Uploaded July 2026 | Updated November 2026, 2 days ago
► Full written breakdown: theoreticallymedia.com/blog/seedream-5-pro-layers-seedance-2-5-leaks
ByteDance's new Seedream 5.0 Pro is an AI image model with a genuinely new trick: it can generate in actual layers — plus up to 10 reference images and native 4K. It's also kind of a mess. I stress-test it (blobby faces and all), dig into the layer editor on Dreamina, and call whether it's a real Nano Banana / GPT Image rival.
Plus: Seedance 2.5 slips again but the new leaked clips look incredible, ByteDance drops an official Seedance short film, LingBot-World 2.0 puts a real-time open-source world model on Hugging Face, and Anam's CARA-4 gives AI voice agents a face.
Thanks to Teachable for sponsoring today's video! Sign up for my AI Academy live session: teachable.com/events/build-your-own-ai-film-studio-the-production-company-on-your-desktop?utm_source=TheoreticallyMedia&utm_medium=partnership&utm_campaign=aiacademy&utm_content=youtube
LINKS:
► ByteDance's Seedance short film (full, with audio): https://x.com/BytePlusGlobal/status/2077321849806234080?s=20
► LingBot-World 2.0 (GitHub): github.com/robbyant/lingbot-world
► LingBot-World 2.0 online demo: https://www.reactor.inc/
► Seedream 5.0 Pro on Dreamina: dreamina.capcut.com
► Anam CARA-4: anam.ai/cara4
NEWSLETTER: theoreticallymedia.com/newsletter
Chapters
00:00 Intro: Still Waiting on Seedance 2.5
00:22 Seedream 5.0 Pro: ByteDance's New "Thinking" Image Model
00:41 The Feature Stack: Native Text, 10 Refs & Layer Generation
01:32 First Test: The Man in the Blue Business Suit
02:07 Punch-In Problems: Return of the Blobby Faces
02:48 The Trademark Logo Problem
03:06 The UK Living Room Detail Test
03:53 Reference Tests: The Oni Stuntmen
04:42 Flamethrower Girl: Where References Get Weak
05:20 Tuned Prompts: Technically Fine, Kind of Bland
06:03 Where Seedream Shines: Close-Ups, Genres & Environments
06:49 Wong Kar-wai by Way of Cyberpunk
07:07 The Dreamina Layers Editor: Marketing vs. Reality
08:04 Adding a Layer: The Cyberpunk Cab Problem
08:38 Remove Background's Flipped Masking
08:58 Compositing Flamethrower Girl (A Failed Relight)
10:07 Seedream 5.0 Pro: The Verdict
10:51 Seedance 2.5 Delayed Again (July 20?)
11:04 Leaked Clip 1: The Train Station Reunion
12:29 Leaked Clip 2: Yes, Chef
13:20 Leaked Clip 3: "We Made It"
14:06 ByteDance's Official Seedance Short Film
14:45 Physics Nitpicks & That GPT-Image Grain
15:21 Sponsor: My Teachable AI Academy Live Session
16:48 Real-Time World Models: LingBot-World 2.0
17:25 Text-Triggered Events & the "Brain/Cerebellum" System
17:47 Hands-On: Siege of the Keep
18:47 Where the World Starts to Melt
19:20 The Respawn Trick & Holding Coherence
20:19 World Models as Virtual Sets (+ Where to Get It)
20:54 Anam CARA-4: Real-Time AI Avatars
21:20 AI Voices Are About to Get Faces
21:52 Wrap-Up
► Full written breakdown: theoreticallymedia.com/blog/seedream-5-pro-layers-seedance-2-5-leaks
ByteDance's new Seedream 5.0 Pro is an AI image model with a genuinely new trick: it can generate in actual layers — plus up to 10 reference images and native 4K. It's also kind of a mess. I stress-test it (blobby faces and all), dig into the layer editor on Dreamina, and call whether it's a real Nano Banana / GPT Image rival.
Plus: Seedance 2.5 slips again but the new leaked clips look incredible, ByteDance drops an official Seedance short film, LingBot-World 2.0 puts a real-time open-source world model on Hugging Face, and Anam's CARA-4 gives AI voice agents a face.
Thanks to Teachable for sponsoring today's video! Sign up for my AI Academy live session: teachable.com/events/build-your-own-ai-film-studio-the-production-company-on-your-desktop?utm_source=TheoreticallyMedia&utm_medium=partnership&utm_campaign=aiacademy&utm_content=youtube
LINKS:
► ByteDance's Seedance short film (full, with audio): https://x.com/BytePlusGlobal/status/2077321849806234080?s=20
► LingBot-World 2.0 (GitHub): github.com/robbyant/lingbot-world
► LingBot-World 2.0 online demo: https://www.reactor.inc/
► Seedream 5.0 Pro on Dreamina: dreamina.capcut.com
► Anam CARA-4: anam.ai/cara4
NEWSLETTER: theoreticallymedia.com/newsletter
Chapters
00:00 Intro: Still Waiting on Seedance 2.5
00:22 Seedream 5.0 Pro: ByteDance's New "Thinking" Image Model
00:41 The Feature Stack: Native Text, 10 Refs & Layer Generation
01:32 First Test: The Man in the Blue Business Suit
02:07 Punch-In Problems: Return of the Blobby Faces
02:48 The Trademark Logo Problem
03:06 The UK Living Room Detail Test
03:53 Reference Tests: The Oni Stuntmen
04:42 Flamethrower Girl: Where References Get Weak
05:20 Tuned Prompts: Technically Fine, Kind of Bland
06:03 Where Seedream Shines: Close-Ups, Genres & Environments
06:49 Wong Kar-wai by Way of Cyberpunk
07:07 The Dreamina Layers Editor: Marketing vs. Reality
08:04 Adding a Layer: The Cyberpunk Cab Problem
08:38 Remove Background's Flipped Masking
08:58 Compositing Flamethrower Girl (A Failed Relight)
10:07 Seedream 5.0 Pro: The Verdict
10:51 Seedance 2.5 Delayed Again (July 20?)
11:04 Leaked Clip 1: The Train Station Reunion
12:29 Leaked Clip 2: Yes, Chef
13:20 Leaked Clip 3: "We Made It"
14:06 ByteDance's Official Seedance Short Film
14:45 Physics Nitpicks & That GPT-Image Grain
15:21 Sponsor: My Teachable AI Academy Live Session
16:48 Real-Time World Models: LingBot-World 2.0
17:25 Text-Triggered Events & the "Brain/Cerebellum" System
17:47 Hands-On: Siege of the Keep
18:47 Where the World Starts to Melt
19:20 The Respawn Trick & Holding Coherence
20:19 World Models as Virtual Sets (+ Where to Get It)
20:54 Anam CARA-4: Real-Time AI Avatars
21:20 AI Voices Are About to Get Faces
21:52 Wrap-Up




![Seedance 2.0: Released & Unlimited!!
Seedance 2.0 has finally landed in the US — meaning its now fully global. Today Im breaking down where youre getting the best bang for your buck (hint: unlimited plans are hard to beat), what the current content restrictions look like, workarounds Ive found so far, and some good news on the moderation front. Plus, we check back in on PixVerses updated real-time video model — and yeah, its getting pretty cool.
A BIG Thanks to Artlist for Sponsoring Todays Video: Try them out at:
https://artlist.io/artlist-70446/?artlist_aid=TheoreticallyMedia_4024&utm_source=a[…]um=TheoreticallyMedia_4024&utm_campaign=TheoreticallyMedia_4024
🔗 Links mentioned in this video:
RunwayML: https://runwayml.com/
(FYI: Promo Code THEO25 for 25% off)
Kling 3.0 Tutorial (Guest Host Spot): https://youtu.be/WmC1nqz10Qk?si=ILgkmDPSxFqj1VcF
PixVerse Real-Time Video: https://app.pixverse.ai/home
📬 Subscribe to the newsletter for compiled workarounds and tips: https://theoreticallymedia.beehiiv.com/
Chapters:
00:00 - Intro
00:30 - Seedance 2.0 US Release Confirmed
00:43 - Image-to-Video & Face Filtering Changes
01:03 - No More Sketchy Third-Party Sites
01:37 - Best Bang for Your Buck: Runway
02:16 - Runway Unlimited Plan Breakdown
03:09 - Current Restrictions & Limitations
03:27 - What Works: Flamethrower Girl & Lyra
04:00 - Whats Blocked: Tom & Sunny
04:12 - Kling 3.0 Omni Reference Comparison
04:33 - Character Swap Workaround
05:22 - Animated Outputs: No Restrictions
05:41 - Text-to-Video Location Trick
06:04 - Chaining Scenes with Continue
06:32 - Prompt Experimentation Tips
07:35 - The Log Lady Fix
07:59 - Creative Partner Program & Realistic Faces
08:54 - Workarounds Roundup & Newsletter
09:05 - PixVerse Real-Time Video Update
09:15 - Lyria 3 Pro With Artlist!
13:10 - PixVerse C1 + Real-Time Avatar Update
13:56 - How to Access the Real-Time Generator
14:26 - Building a Backrooms World
14:53 - Live Real-Time Demo
16:31 - Will We Miss the Wonky AI Era?
17:18 - Try PixVerse Real-Time (Free)
17:26 - Outro Seedance 2.0: Released & Unlimited!!](https://i.ytimg.com/vi/rwnHJ6qlbUA/mqdefault.jpg)





