Pinned Tweet
I animated the Star Wars Force Awakens Trailer using AI š (š on).
This is a culmination of 4 months of work, tuning 1000s of parameter combinations & running 100s of A/B tests.
This is just a glimpse into what is to come in the future & represents where we are at with open-source Vid2Vid AI.
This is just the first piece of the puzzle solved. Next, I'll be moving to creating my own AI 3D worlds with my own characters & storylines.
Follow along as I post on my progress & how you can do the same.
Big thanks to all the AI creators on X who have helped & inspired me to create & share either personally or through their content, prompting & building tips & tricks.
I would highly recommend you follow themš
@8bit_e
@bri_guy_ai
@olafsells
@iamneubert
@nickfloats
@_Borriss_
@bennash
@hugovntr
@CharaspowerAI
@ProperPrompter
@dr_formalyst
@HBCoop_
@Artedeingenio
@CorridorDigital
@CrewCorridor
@ciguleva
HAYDN retweeted
If you need consistent, reoccurring scenes for your story it's never been easier.
A thread on how I go about blocking out a 3D scene that I can then use for any reoccurring environment shots.
As always - if this helps you please kindly like and share! š©µ
š§µBELOW š§µ
Iām dreading Google Next
Iāve heard from their team they will be announcing some crazy AI releases.
Which only means one thingā¦
This type of post x1000 on my feed:
āChatgpt killerā¦ā
āKlingās deadā¦ā
Ergh, I need a bot that auto blocks these guys
1
98
How to be more effective with AI:
When we write prompts, we assume that the LLMs can interpret the vision we have in our head.
This is what typically leads to poor outputs and not getting what you want (especially when coding).
The same is true when we receive a bad brief:
Client: 'We want to drive excitement for the brand'
Me: 'Ok, how about this...'
Client: 'That's not how we define excitement...'
The Solution:
Input your project or idea and ask the AI to run a brainstorming session with you, question by question, to delve into the topic in more detail.
The Result:
A complete and detailed project brief, which removes ambiguity.
The bonusĀ is that you can keep referring to this brief and stop wasting time explaining your project.
85
How to build an app faster than boiling the kettle:
1. Download the new @Replit mobile app
2. Prompt with what you want to build
3. Let the AI elves work their magic
4. Press Deploy
Done
The best partā¦.itās free.
Hereās a game baked in less than 60s.
Iām off for my tea now. āļø
191
HAYDN retweeted
This is HUGE
The AI App store is here
Ask anything you want to do with AI
With ~400k Apps, this is the best place to find the AI apps you need
developers can build apps, users can try them out and find new apps with AI search
What impact will Deepseek r1 have on consumers?
Below is our detailed, research-driven report on the impact that DeepSeek R1 is expected to have on consumers, integrating recent insights from social conversations and industry analyses:
⢠Market Trends
- DeepSeek R1 represents a breakthrough in cost-efficient AI development. Released on January 20, 2025, with a development budget of only about $6 millionācompared to the billions typically spent by U.S. companiesāits emergence signals a shift toward more resource-efficient, open-source AI models.
[Cyrus Janssen, YouTube, 2025-01-26, "China's AI Breakthrough with DeepSeek" ā youtube.com/watch?v=OC2J-0vlā¦]
- The open-source nature of DeepSeek R1 is creating buzz across social media from influencers and tech voices alike. This model is being positioned as a ādemocratiserā of AI technology, making advanced features more accessible to a broader range of developers and, ultimately, consumers.
[CNBC, YouTube, 2025-01-24, "How Chinaās New AI Model DeepSeek Is Threatening U.S. Dominance" ā youtube.com/watch?v=WEBiebbeā¦]
- With mention counts reaching high levels in our recent data clusters and discussions spread across major platforms such as TikTok and YouTube, the conversation around DeepSeek R1 reflects both excitement over its technological promise and a strategic rethinking of AIās cost structure in consumer applications.
⢠Customer Preferences
- Consumers increasingly value personalised, efficient, and cost-effective experiences. With DeepSeek R1ās reduced development costs and open access, there is a strong expectation that future consumer devices and applications will integrate high-performance AI features without driving up price points.
- Early discussions indicate that tech enthusiasts appreciate the potential for more affordable AI in everyday productsāwhether that means smarter virtual assistants, improved photo editing tools, or more intuitive user interfaces.
- There remains, however, a degree of caution on data privacy and reliability, as consumers are mindful of how open-source models manage and safeguard personal information.
⢠Competitive Landscape
- DeepSeek R1 is challenging conventional U.S. AI models developed by giants like OpenAI and Google. The efficiency and open-source approach of DeepSeek R1 pressure established firms to reconsider their high-cost, proprietary development strategies.
- This emergence is driving competitive realignment; U.S. companies may soon face increased pressure to innovate with lower-cost solutions if they wish to maintain market share in devices and applications that consumers ultimately interact with.
- In a broader global context, this development also contributes to rising geopolitical tensions as U.S. tech leaders reassess investments in AI infrastructure.
⢠Potential Growth Areas
- Open-Source AI Integration: The cost-effectiveness and modularity of DeepSeek R1 can spur the development of AI-enhanced consumer productsāfrom smartphones and smart appliances to digital assistantsāthat are more affordable yet powerfully capable.
- Enhanced Consumer Applications: Products built on efficient AI platforms can offer more personalised recommendations, improved natural language processing, and real-time contextual assistance.
- Developer-Driven Innovations: As the open-source model allows customisation and rapid prototyping, startups and smaller developers have a greater opportunity to introduce novel AI applications that directly address consumer needs.
- Ethical and Data Privacy Measures: With consumers increasingly concerned about how their data is used, companies that can combine advanced AI functionality with transparent data practices and strong privacy safeguards will likely resonate well.
⢠Insight Synthesis
- Our research indicates that DeepSeek R1ās breakthrough in efficiency and open-source accessibility is set to lower the barriers for integrating high-performance AI into consumer products. This shift means that everyday devicesāfrom smartphones to home assistantsācould soon offer sophisticated AI features without a corresponding increase in price.
- The democratisation of AI technology will likely spur innovation, as more companies (especially smaller players) can build and tailor applications to consumer needs, resulting in a richer, more competitive market.
- However, while consumers stand to benefit from improved functionality and lower costs, there is also a clear need to address lingering concerns over data security and ethical AI practicesāfactors that can shape overall adoption and trust.
⢠Actionable Recommendations
- For Consumer Electronics Brands:
⢠Accelerate research and integration of open-source AI models like DeepSeek R1 into product development to boost functionality while controlling costs.
⢠Highlight benefits such as advanced personalization, real-time assistance, and improved performance in marketing materials to create appealing value propositions for tech-savvy consumers.
- For Software/Application Developers:
⢠Leverage DeepSeek R1ās open-source framework to swiftly prototype new consumer apps or enhance existing ones with cost-effective yet powerful AI features.
⢠Emphasize transparency in data management and robust privacy protections in messaging to build user trust.
- For Marketers and Strategists:
⢠Craft narratives around the democratization of AI and the resulting affordability, showcasing how deep integration of effective AI features improves user experience without premium pricing.
⢠Monitor ongoing discussions on platforms like TikTok and YouTube to adapt messaging that directly addresses consumer privacy concerns and overall AI reliability.
- For Policy and Regulatory Stakeholders:
⢠Increase focus on developing guidelines that ensure ethical usage and data privacy in open-source AI models to protect consumers as more products integrate such technologies.
In conclusion, our research suggests that DeepSeek R1 will empower consumers with more advanced, accessible, and cost-effective AI-driven experiences while pushing industry players to innovate more efficiently in a rapidly shifting competitive landscape.
I
Sources:
⢠[Cyrus Janssen, YouTube, 2025-01-26, "China's AI Breakthrough with DeepSeek" ā youtube.com/watch?v=OC2J-0vlā¦]
⢠[CNBC, YouTube, 2025-01-24, "How Chinaās New AI Model DeepSeek Is Threatening U.S. Dominance" ā youtube.com/watch?v=WEBiebbeā¦]
60
Why read the news and updates youāre interested in, when you can listen on the go?
Introducing Daily Listen: an experimental AI-powered audio show giving you short, timely updates about things YOU care about, right in your Google app. š§Try it & tell us what you think!šgoo.gle/4hxKBml
67
Awesome video comparison of the various ai models.
Veo 2 š
š¬ Video Model Comparison: Text to video
⢠Google Veo 2
⢠OpenAI Sora
⢠Hunyuan Video
⢠Pika 2.1
⢠Alibaba Wanx 2.1
⢠Runway Gen-3
⢠Kling 1.6
⢠Luma Ray2
⢠Hailuo T2V-01
I used the same prompt on each model for 2 generations and chose the best results below.
Prompt: A majestic white tiger prowling silently through a dense jungle, sunlight piercing through the foliage to illuminate its fur; shot with a telephoto lens for depth, using soft natural lighting and creating a mysterious yet tranquil atmosphere.
Prompt credit: @LudovicCreator
Overall, the results are impressive for each one. We have many choices for AI video!
I could have tested more models, but this visual is crowded enough. Drop your results in the comments š
1
2
152
DeepSeek is a DeepFake
Recent developments suggest that DeepSeek's latest breakthrough challenges the long-held belief that cutting-edge AI must rely on extensive hardware infrastructures.
According to reports, DeepSeek trained their large language model (LLM) for approximately $6 million, a fraction of the billions traditionally spent by Western tech giants (reuters.com).
The financial markets have certainly felt the impact: Nvidia's stock reportedly plummeted by nearly $600 billion on Monday, following claims that DeepSeek managed its training with just 2,000 industrial-scale GPUs instead of the 25,000 it took for GPT-4 (genspark.com).
DeepSeek credits much of this success to its frontier "software optimisation." (reuters.com, 2025)
However, not everyone is convinced by the official narrative.
OpenAI alleges that DeepSeek used a technique called "distillation"āessentially using ChatGPT's own outputs as training dataāto build its AI.
Distillation can sharply reduce hardware needs and training time but raises crucial questions around intellectual property and data ethics.
If you prompt DeepSeek, it appears to reveal that it was trained on OpenAI's data (see image of my DeepSeek convo).
Why This Matters for Western AI companies:
Cost Disruption
If a start-up can train an advanced model at a fraction of the cost, market entry barriers could drop significantly. Established players might need to justify ever-growing R&D budgets to stakeholders, especially as savvy contenders prove you can do more with less.
Legal and Ethical Challenges
Distilling proprietary data from competitors' AI models treads a grey line. Future lawsuits or policy frameworks could shape how (and whether) data from existing AI is used to bootstrap new models.
But if OpenAI trained its model on data that it didn't own (claimed), then does it have a leg to stand on?
Speed to Market
Faster (and cheaper) training translates to agile iteration and product deployment.
Data Protection and IP
The possibility that DeepSeek used ChatGPT content without permission underlines a larger conversation on data governance.
Implications for Global Competition
Should DeepSeek's approach prove both lawful and commercially viable, Western tech firms might see more entrants around the world adopting similar methodologies.
Sources
lnkd.in/epkqeXHv
lnkd.in/euCZDRqX
1
84
My side hustleās X content runs on autopilot.
Using a linear sequence of API calls and AI agents, every piece of content, including infographics, is 100% AI-generated.
The best part is that is can be adjusted to fit the needs of any business.
If you want an in depth explainer, like this message.
Feeling mentally sluggish? š§ āØ
Weāve distilled advanced insights from @hubermanlab into a convenient routine card to boost your brainās health, longevity, and performance!
Your personalised brain optimisation toolkit is here ā¬ļø
1. Resistance Training:
- Engage in 2-3 weekly sessions, focusing on compound exercises that activate multiple muscle groups.
- Such exercises increase norepinephrine and dopamine, enhancing cognitive focus and neuroplasticity.
2. Cardiovascular Training:
- Perform HIIT 2-3 times a week for 20-30 minutes to increase cerebral blood flow.
- Opt for cycles like 4x4 or 6x6 depending on your fitness level.
- Include 150 minutes of moderate-intensity cardio weekly, ideally in the morning for optimal neurotrophic factor release.
- Add a weekly long, slow distance session to improve cardiovascular efficiency and cerebral blood flow.
3. Short Intense Sprints:
- Use these to enhance brain function and cognitive flexibility through catecholamine release.
4. Bone Loading & Jumping:
- Incorporate jumping exercises to release osteocalcin, promoting neurogenesis and improving coordination.
5. Timing:
- Morning workouts enhance cognitive performance, regulate circadian rhythms, and elevate mood.
- Consistency is keyāintegrate various exercises weekly to maintain neurotrophic factor levels.
6. Supplementation:
- Omega-3 fatty acids (1-2 grams daily) boost BDNF and reduce inflammation.
Check out the video here: youtube.com/watch?v=J2oqi41dā¦
96
Coding with AI tip:
Stop blindly copying and pasting chunks of code.
Start manually typing out the code and ask what each line of code is doing.
It may take longer, but you can troubleshoot it far faster when it fails (& it will).
3
92
Whatsapp messages are gonna pop off cause of this
Video & image prompting is here with Suno Scenes! Make a Scene on our mobile app at suno.com/ios
Whether it's a stunning landscape, a funny meme, a candid selfie, or a quick video of your daily adventures, Suno Scenes helps you create unique songs from your favorite photos and videos - all from your phone.
Download the app now & tag us in your content: #sunoscenes and @suno_ai_ on X and @.sunomusic on IG and TikTok.
1
131
How to scale your micro app with a simple tool:
I use @airtable as my database for proof of concepts.
It's quick, cheap, & user-friendly.
But, it's not designed to handle 1000s of requests.
Eventually, you need to switch to a PostgreSQL database.
With this tool, you can automatically transform your Airtable base into a PostgresSQL table.
Removing a ton of manual work
Enjoy āļø
uibakery.io/move-airtable-toā¦
1
79
How to build AI apps in a weekend:
I build a lot, but struggle to launch quickly.
@nutlope solves the issue with simplicity, GPT, & @vercel.
Key takeaway:
āIf you canāt describe it in 5 words then you wonāt build it in a weekendā
Gave a talk on how build AI apps in a weekend that go viral and scale to millions last year. AMA!
Thanks for having me @aiDotEngineer!
118
How to Instantly Turn Your 3D Models into Emotional, Talking Characters...with just Audio & AI:
I find that the reason why it's so hard to buy into AI-created characters is that they lack emotion...and don't blink.
(additional fingers don't help too)
However, I've found a quick solution from NVIDIA that can help to solve this issue for free.
Using their free Audio2Face program, all you need to do is:
1. Build/find/generate a 3D model
2. Record dialogue audio (You could use OpenAI's new real-time voice models, which can now emote)
3. Import your model into Audio2Face
4. Place markers on the model to indicate where facial features are
5. You can now change the emotion of your model to be happy/sad/angry/tired (the options are endless), which transforms the face in real time!
6. Upload your dialogue audio
7. Press play and watch your character come to life
In the example below, I have just used a static character.
However, you can import the outputs into Unreal Engine & place them on moving subjects.
Note: Audio2Face only works with Nvidia GPUs
83
Interesting how some posts perform a lot better on LinkedIn
Same post on X in quoted tweet.
How to control the outputs of AI video generation:
The key is to combine traditional tools with AI.
Tools like @UnrealEngine make this very easy.
In a day, I learned to create a driving game in UE5, filmed a sequence & used video style transfer with @runwayml
Output: š½
69
Correction to the below:
I initially misinterpreted the win rate for @AIatMeta ās new video gen model,
Movie Gen.
Upon review, Movie Gen outperformed Runway Gen3, LumaLabs, and OpenAI Sora across most metrics but had mixed results when compared to Kling1.5.
Big AI Video Gen release from @AIatMeta
In their paper, they compared publicly generated videos from other video competitors like @runwayml & @Kling_ai.
Meta's new model came 2nd, with @LumaLabsAI coming out on top in overall quality š„
1
121
How to control the outputs of AI video generation:
The key is to combine traditional tools with AI.
Tools like @UnrealEngine make this very easy.
In a day, I learned to create a driving game in UE5, filmed a sequence & used video style transfer with @runwayml
Output: š½
If you're interested in learning more, I would recommend checking out @MartinNebelong
He talks a lot about this field & is a great learning resource
1
45













