVideo/tree/master/example_workflows/2.5 ► LTX Desktop: https://ltx.io/ltx-desktop ► Tag @ltx_io if you make something. LINKS OmniHuman 2.0: https://www.byteplus.com/en/product/OmniHuman The gala demo: https://x.com/TheoMediaAI/status/2109643192920404199?s=20 Pico Space Pro: https://roadtovr.com/pico-vision-pro-competitor-delayed-cancelled-event/ Krea 3: https://www.krea.ai/blog/introducing-krea-3 The Information on Sora: https://www.theinformation.com/newsletters/ai-agenda/openais-plan-revive-sora The arrest: https://techcrunch.com/2026/10/13/fbi-arrests-openai-employee-accused-of-planting-hidden-recording-device-in-conference-room/ World Gacha: https://tools.oyasumi-gamedev.com/worldgacha/ Newsletter: https://theoreticallymedia.com/newsletter/ CHAPTERS 00:00 ByteDance's world model, Krea 3 and the return of Sora 00:16 OmniHuman 2.0: ByteDance's lip-sync avatar is back 00:57 The teeth, fake Karl Urban, OmniHuman 2.0 vs Seedance 02:42 First look: ByteDance's unnamed world model 03:32 New camera angles inside a real movie: King of Comedy 03:59 The UI: 3D controls, the portal demo, camera paths 04:49 Why it matters: video, 3D and world models converge (2027) 05:16 Pico Space Pro: ByteDance's Vision Pro? 05:37 Krea 3: an editor more than a generator 06:48 Grain removal, and Flamethrower Girl's outfit change 07:54 Precise edit: Business Shark Flamethrower Girl 08:08 First frame / last frame: the diner and the pie 09:03 Where Krea 3 falls short (plus a Planet Hell WIP) 09:57 Sora is coming back (but first, a word from LTX) 10:11 LTX-2.5 post tools: Alpha Gen, SDR to HDR, 4K/8K 12:45 Sora: the plug, Peebles, and The Information's report 13:39 The Sora Stealth startup 14:22 Will We Get Sora Again? 14:49 World Gacha: free cities you can wander and export to Blender #ByteDance #Sora #Krea3 #OmniHuman #LTX #AIVideo #AINews