GPT-Image-2
Three new models appeared in the arena and then disappeared: maskingtape, packingtape, gaffertape.
Most likely, this is a new version of OpenAI’s image generator.
The models are extremely strong in world knowledge (just look at the anatomy) and also incredibly good at rendering very fine text (YouTube screenshots and code were generated).
It looks like a breakthrough for illustrative graphics.
Importantly, under the hood it’s not the old 4o, but one of the new models. I wonder which one it is?
Three new models appeared in the arena and then disappeared: maskingtape, packingtape, gaffertape.
Most likely, this is a new version of OpenAI’s image generator.
The models are extremely strong in world knowledge (just look at the anatomy) and also incredibly good at rendering very fine text (YouTube screenshots and code were generated).
It looks like a breakthrough for illustrative graphics.
Importantly, under the hood it’s not the old 4o, but one of the new models. I wonder which one it is?
🎉1.11K🔥1.1K❤1.09K👍1.06K
We are COOKED! gpt-image-1.5 vs gpt-image-2 it's so over!
OpenAI shipped again!
👇👇👇
Try all best models in VORTEX AI platform
https://vortex.channel/create.php
OpenAI shipped again!
👇👇👇
Try all best models in VORTEX AI platform
https://vortex.channel/create.php
👍459🔥136🎉130❤113
This media is not supported in your browser
VIEW IN TELEGRAM
Sync-3 is positioning itself as having the best lip sync on the market.
First, the good: you can upload your own voices and audio files, and the system will process them and sync the lips. The key point is that the model is language-agnostic — it doesn’t care what language the input audio is in.
Now, the downsides: It’s insanely expensive. If you use it via Fal, it costs about $8 per minute: https://fal.ai/models/fal-ai/sync-lipsync/v3
If you use it through their own website, the pricing model is pretty unclear: there’s a subscription PLUS usage-based charges, but in the end it comes out to roughly the same as Fal: https://sync.so/sync-3
On Twitter, they admit their target segment is B2B and enterprise, and that an ad agency will gladly pay $8 per minute for localization without blinking. But in that case, their “Hobby” plan looks odd — it’s way too expensive for casual experimentation at home.
First, the good: you can upload your own voices and audio files, and the system will process them and sync the lips. The key point is that the model is language-agnostic — it doesn’t care what language the input audio is in.
Now, the downsides: It’s insanely expensive. If you use it via Fal, it costs about $8 per minute: https://fal.ai/models/fal-ai/sync-lipsync/v3
If you use it through their own website, the pricing model is pretty unclear: there’s a subscription PLUS usage-based charges, but in the end it comes out to roughly the same as Fal: https://sync.so/sync-3
On Twitter, they admit their target segment is B2B and enterprise, and that an ad agency will gladly pay $8 per minute for localization without blinking. But in that case, their “Hobby” plan looks odd — it’s way too expensive for casual experimentation at home.
👍122🎉117❤114🔥113
This media is not supported in your browser
VIEW IN TELEGRAM
Guys, this is big. Seedance 2.0 just landed on Vortex with no limitations, from any country, and you can now create longer, bigger, cinematic AI videos right now.
Try now on https://vortex.channel/
Try now on https://vortex.channel/
👍364🎉118❤113🔥113
Media is too big
VIEW IN TELEGRAM
MiniMax Music 2.6
Suno is still far ahead, especially in terms of functionality.
What’s new in version 2.6: → First audio file in less than 20 seconds. → Improved low-to-mid frequency response: denser bass for house, trap, and drum and bass. → Style transfer and remixing.
The 14-day free global beta testing starts today (limit of 500 songs per day).
Suno is still far ahead, especially in terms of functionality.
What’s new in version 2.6: → First audio file in less than 20 seconds. → Improved low-to-mid frequency response: denser bass for house, trap, and drum and bass. → Style transfer and remixing.
The 14-day free global beta testing starts today (limit of 500 songs per day).
👍357❤132🔥118🎉112
This media is not supported in your browser
VIEW IN TELEGRAM
Unitree H1: 10 meters per second
Usain Bolt can still outrun robots—for now. He hit 10.44 m/s in the 100-meter sprint.
You’d better start getting used to being polite to them—both in chat and on the street. There’s nowhere left to run 😃
Usain Bolt can still outrun robots—for now. He hit 10.44 m/s in the 100-meter sprint.
You’d better start getting used to being polite to them—both in chat and on the street. There’s nowhere left to run 😃
👍501🎉150❤143🔥118
This media is not supported in your browser
VIEW IN TELEGRAM
Generation of real 3D worlds
Hunyuan is teasing a new world generator.
Unlike Google’s Genie 3, they’re promising not just a video image, but actual 3D environments that can be integrated into games, for example.
Hunyuan is teasing a new world generator.
Unlike Google’s Genie 3, they’re promising not just a video image, but actual 3D environments that can be integrated into games, for example.
👍433🔥88❤84🎉83
GPT-Image-2 is starting to roll out in ChatGPT.
Input: 4 photos from a clothing catalog and the prompt: “street style shot of a woman in this outfit.”
By the way, if you want to check whether it’s been rolled out to you, just try asking for an image in 16:9 or 21:9. The old version only outputs 3:2.
Input: 4 photos from a clothing catalog and the prompt: “street style shot of a woman in this outfit.”
By the way, if you want to check whether it’s been rolled out to you, just try asking for an image in 16:9 or 21:9. The old version only outputs 3:2.
👍291❤85🔥83🎉81
Happy Oyster World Generator
If you didn’t receive an invite code, this is what it looks like from the inside.
If you didn’t receive an invite code, this is what it looks like from the inside.
👍1.4K❤404🔥379🎉364
It's OVER.
GPT Image 2 is out! It's insane at text now. Posters, labels, typography, all clean. Still somehow loses a street fight against counting the Rs in strawberry. It’s so OVER! Great at info graphics and small texts and thumbnails too. is it available to pro users? GPT Image cooks logos, signs, and clean typography now. Then strawberry shows up and the whole illusion collapses. It’s OVER.
GPT Image 2 is out! It's insane at text now. Posters, labels, typography, all clean. Still somehow loses a street fight against counting the Rs in strawberry. It’s so OVER! Great at info graphics and small texts and thumbnails too. is it available to pro users? GPT Image cooks logos, signs, and clean typography now. Then strawberry shows up and the whole illusion collapses. It’s OVER.
👍291🔥71🎉61❤52
Nano Banana killer is here? Exciting news - GPT-Image-2 by
@OpenAI
has claimed the #1 spot across all Image Arena leaderboards!
@OpenAI
has claimed the #1 spot across all Image Arena leaderboards!
👍271🎉73🔥69❤52
A bit more about ChatGPT Image 2.0
First, it’s now available via API (Fal, Higgsfield) and on Vortex: https://vortex.channel/
There’s a generation quality setting (High, Medium, Low). Default is High. Pricing depends on quality and resolution.
https://fal.ai/models/openai/gpt-image-2
About 4K — unclear. Officially it’s 2K, though some aggregators mention 4K. The funniest part: I generated a 1K image in ChatGPT and it goes:
— Want me to upscale it to 4K?
— Sure!
— Here you go!
— Wait, how?!
— I just ran some Python with Lanczos.
Second, the model beats Banana and everyone else on the Arena. No model has had such a lead before.
Third, it generates panoramas well. With 1:3 support, that’s useful. Banana can too, but here you don’t see seams.
Fourth, it can display time on analog clocks, though the hour hand can still be off.
Fifth, it now handles alphabets properly.
Meanwhile, Google I/O is coming soon—rumor has it Nano Banana 3 might drop there.
First, it’s now available via API (Fal, Higgsfield) and on Vortex: https://vortex.channel/
There’s a generation quality setting (High, Medium, Low). Default is High. Pricing depends on quality and resolution.
https://fal.ai/models/openai/gpt-image-2
About 4K — unclear. Officially it’s 2K, though some aggregators mention 4K. The funniest part: I generated a 1K image in ChatGPT and it goes:
— Want me to upscale it to 4K?
— Sure!
— Here you go!
— Wait, how?!
— I just ran some Python with Lanczos.
Second, the model beats Banana and everyone else on the Arena. No model has had such a lead before.
Third, it generates panoramas well. With 1:3 support, that’s useful. Banana can too, but here you don’t see seams.
Fourth, it can display time on analog clocks, though the hour hand can still be off.
Fifth, it now handles alphabets properly.
Meanwhile, Google I/O is coming soon—rumor has it Nano Banana 3 might drop there.
👍1.13K🔥183🎉160❤148