Anthropic just released a slew of Claude connectors for the big creative apps — Blender, the Adobe suite, SketchUp, and more — and the hot takes immediately rolled in declaring 3D artists, video editors, and photo retouchers obsolete. Time for a reality check.
In this one I take Claude's new MCP integrations on a tour through actual creative work: handing it Blender Guru's famous donut tutorial to recreate from scratch, putting the Adobe Express connection through its paces with a Flamethrower Girl reframe and a white balance pass, and asking SketchUp to design me a one-bedroom apartment. Results range from "honestly not bad" to "drunk intern at 4 AM" — and the Adobe rollout in particular isn't quite what the announcement teased.
I also take a look at where one-shot agentic AI video workflows currently sit, by running the same script through an agent with zero direction and comparing it to my own hand-edited 30-second short. Spoiler: the rhythm of a longer piece is still the work of a skilled human, with or without AI in the chain.
If you've been wondering whether these connectors are actually coming for creative jobs — or where they're genuinely useful as an assistant — this one's for you.
⏱ CHAPTERS
00:00 — "Claude Killed Every Creative Job" (the hot takes)
00:38 — Anthropic's new Claude connectors, briefly
01:01 — The big three: Blender, Adobe & SketchUp
01:36 — Setup: Desktop Claude, MCP & computer use
01:52 — What MCP actually is (and why this isn't new)
02:21 — The mission: Recreate Blender Guru's Donut
02:42 — Attempt #1: "Make me a donut in Blender"
03:26 — The Mad Max donut
04:14 — Macro photography & the thimble coffee cup
04:36 — PBR textures finally arrive
05:47 — The magenta crash-out (drunk intern phase)
06:36 — The verdict: 2 hours, 60% of tokens
07:05 — The "Will Smith eating spaghetti" stage of AI 3D
07:52 — Where Blender + AI could actually go
08:28 — The Adobe connectors (...I have notes)
08:50 — Surprise: it's just Adobe Express
09:06 — Reframing Flamethrower Girl (3:14 later…)
09:55 — The white balance fail
10:13 — Doing it in Photoshop in 13 seconds
10:33 — SketchUp builds me an apartment
10:51 — The $3K NYC apartment with no bathroom door
11:23 — Ranking the three connectors
12:08 — Claude as a software tutor (actually decent?)
12:23 — My hand-edited 30-second short
13:53 — Same script, agent on full auto
14:23 — Why agentic video still feels choppy
14:49 — Where this is actually heading
🔗 LINKS
📩 Theoretically News (the newsletter): https://theoreticallymedia.beehiiv.com/
🤝 Business inquiries: tim@smoothmedia.co
👇 Tell me in the comments — if you've used the Claude Blender, Adobe, or SketchUp connectors yet, what'd they actually nail and where did they fall apart?
Artlist just dropped Artlist Studio — a full cinematic AI workflow that puts every model you actually use (Nano Banana Pro, Seedance 2, Kling, Veo 3.1, Lyria 3, ElevenLabs, and more) into one creative flow state. In this hands-on walkthrough, I build a short scene with consistent characters, voices, and locations to find out where it shines and where it still needs polishing.
🎬 Try Artlist Studio: https://artlist.io/artlist-70446/?artlist_aid=TheoreticallyMedia_4024&utm_source=affiliate_p&utm_medium=TheoreticallyMedia_4024&utm_campaign=TheoreticallyMedia_4024
What I cover:
The full Artlist AI Toolkit lineup (image, video, voice, music)
Building consistent characters from scratch (or using their templates)
Locking in character voices — and what's coming with custom voices
Generating locations that play nice with your characters
The Framing tab: first-frame generation with Nano Banana Pro, GPT Image 2, Flux 2
Honest credit-cost breakdown (and which models are the actual bargains)
The Directing tab: Seedance, Kling 2.6, and Grok Imagine for video
Hidden camera and lens prompts (Arri Alexa 35, Sigma, Helios, Lomo)
Structured prompts vs. freehand — which one's worth your time
A weird experimental warp transition trick
Models and tools mentioned in this video:
Artlist Studio • Nano Banana Pro • GPT Image 2 • Flux 2 • Seedance 2 • Veo 3.1 • Sora 2 • Kling 2.6 • Grok Imagine • ElevenLabs • MiniMax • Sonic 2 • Lyria 3 Pro • Midjourney 6.1
⏱️ Chapters:
00:00 Artlist Studio is here
00:30 Artlist Hops into the AI Game
01:31 Introducing Artlist Studio
02:14 NPC: A Super Short Film
03:10 Walkthrough: the Studio UI
03:32 Building a custom character
04:57 Choosing a character voice
05:12 Adding a secondary character (NPC)
06:01 Building locations
06:42 Framing: first-frame generation
07:22 Model picker and credit costs
08:28 Resolution: 1K vs 2K vs 4K
09:05 Camera and lens controls
10:03 Aspect ratios
10:18 Directing: generating video
10:38 Video models: Seedance, Kling, Grok
12:06 Shot types and structured prompts
12:44 Working with the timeline
13:23 Experimental warp transitions
14:20 Final thoughts and what's next
15:11 Outro
Got a tool or workflow you want me to break down next? Drop it in the comments — I read every one and pass the best feedback straight to the teams building these tools.
#AIFilmmaking #ArtlistStudio #AIVideo
Alibaba quietly launched HappyHorse-1.0 and it shot to #1 on the Artificial Analysis video leaderboard, with a lot of people calling it a Seedance killer. Short answer: it's not — at least not yet. But the team behind Happy Horse is led by Zhang Di, the architect of Kling 1.0 and 2.0, who left Kuaishou last fall, joined Alibaba in November, and shipped this in roughly five months. So yeah, this is a pony worth keeping an eye on.
Also in this one: Topaz drops Starlight Precise 2.5 (and surprise-launched a new creative model mid-shoot), Kling goes native 4K (real 3840×2160, no sneaky upscaler), and Netflix's Eyeline Labs releases Vista4D — an open-source 4D reshooting framework that lets you retarget the camera on existing footage. Open source. From Netflix. 2026 is weird.
LINKS:
Happy Horse: https://www.happyhorse.com
TOPAZ: https://topazlabs.com/ref/2518/
Kling: https://klingaiaffiliate.pxf.io/qz3955
Vista 4D: https://github.com/Eyeline-Labs/Vista4D
Signup for my Newsletter: https://theoreticallymedia.beehiiv.com/
Chapters:
00:00 Big AI video updates
00:55 Happy Horse from Alibaba Arrives
01:10 Leaderboard hype and early buzz
01:30 Who made Happy Horse?
01:56 Happy Horse testing begins
02:18 First text-to-video results
02:38 Prompting discovery: keep it short
02:59 Prompt format and token spam
03:23 Why brevity matters
03:43 Image-to-video test: Twin Peaks diner
04:08 Lip sync and voice issues
04:40 Prompt adherence wins
04:57 Kung-fu/action limits
05:11 Can Happy Horse handle action?
05:39 Current limitations and missing features
05:58 1080p, 15-second generations, and batch output
06:08 Reference / Omni mode
06:38 Reference mode problems
06:57 Prompt troubleshooting
07:17 Reference mode working well
07:57 Is Happy Horse a Seedance killer?
08:18 Topaz update begins
08:49 Starlight Precise 2.5 test
09:12 4K upscale detail check
09:23 Cleaning without changing faces
09:44 Skin texture test
09:59 Another Seedance upscale test
10:15 Precise mode vs creative detail
10:36 The unhinged Bruce Lee Terminator test
11:05 Starlight Precise 2.5 is available
11:17 Astra Creative 2 drops
11:36 Kling native 4K generation
11:55 Robotech/Macross 4K test
12:18 Why native 4K matters
12:34 Stock and travel footage use case
13:03 When to actually use 4K
13:18 Kling 4K verdict
13:35 Every AI video model has strengths
13:56 Netflix Vista4D
14:13 How Vista4D works
14:25 Compared with Google Flow and Veo 3 experiments
14:40 Camera control limits
15:01 Video outpainting and Wan 2.1
15:11 Where to try Vista4D
15:22 Wrap-up
Here's some insane Seedance 2.0 prompts & tips you need to know! The Prompts are broken down so you can steal them, modify them, and build your own. From snap-stop time shockwaves to first-frame/last-frame storytelling, time-loop sequences, and an invisible VFX shot from the 1997 film Contact — this one's a tactical breakdown, not just a prompt dump.
Plus we head back over to Martini for a feature I haven't seen anywhere else (and yes, you can skip the waitlist 👇).
🍸 Skip the Martini waitlist → https://www.martini.film/theoretically-media
And if you use the link:
$30 plan → $20 bonus credits ($50 total for the first month)
$150 plan → $50 bonus credits ($200 total for the first month)
📩 Get this week's newsletter (with all the prompts copy-paste ready) → https://theoreticallymedia.beehiiv.com/
🎬 Don Burgess on the Contact mirror shot →https://youtu.be/HQRu9cz5L9E?si=B47-tPxdbKUS8NjT
CHAPTERS
00:00 Intro — Why These Seedance Prompts Are Going Viral
00:39 - Prompt Intro
00:56 Snap Stop-Time Prompt —
02:04 - My Stop Time Test
02:41 - First Frame / Last Frame Trick — Emergent Storytelling
03:19 - My Pirate Test
03:59 - Weapons Test
04:29 - Real Faces Appearing Now
04:59 - Day-In-The-Life Time Loop
05:59 - GPT Image 2 Character
06:31 - My Loop Test
06:51 - Rerolling
7:44 "Contact" Mirror
08:41 - Seedance Recreation
09:13 - My Contact Test
09:36 - How Many ReRolls?
10:08 Martini's New Studio Feature
10:48 Step Into Set Improvements
11:45 - Things get Wild here
12:00 - Step Into Set With Characters
12:45 - Character Turnarounds
13:22 - Character Space Controls
14:10 - Why This is So Good
14:55 - Turnaround Editing
15:48 - The Multiplayer Feature
16:19 - Why Multiplay?
17:08 Skip the Waitlist
18:04 Final Thoughts & What's Next
OpenAI just dropped Image 2 — their new autoregressive image generation model inside ChatGPT. But is it actually good enough to dethrone the banana? 🍌🔥
In this video, I put GPT Image 2 through a full battery of standardized tests — from wine glasses and analog clocks to pelicans riding bicycles — and compare it head-to-head against Nanobanana and other leading AI image generators. I cover what's improved (aspect ratios, text generation, character consistency), where it falls short (artifacts, crunchy textures, inconsistent guardrails), and share a critical tip that will save you hours of frustration.
Whether you're using AI image generation for thumbnails, concept art, movie posters, or just having fun — this breakdown will help you understand exactly where Image 2 fits in the current landscape.
🔥 What's Inside:
Full standardized test suite (wine glass, pelican, avocado armchair)
Aspect ratio freedom — finally!
Text generation accuracy (strawberry test, chalkboard test)
Image referencing & character consistency vs Nano Banana
IP guardrails & prompt pushback — the weird mixed bag
The artifact problem — and the simple fix
Spatial awareness & 180-degree rule test
Style transfer: illustrated to photorealistic
Final verdict: Banana killer or close second?
NEWSLETTER! https://theoreticallymedia.beehiiv.com/
⏱️ CHAPTERS:
00:00 OpenAI's Challenger to the Banana Throne
00:20 Why OpenAI Needed This Win
01:12 What We're Testing Today
01:26 Aspect Ratios — Finally Unlocked
01:54 Standardized Test: Wine Glass & Clock
02:43 Standardized Test: Pelican on a Bicycle
03:14 The Ultimate Combo Test
03:37 DALL-E 1 vs Image 2 — Five Years Later
04:03 Aspect Ratio Deep Dive (1988 Mall)
04:54 Ultra Widescreen & Spaghetti Westerns
05:14 Text Generation — Hit or Miss
05:58 The Strawberry Counting Test
06:28 Image Referencing & Flamethrower Girl
07:17 Real People & Thumbnail Generation
07:38 IP Guardrails — The Weird Mixed Bag
08:40 Nonsensical Pushback
09:12 The Artifacting Problem
09:58 Image 2 Explains Why It's Struggling
10:34 The Fix — Just Open a New Chat
11:10 Spatial Awareness & the 180° Rule
12:34 Upscaling Workaround with Magnific
12:43 Style Transfer Test
13:27 The Eight Fingers Challenge
13:50 Final Verdict — Is It a Banana Killer?
#openai #chatgpt #image2 #gptimage #imagegeneration #artificialintelligence #aicreative
Great visuals deserve great audio — and until now, AI filmmakers have been stuck scoring silent scenes with royalty-free tracks and bolting on sound effects after the fact. Ace Studio changes that. It's an AI-native digital audio workstation built for filmmakers, with a Video Composer agent that watches your footage and builds a soundtrack around it, one-shot AI sound effect generation, stem splitting, vocal synths, and a full DAW underneath when you want to get hands-on.
In this tour, I run every major feature through real generations from the channel — the FBI diner scene, Renfield the Pirate, Flamethrower Girl, a silent Seedance cockpit sequence, and Malloy the detective — to show you exactly what this thing can do for AI video work.
🎁 Try Ace Studio (exclusive link + discount):
👉 https://acestudio.ai/?promo=TheoreticallyMedia
💸 Discount Code: TheoreticallyMedia
🔑 What's covered:
- AI Video Composer — automatic scoring that reads your footage
- One-shot AI sound effects (Foley, ambience, SFX stacking)
- The AI agent that scores + SFXs an entire video in a single prompt
- Building soundscapes from scratch on a silent generation
- Audio effects, reverb, and source audio manipulation
- AI instruments + MIDI keyboard input
- VST3/AU bridge to Ableton, FL, Logic, Studio One
- Stem Splitter — pull isolated audio out of ANY AI generation
- AI vocal synths with custom lyrics
This is easily one of the most complete audio solutions I've seen built specifically with AI filmmakers in mind, and the stem splitter alone is worth the price of admission for anyone who's ever prompted "no music" and had the model ignore you anyway.
⏱️ CHAPTERS
00:00 Sound is Half Your Picture
00:41 Meet Ace Studio — AI DAW for Filmmakers
01:20 Video Composer: AI That Scores Your Scenes
01:31 Test 1 — Scoring the FBI Diner Scene
02:47 Test 2 — Renfield the Pirate + Reverb FX
04:26 AI Sound Effects (Flamethrower Girl)
06:26 The AI Agent: One-Shot Full Soundscape
08:16 Building Audio from a Silent Seedance Scene
09:45 AI Instruments + MIDI Keyboard
10:48 DAW Bridge (VST3/AU for Power Users)
11:10 Stem Splitter — The Hidden Superpower
12:46 AI Vocal Synths with Custom Lyrics
13:50 Final Thoughts
📬 Newsletter: https://theoreticallymedia.beehiiv.com/
#AceStudio #AIMusic #AIFilmmaking #AISoundEffects #AIVideo #AIAudio #AIDAW #AITools #GenerativeAI #Filmmaking #AIFoley #Seedance #MusicProduction #AIForCreators #TheoreticallyMedia
Seedance 2.0 has finally landed in the US — meaning it's now fully global. Today I'm breaking down where you're getting the best bang for your buck (hint: unlimited plans are hard to beat), what the current content restrictions look like, workarounds I've found so far, and some good news on the moderation front. Plus, we check back in on PixVerse's updated real-time video model — and yeah, it's getting pretty cool.
A BIG Thanks to Artlist for Sponsoring Today's 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 - What's 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
OpenAI's next image model — likely GPT-Image 2 — is already being tested under stealth names on the Arena leaderboards, and the results are impressive. Today we break down the mystery models (Masking Tape Alpha, Gaffer Tape Alpha, Packing Tape Alpha), compare OpenAI's new image generation head-to-head with Nano Banana 2, and explain why the upcoming "Spud" update isn't just another image model — it's the image capability of a new autoregressive multimodal thinking model that could change everything.
Plus: Milla Jovovich open-sourced an AI memory system called Mem Place, a mysterious video model called "Happy Horse" just dethroned Seedance 2.0 on the leaderboards, Pixverse dropped their cinematic C1 model, and Galileo Zero introduces a "world critic" for AI video quality control.
🔗 Links mentioned in this video:
Mem Place (GitHub): https://github.com/milla-jovovich/mempalace
Galileo Zero waitlist: https://physionlabs.ai/blog/galileo-0
Pixverse C1: https://app.pixverse.ai/
Friday's newsletter (Wen 2.7 coverage): https://theoreticallymedia.beehiiv.com/
📩 Join the Theoretically Media newsletter for daily AI creative tool updates: https://theoreticallymedia.beehiiv.com/
#AIVideo #OpenAI #ImageGeneration #AITools #AINews #GenerativeAI #NanoBanana #Seedance #AIFilmmaking #CreativeAI
CHAPTERS:
00:00 — Is OpenAI's New Image Model a Banana Killer?
00:29 — Mystery Models Hit the Arena Leaderboards
01:09 — AI-Generated YouTube Homepage Breakdown
02:24 — Detailed Image Analysis: Receipts, Surfers & Maps
03:10 — Nano Banana 2 vs GPT Image Model Comparison
03:34 - Text Rrendering in a Store Front
04:03 - MAPS!
04:35 — AI-Generated YouTube (again)
05:17 — GTA VI: Will We Get AGI First?
05:57 — Autoregressive Thinking Models Explained
07:22 — Milla Jovovich Open Sources AI Memory (Meme Place)
07:58 — "Happy Horse" Dethrones Seedance 2.0
08:31 - Wan 2.7 thoughts
08:46 — Piixverse Drops Cinematic C1 Model
08:58 - Pixverse Does what Seedance Can't...or Won't.
10:01 — Piks C1: Multi-Frame Input Demo
10:37 — Galileo Zero: A World Critic for AI Video
12:28 — Closing & What's Coming This Week
Seedance 2.0 just went global — well, mostly. We break down exactly what the release means (and doesn't mean), including API pricing, business account requirements, image rights contracts, and one platform where you can try it right now.
Plus: Tencent's Wen 2.7 is about to drop with first/last frame generation, 9-grid image-to-video, voice referencing, and instruction-based editing. Could it take the #2 spot from Kling 3.0?
We also test Magnific's new video upscaler and Topaz's Astra model for AI video upscaling — side-by-side comparisons included.
And then things get weird: Pika just launched real-time video chat with AI agents through Pika Me. So naturally, I sat down for a live face-to-face interview with Flamethrower Girl. It's janky. It's fascinating. It's the future.
Thanks to today's sponsor: Wispr Flow!
Head over and get a month of Pro with my code: THEORETICALLY
https://ref.wisprflow.ai/theoretically
🔗 LINKS:
Venice.ai (Seedance 2.0 access) — https://venice.ai
Pika Me — https://pika.me
Magnific Video Upscaler — https://magnific.ai
Topaz Video AI (Astra) — https://topazlabs.com/ref/2518/
Wen 2.7 (Tencent) — https://wan.video/
FREE NEWSLETTER! (News, Tips, & More!)
https://theoreticallymedia.beehiiv.com/
📌 CHAPTERS
00:00 — Intro
00:37 — Seedance 2.0 Global Release (Minus US & Japan)
00:56 — API Rollout & Country Availability
01:06 — Business Account Requirements & Pricing
01:21 — Image Rights & Likeness Contracts
01:54 — Enterprise-Level Pricing Breakdown
02:28 — Venice.ai: Seedance 2.0 Access Now
03:26 — Wen 2.7 From Tencent: What We Know
03:52 — Expected Wen 2.7 Features
04:07 — Can Wen 2.7 Dethrone Kling 3.0?
04:44 — Kling 3.1, Veo 4 & What's Coming
04:54 — Magnific Video Upscaler Test
05:27 — Magnific Results: 720p to 2K
06:22 — Topaz Astra AI Upscaler Test
6:48 — Topaz Auto Scene Detection Feature
07:04 — Wispr Flow
10:28 — Pika Me: AI Agent Video Chat
11:13 — Live Interview With Flamethrower Girl
15:05 — Post-Interview Breakdown: Latency, Voice & Jank
15:50 — Running Pika Me Through Your Own Agent
15:59 — Final Verdict on AI Video Chat
16:23 — Outro
🔔 Subscribe for daily AI creative tool breakdowns, reviews, and workflows: https://www.youtube.com/@TheoreticallyMedia?sub_confirmation=1
#AIVideo #Seedance2 #AIAgent #AIAvatar #AIFilmmaking #Magnific #TopazAI #AIUpscaling #AINews #GenerativeAI #AICreativeTools #TheoreticallyMedia #Kling #aitools2026
Recraft V4 is here and the new Recraft Studio might be the most underrated platform in AI image generation right now. Today I'm walking through everything that's new — the V4 model head-to-head with Nano Banana, the revamped Studio interface, vector/SVG generation and editing, node-based workflows with mockup deformation, exploration and agentic prompting modes, and the wild new OpenClaw integrations!
Try Recraft for FREE: https://go.recraft.ai/Theoretically_Media
🔑 Key Topics:
— Recraft V4 vs Nano Banana 2 — side-by-side comparison and cost breakdown
— Recraft Studio interface walkthrough — generation modes, model picker, palettes
— Vector/SVG output and raster-to-vector conversion in Illustrator
— Node-based workflows — mockup deformation for logos on clothing and surfaces
— Exploration mode and agentic conversational prompting
— OpenClaw integration- First Time I've seen that!
#Recraft #RecraftV4 #RecraftStudio #AIImageGeneration #AIDesign #SVG #VectorArt #NanoBanana #OpenClaw #MCP #Claude #GenerativeAI #AITools #TheoreticallyMedia
—
🔔 Subscribe for daily coverage of Creative AI tools: https://www.youtube.com/@TheoreticallyMedia?sub_confirmation=1
📧 Business inquiries: tim@smoothmedia.co
—
CHAPTERS
0:00 Intro
0:30 What Is Recraft?
1:02 Video Models Added (But We're Doing Images)
1:46 Recraft V4 Model Overview
2:17 V4 Design Philosophy
2:41 Recraft V4 vs Nano Banana 2
3:59 V4 Cost Advantage
4:21 Recraft Studio Interface
4:35 Prompt Bar & Generation Modes
4:54 Model Picker & Available Models
5:24 Style Library Status
5:55 Output Resolutions & Credit Costs
6:36 V4 Image Showcases
7:02 Vector/SVG Generation
7:17 Exporting SVGs to Illustrator
7:58 Vectorizing Raster Images
8:53 Flamethrower Girl Vector Test
9:17 Node-Based Workflows
9:53 Mockup Workflow Demo
10:42 Logo Deformation on Clothing
11:31 Workflow Templates & Editing Tools
11:59 3D Me Template
12:16 Manual Prompting Deep Dive
12:52 Exploration Mode
13:08 Agentic Mode
13:37 OpenClaw Integration
14:24 Recraft MCP for Claude
15:14 Final Thoughts & Free Tier
#Recraft #RecraftV4 #RecraftStudio #AIImageGeneration #AIDesign #SVG #VectorArt #NanoBanana #OpenClaw #MCP #Claude #GenerativeAI #AITools #TheoreticallyMedia
Here's the full AI Film masterclass breakdown — every tool, every technique, every mistake, and exactly what it cost.
This is a complete production walkthrough of "Dragon Blue," my AI short film made with Seedance 2.0, Nano Banana Pro, Claude Cowork, and the Luma Agent board. I cover the entire pipeline from pre-production and reference image generation through video generation, post-production, and final delivery — including the multilingual prompting tricks, safety filter workarounds, and Omni model techniques that made this workflow possible.
Whether you're exploring AI filmmaking for the first time or looking to level up your production pipeline, this video breaks down a real workflow you can replicate with the tools available right now.
🎬 WATCH DRAGON BLUE (Clean Version):
https://youtu.be/dRjN6Cr2Z00
🔧 TOOLS USED IN THIS VIDEO:
Seedance 2.0 (on Dreamina) — https://dreamina.capcut.com/ai-tool/home?utm_source=Officiaaccount&utm_campaign=sd2&utm_content=36x
Claude / Cowork by Anthropic — https://claude.com/product/cowork
Luma Agent Board — https://lumalabsai.partnerlinks.io/dd1jzuzx6o87
Topaz Video (Upscaling) —https://topazlabs.com/ref/2518/
Suno (AI Music) — https://suno.com
Adobe Podcast Enhance Speech — https://podcast.adobe.com/en/enhance
📰 NEWSLETTER (Full prompt templates, scene designs & Claude project files):
https://theoreticallymedia.beehiiv.com/
⏱️ CHAPTERS:
0:00 Intro
0:42 Watch: "Dragon Blue" AI Short Film
4:27 Kill Bill Vibes & How This Started
5:02 The Tools: Seedance 2.0, Nano Banana Pro, Claude & More
5:36 Claude Cowork as a Production Office
7:59 Reference Images with the Luma Agent Board
9:07 Scripting & Nano Banana Prompt Templates
9:58 Spray & Select Workflow for Image Generation
11:13 Color Palette & Scene Design Choices
12:26 Moving into Seedance 2.0 (Dreamina Omni Model)
13:19 Struggles: Getting Out of the Car Scene
13:53 Hero Moment #1: The Fight Scene
15:10 Hero Moment #2: The Silhouette Fight (Oner Attempt)
16:10 Multilingual Prompting Trick (Chinese Prompts)
17:07 Hero Moment #3: Katana Girl in the Sunglasses
17:29 Japanese Dialog & Multilingual Generation
17:54 Safety Filter Frustrations
18:21 Post-Production & Editing Workflow
19:00 Continuity Errors (Yes, I Saw Them Too)
19:38 Production Time & Full Cost Breakdown ($187)
21:19 AI vs Hollywood Budget Comparison
22:24 "Is This a Real Movie?" — The Big Picture
23:25 Final Thoughts: Just Make Something
#AIFilmmaking #Seedance2 #AIShortFilm #NanoBananaPro #ClaudeAI #LumaAI #AIVideoGeneration #AIFilm #Dreamina #AITools #TheoreticallyMedia #AIWorkflow #AIMovieMaking #IndieFilmmaking #GenerativeAI #AICreativeTools #Seedance #AIProduction #VideoGeneration2026 #AIFilmBreakdown
"Dragon Blue" is a hyper-stylized action thriller about vengeance and katanas, what more could you ask for? Every shot was generated with Seedance 2.0, and I'm really proud of how this one turned out.
I'll be back shortly with a full Masterclass on the workflow for this film, and we've got a LOT to go over, so make sure to subscribe to the channel!
Dreamina — https://dreamina.capcut.com
📬 Newsletter —https://theoreticallymedia.beehiiv.com/
Seedance 2.0 has finally released — sort of. After months of delays, lawsuits, and rumors of cancellation, ByteDance has made the model available on CapCut and Dreamina... in seven countries. We break down what's actually available, the new guardrails (including real face restrictions and C2PA watermarking), whether VPNs are worth the headache, and what this all means for a wider Western release.
Plus, Luma Labs dropped Uni One — a new thinking image model — and when you pair it with their boards and agents feature, the storyboarding and pre-production workflow gets genuinely powerful. We walk through a full pipeline from character reference to storyboard to video generation using Seedance 2.0 and Luma's canvas tools.
Seedance 2.0 samples, prompting observations, the Omni model on CapCut, Luma Uni One, Luma Boards and Agents, thinking image models, AI video generation workflow, storyboard-to-video pipeline, region restrictions, C2PA content credentials, IP guardrails.
🔗 LINKS & RESOURCES
CapCut — https://www.capcut.com
Dreamina — https://dreamina.capcut.com
Luma Labs —https://lumalabsai.partnerlinks.io/dd1jzuzx6o87
Artificial Analysis Leaderboard — https://artificialanalysis.ai
📬 Newsletter —https://theoreticallymedia.beehiiv.com/
🎬 Short Film (drops tomorrow) — [LINK WHEN LIVE]
CHAPTERS
0:00 — Seedance 2.0 Has Released (Kinda)
0:49 — The Full Seedance Saga Recap
1:50 — Where It's Available (and Where It Isn't)
2:28 — New Guardrails: Faces, IP Blocking, C2PA
3:02 — When Will It Go Wide?
3:25 — First Samples: Text to Video
4:35 — Famous Faces Are Off the Table
5:15 — Image to Video & The Omni Model
6:09 — Honest Takes: What Works, What Doesn't
7:10 — Morphing, Decoherence & Editing Around Weird Choices
8:02 — CapCut's Canvas Approach
8:55 — Luma Uni One: Thinking Image Model
9:49 — Luma Boards + Agents: The Real Power Move
11:16 — Combining Thinking Image + Thinking Canvas + Thinking Video
11:59 — Is Seedance 2.0 the Best Video Generator Right Now?
12:50 — Sign-Off + Short Film Tomorrow
Midjourney V8 Alpha is here — and it's complicated. In this video, I burn through my fast hours to give the new model a fair shake. We run it through the standard tests (toast, blue suit guy, Flamethrower Girl), crank stylize from 100 to 1000, explore the new Style Creator tool, test HD mode and Quality 4, try personalization codes, and ultimately figure out what Midjourney is actually building toward. Plus a Nano Banana Pro rescue workflow for taking V8's best stylistic outputs and making them production-ready.
🔗 LINKS
Midjourney: https://midjourney.com
Midjourney V8 Alpha: https://alpha.midjourney.com
⏱ CHAPTERS
00:00 Midjourney V8 Is Here
00:24 A Brief History of Midjourney Updates
00:56 What's New in V8 Alpha
01:17 The Toast Test
01:53 Stylize Settings Deep Dive
02:25 Blue Suit Guy at Stylize 100–1000
03:53 HD Mode & Quality 4 (Don't Do This)
04:23 V8 Alpha Quirks & Observations
04:48 Personalization Codes & Profiles
05:38 Cyberpunk Woman Prompt Roller Coaster
07:24 Style References in V8
08:31 Style Creator: Where It Gets Weird
09:45 V8 as a Stylistic Exploration Tool
10:14 Flamethrower Girl in V8
11:47 Nano Banana Pro Rescue Workflow
12:11 V8 Alpha Costs & Technical Details
13:31 Where Midjourney Is Really Headed
15:00 Final Verdict: Did They Cook?
#midjourney #midjourneyv8 #aiart #aivideo #imagegeneration #creativai #nanobanana #theoretically media
Nvidia’s GTC 2026 keynote just dropped some massive AI hardware reveals! From the insane Vera Rubin super platform to the game-changing DLSS 5 neural rendering and the localized NemoClaw AI agent, here is everything AI creatives and gamers actually need to know.
Jensen Huang took the stage to map out the next two years of AI infrastructure. While massive data centers and million-dollar racks like the NVL72 seem out of reach, this tech directly trickles down to reduce the speed and cost of your everyday AI video and image generation.
In this video, I cut through the noise of the 2.5-hour keynote to break down the actual specs of the Rubin GPU and Vera CPU, why gamers are debating the "GPT moment for graphics" with DLSS 5, and how Nvidia’s NemoClaw is stepping up to fix OpenClaw by bringing secure, local AI agents directly to your PC.
If you want to stay ahead of the curve in the AI and creative space, make sure to hit that subscribe button! 🛎️
👇 Mentioned in this Video:
🔗 Watch my breakdown of the ByteDance Seed 2.0 "Thinking" Video Model: https://youtu.be/yLQClFqzHOU
⏳ VIDEO CHAPTERS:
00:00 Nvidia GTC 2026 & AI Infrastructure
00:45 Vera Rubin AI Supercomputer Specs Revealed
03:14 How Vera Rubin Lowers AI Token Costs
04:08 ByteDance Seed 2.0 & Faster AI Video Generation
05:32 DLSS 5: Real-Time Neural Rendering Explained
06:33 The DLSS 5 "Uncanny Valley" Gamer Backlash
08:43 OpenClaw AI Agents Explained
09:29 NemoClaw: Nvidia's Secure Local AI Agent
10:36 Cloud vs. Local AI Models & The DGX Spark
11:28 How Will Google Respond at I/O?
💡 KEY TAKEAWAYS FROM GTC 2026:
• Vera Rubin Super Platform: Combining the Rubin GPU and Vera CPU with the NVL72 rack system to create 260TB/s of throughput. While it targets data centers, it will deliver 10x more throughput per watt—meaning cheaper and faster AI token generation for everyone by 2027.
• DLSS 5 (Deep Learning Super Sampling): Real-time neural rendering is here. It's essentially running a creative AI upscaler (like Magnific or Topaz) over the game engine in real-time. Is it the GPT moment for gaming graphics, or just uncanny valley?
• NemoClaw: Nvidia’s answer to OpenClaw. A secure, local AI agent you can install with one command line. Powered by Nemotron models (Nano, Super, and Ultra), it acts as your personal "Jarvis" without compromising your PC's security.
#Nvidia #GTC2026 #VeraRubin #DLSS5 #NemoClaw #OpenClaw #ArtificialIntelligence #AIAgents #TechNews
If you've ever bounced off ComfyUI because the node graphs gave you anxiety — this is the update to pay attention to. App Mode turns any workflow into a clean app interface with one click. No nodes. No spaghetti. Just inputs, outputs, and a run button. ComfyHub is the new home for sharing those apps, and Comfy Cloud means you don't even need the hardware.
Plus: Sora adds References, Sora 1 sunsets tomorrow, video gen is moving to ChatGPT, and Claude is generating video by writing code that draws every frame. Welcome to the week.
Big thanks to Wispr Flow for Sponsoring Today's Video: Sign up for a FREE 14 day trial here: https://ref.wisprflow.ai/theoretically
⏱️ CHAPTERS
0:00 - Open
0:38 - ComfyUI App Mode, App Builder & ComfyHub
1:03 - ComfyHub Walkthrough — Workflows, Apps & External Models
2:32 - You Don't Need a Beast PC — Comfy Cloud
3:25 - ComfyUI Apps — How They Work (No Nodes Required)
5:11 - Graph Mode — The Murder Board Is Still There
5:42 - Quick Look: FireRed Image Editor + Flamethrower Girl
6:32 - Building Your Own ComfyUI App
8:16 - Running Complex Workflows Without the Anxiety
9:39 - Mixing API + Local Models in One Workflow
10:34 - Who Is ComfyUI For Now?
11:15 - Wispr Flow
14:25 - Sora: References, ChatGPT Integration & Sora 1 Sunset
14:33 - Sora References — Characters, Styles & Renfield Test
15:45 - Sora Video Gen Coming to ChatGPT
16:11 - Sora Installs Down 45% — What Happened
16:47 - Farewell Sora 1 — Sunset Tomorrow
17:54 - Claude Makes Videos (Sort Of)
18:40 - Claude Video Use Cases + LLM Sandbox Short Film
19:22 - How to Try It + Sign-Off
🔗 LINKS
ComfyUI App Mode Blog Post: https://blog.comfy.org/p/from-workflow-to-app-introducing
ComfyHub (Preview): https://comfy.org/workflows
Sora 1 Sunset FAQ: https://help.openai.com/en/articles/20001071-sora-1-sunset-faq
LTX Comfy Install Video: https://youtu.be/5l4XumW4fVQ
#ComfyUI #AppMode #ComfyHub #AIVideo #Sora #OpenAI #Claude #Anthropic #AITools #GenerativeAI #ComfyUITutorial #AIFilmmaking #NodesOptional #TextToVideo #AIWorkflow #ComfyCloud #SoraUpdate #ClaudeAI #CreativeAI #AIVideoGeneration
LTX Desktop just dropped — a free, open source, fully local non-linear video editor built on the LTX 2.3 engine. Today we're going through the whole thing: how to install it, what it can do, what it can't do, and why I think this matters more than most people realize.
We're also running through the LTX 2.3 model updates including the rebuilt VAE, image-to-video fixes, native portrait video, and audio quality improvements.
LTX Desktop: https://ltx.io/ltx-desktop
LTX 2.3 on Hugging Face: https://huggingface.co/Lightricks/LTX-2.3
LTX API: https://app.ltx.studio/ltx-2-playground/t2v?anonymousId=AC7DB444-6FC2-46E5-AFE9-C034F41A95DF
ComfyUI Beginner Guide:
https://youtu.be/5l4XumW4fVQ
FireRed Image Edit: https://github.com/FireRedTeam/FireRed-Image-Edit
0:00 Intro
0:44 LTX 2.3 Model Improvements
1:21 Portrait Video & Audio Quality
1:27 Hugging Face & ComfyUI Support
1:49 Side Notes: FireRead Image Edit
2:11 The Big Story: LTX Desktop
2:42 How to Install LTX Desktop
3:18 PC Installer Fix (Run as Admin)
3:55 Text Encoder: API vs Local
4:17 API Key Setup
4:32 The VRAM Elephant in the Room
5:28 Mac & Low VRAM: API Generation
5:38 Open Source VRAM Fix
6:01 Playground & Gen Space
7:00 The Video Editor
7:49 Basic Features & Extra Bells
8:42 AI Native: Regenerate & Reroll Shots
10:04 Image to Video on the Timeline
10:43 Bridge Shots: Fill with Video
11:56 Retake Feature
13:05 V1 Assessment: Not Replacing Your Editor
13:35 Why Open Source Matters Here
14:32 Will Traditional Editing Be Automated?
15:08 AI Native NLEs: A New Category
15:27 Outro
#LTXDesktop #AIVideo #VideoEditing #OpenSource #LTX #AITools #AIVideoEditor #CreativeAI #LocalAI
Two brand new AI video platforms just dropped — and they couldn't be more different. First up is Pai from Utopia Studios, an agentic AI video generator that outputs a minute or more of video from a single prompt with narrative continuity, character consistency, and IP-safe generation. Then we dive into Martini, a canvas-based AI video production platform that lets you walk through 3D virtual sets, drag-and-drop references across models like Kling, Veo, Sora, and Seedance, and rough cut your project in a built-in editor with XML export. Big thanks to Martini for sponsoring today's video and hooking you all up with early access and free credits.
Martini (Early Access + Free Credits): https://www.martini.film/early-access?code=BPKBY1
Pai: https://www.utopaistudios.com/pai
0:00 Intro
0:32 Pai from Utopia Studios
1:27 - Cargo Haulers Short Demo
2:19 Agentic Workflow & Character Creation
3:23 Storyboarding & Keyframe Editing
4:38 - Western Short Part 1
5:22 Generation Results & Notes
6:01 Flamethrower Girl Short
6:52 The Multi-Generation Trick
8:00 IP Safety & Character Detection
8:47 Pai Overall Thoughts
9:09 - Subway Kung Fu Fight
9:34 - Seedance Bash
10:03 Editing Is Still the Most Important Skill
10:19 Martini — A Different Approach
10:39 Canvas-Based Interface Walkthrough
11:08 Image Generators — NBP Pro, Flux 2 Max, Nano Banana 2
11:47 Video Generation — Kling, Veo, Sora, MiniMax & More
12:10 Tool Agnostic Philosophy
12:38 Noir Tuesday — The UK Remake
13:56 Project Setup & Character References
14:10 Step Into Set — 3D Virtual Sets
15:02 Snapshot to Nano Banana Upscale
15:46 Virtual Sets + First Frame/Last Frame Generation
16:28 Virtual Set Limitations — Not a World Model
17:12 Drag-and-Drop Frame References
17:32 Built-In Editor & Timeline
18:13 Unlink Audio, Ripple Delete & More
18:35 Export — MP4 & XML to Your NLE
19:03 Martini Final Thoughts
19:14 Early Access, Free Credits & Wrap Up
#AIVideo #AIVideoGeneration #TheoreticallyMedia #Pai #Martini #UtopiaStudios #Kling #Seedance #Sora #NanoBanana #AIFilmmaking #AITools
Nano Banana 2 is HERE — Google just dropped Gemini 3.1 Flash Image and it's packing a feature no other image model has: Advanced World Knowledge. That means it can pull from real-time web information to generate images about things happening RIGHT NOW. I put it through the full battery of tests — text rendering, manga translation, cinematic prompts, stylistic consistency, camera lens simulation, and more. Plus, the thinking mode had a full-on psychotic break (you'll want to see that).
Also: Seedance 2.0 has finally dropped... in CapCut. And there's a wild rumor about the model weights being leaked. Full details inside.
🔗 Try Genspark: https://www.genspark.ai/?utm_source=yt&utm_campaign=TheoreticallyMedia
📌 WHAT'S COVERED:
Nano Banana 2 (Gemini 3 Flash Image) — full breakdown and hands-on tests
Advanced World Knowledge — real-time information in image generation
Thinking Mode results (and one spectacular meltdown)
Seedance 2.0 now available in CapCut — pricing and details
Seedance 2.0 model weights allegedly leaked
⏱️ CHAPTERS:
0:00 — Intro
0:32 — Google's Impressive Release Streak
0:51 — Nano Banana Naming Explained (NB1 vs Pro vs NB2)
1:28 — Nano Banana 2 Key Features
2:25 — Wine Glass & Clock Test
2:59 — Pelican Riding a Bike (SVG Test)
3:41 — Thinking Mode Pelican Test
4:25 — Cinematic Presets & Truck Driver Prompt
5:12 — Underwater 90s Bedroom Test
5:59 — Text Rendering (Tale of Two Cities)
6:44 — Manga Translation Test (Akira)
7:37 — Stylistic Consistency & Aspect Ratio
8:26 — Extra Finger Thumbnail Test (MJ V8 Teaser)
9:23 — Creativity Test (Flamethrower Girl's Day Off)
9:54 — Camera Lens Simulation
10:21 — Gemini's Thinking Mode "Psychotic Break"
11:33 — Genspark (Sponsor)
15:50 — Advanced World Knowledge (The Standout Feature)
16:31 — Punch the Baby Monkey Test
17:38 — Rollout & Availability (141 New Countries)
18:18 — Seedance 2.0 Drops in CapCut
19:46 — Seedance 2.0 Weights Leaked?
20:32 — Outro
#NanoBanana2 #Gemini3Flash #Genspark #WorkWithGenspark #GoogleAI #AIImageGeneration #Seedance2 #AIArt #GenerativeAI #TheoreticallyMedia #AITools #Gemini
Conectar
