Tjay

Scale your business with

[Tjay @type_tjay]

https://t.co/ZgwTmupHyd

[Tjay @type_tjay]

@rareZuhair @macyagilliam @aridutilh He the best we got

[Tjay @type_tjay]

Superhuman is a hot steaming pile of shit.

Used to be a status signaling device in the valley for the past decade.

Low test move in first place.

[dax @thdxr]

i canceled superhuman. i only use email from opencode now

crazy times

[Tjay @type_tjay]

We already got that covered, not just a concept @peggy_wang

[Ramin Nasibov @RaminNasibov]

Made an app concept that picks a fair spot to meet for groups of friends.

[Tjay @type_tjay]

Why T3 code be sending my data to Beijing - @theo honeypot confirmed?

[Theo - t3.gg @theo]

You guys have no idea how hard I go every day to try and keep data transferred as low as humanly (and agentically) possible in T3 Code

[Tjay @type_tjay]

Fal Supremacist - will die on that hill

[Tjay @type_tjay]

Cost Optimization Guide for Video Models - from someone who spends 70K/month on Tokens

1) What all providers did I test?
- Tested Higgsfield, Runware, KIE, Fal, BytePlus, Runway, Sume, Sunra, Replicate

2) Which is the cheapest?
- Depends on the model you are trying to use.
- Kei is cheapest for all Google models - especially image generation models (I believe they are doing some illegal stuff - by buying retail accounts at 200$ each and then serving API requests through browser automations).
- Btw you can do it yourself too: https://t.co/ijErEKPT3x
- All Byteplus models (Seedance/Seedream) are cheapest on Byteplus, and no other providers support their "Digital Asset Library" - which is the greatest marketing play ever.
- All generations that require consistent characters across generations can be done with references - and Byteplus does not let you generate content with references if they are not part of their digital asset library. Which they charge $1400/month for.
- Higgsfield Soul 2.0 is only available on Higgsfield - and is also the cheapest there - soul is unparalleled quality (not to mention dirt cheap) when it comes to UGC style content.
- Sume is misleading advertisement - they compare their cost with lower resolution generation and call themselves cheaper.

3) Which one is the most reliable?
- Fal is king - I have tested them all - nothing even closely compares to Fal's reliability. In our testing - attempting to make 90s pixar animation style videos with seedance 2.5 - it takes about 17 minutes e2e.
- The same task with others providers takes upto 45 minutes at times - and upon investigation - the majority latency comes from time to request completion on the 3providers end - queues are maxxed out - and they dont have enough capacity - and hence you end up waiting in the queue for all your requests.
- Also the issue of dropped requests is especially high on runware, and sume. These are requests that are stuck in processing forever.
- Conspiracy theory - in our testing - output quality also varies from providers to provider - fal is supreme - runware, replicate is slightly worse - this is empirical but for some reason we are noticing downscaled resolution of the output for some providers.

4) How to Min-Max Token usage without compromising quality
- Generate videos at 480p and upscale with Topaz - it will be 4x cheaper
- Instead of using Seedance 2.5 use Wan 3.0 (64% cheaper) for Human Avatar Workflows - equal quality as seedance for human workflows.
- Instead of using Seedance 2.5 use Kling 4 (34% cheaper) for Animated Workflows - comparable quality as seedance for animated workflows.
- Reserve Seedance 2.5 for only talking head workflows

[Tjay @type_tjay]

> Its ridiculously slow
> Does not respect toolcall output structure
> Inferior in coding to every frontier model
> Not even close to the pareto frontier as far as costs are concerned

> Good at image and video understanding
> Decent at copy writing

Follow for more vibe reviews

[Lumina @LuminaBench]

Gemini 4 Carbon is ridiculously efficient

It barely uses any tokens, it's super cheap and incredibly fast for a model not far off Fable 5.2 in capabilities

One hell of a drop

[Tjay @type_tjay]

Let your agent loose on this one.

Most asymmetrical bet you can make right now for distribution and growth.

[OpenAI @OpenAI]

Now in beta: Businesses in ChatGPT.

Claim your business, manage your information, and get insights on how people engage with you.

Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.

https://t.co/N2GEZX0mFT

[Tjay @type_tjay]

You're gonna carry that weight

[Tjay @type_tjay]

I am never leaving this app: https://t.co/wPgDNBehPS

[Josh Barzon @JoshuaBarzon]

5,000 years of world history on a single timeline. Zoom in. You could spend 30 minutes exploring this.

[Tjay @type_tjay]

@VibeEverything Can you share some of the underlying engine of how you are doing this, also thoughts on open sourcing? I would love to contribute and have some ideas in mind.

[Tjay @type_tjay]

@newomp4 Fax

[Tjay @type_tjay]

Bro doesn’t know about yt-dlp

[newo @newomp4]

I was fed up with "Yt to mp4" websites so i made my own solution

It's better than anything else out there
(fingers crossed haha)

Plus its 100% free (and will be open sourced)

> Downloads anything (even Spotify, X, +1,200 others)
> Supports 4k, 6k, 8k + ProRes and other codecs
> Mac native (based around yt-dlp)
> Lets you clip sections of a full video
> Supports bulk downloading (very useful)
> Under 100mb to download and is very fast
> Drag downloads straight from the app to your timeline

plus so, so, so much more.

I have been using this tool nearly every day for the past couple months and it has streamlined my workflow a ton, hope others find it helpful as well

I will also probably get around to making premiere/ae/davinci plugins for it too.

First though, I want to really perfect the app experience so would love to hear any and all feedback!

[Tjay @type_tjay]

For anyone working with UGC slideshow for TT/IG

Scraping Pinterest images - its better to use warmed up accounts instead of using anonymous scraping.

Pinterest returns worse images for non-logged in users, easily a 2x improvement in quality

[Tjay @type_tjay]

2 min silence for people paying full price for Video Models

[Tjay @type_tjay]

Cost Optimization Guide for Video Models - from someone who spends 70K/month on Tokens

1) What all providers did I test?
- Tested Higgsfield, Runware, KIE, Fal, BytePlus, Runway, Sume, Sunra, Replicate

2) Which is the cheapest?
- Depends on the model you are trying to use.
- Kei is cheapest for all Google models - especially image generation models (I believe they are doing some illegal stuff - by buying retail accounts at 200$ each and then serving API requests through browser automations).
- Btw you can do it yourself too: https://t.co/ijErEKPT3x
- All Byteplus models (Seedance/Seedream) are cheapest on Byteplus, and no other providers support their "Digital Asset Library" - which is the greatest marketing play ever.
- All generations that require consistent characters across generations can be done with references - and Byteplus does not let you generate content with references if they are not part of their digital asset library. Which they charge $1400/month for.
- Higgsfield Soul 2.0 is only available on Higgsfield - and is also the cheapest there - soul is unparalleled quality (not to mention dirt cheap) when it comes to UGC style content.
- Sume is misleading advertisement - they compare their cost with lower resolution generation and call themselves cheaper.

3) Which one is the most reliable?
- Fal is king - I have tested them all - nothing even closely compares to Fal's reliability. In our testing - attempting to make 90s pixar animation style videos with seedance 2.5 - it takes about 17 minutes e2e.
- The same task with others providers takes upto 45 minutes at times - and upon investigation - the majority latency comes from time to request completion on the 3providers end - queues are maxxed out - and they dont have enough capacity - and hence you end up waiting in the queue for all your requests.
- Also the issue of dropped requests is especially high on runware, and sume. These are requests that are stuck in processing forever.
- Conspiracy theory - in our testing - output quality also varies from providers to provider - fal is supreme - runware, replicate is slightly worse - this is empirical but for some reason we are noticing downscaled resolution of the output for some providers.

4) How to Min-Max Token usage without compromising quality
- Generate videos at 480p and upscale with Topaz - it will be 4x cheaper
- Instead of using Seedance 2.5 use Wan 3.0 (64% cheaper) for Human Avatar Workflows - equal quality as seedance for human workflows.
- Instead of using Seedance 2.5 use Kling 4 (34% cheaper) for Animated Workflows - comparable quality as seedance for animated workflows.
- Reserve Seedance 2.5 for only talking head workflows

[Tjay @type_tjay]

@LucasC43966 Dm

[Tjay @type_tjay]

Figured out synthID removal - the internet is officially dead

[Tjay @type_tjay]

How to de-slopify your X feed
> Mute following keywords: Killed, Changes Everything, Game Changer, 99% of People
> Ask your agent to block everyone on the slop list https://t.co/x2UoU2uwX7
> Ask your agent to review 500 posts these slop posters and learn meta patterns of slop
> Block without second thought

[Ariel @arieljrubin]

everyone needs to relax

[Tjay @type_tjay]

@narsagna

[Tjay @type_tjay]

@arlanmarat Need the codes homskie, not just the skill - that would be god sent.