👐 The main thing about GPT-5:
🔜 GPT-5 is an intelligent system — depending on the complexity of the request, the model chooses the depth of reasoning. All current models (from GPT-4o mini to o3 Pro) will be replaced by the corresponding branch of GPT-5.
🔜 Available to everyone today, including free users — if the limit is exceeded, they will be transferred to GPT-5 mini.
🔜 The best model for coding on the market. The Pro version scores 100% in the AIME 2025 mathematical benchmark.
🔜 Even the basic model hallucinates significantly less than GPT-4o and o3.
🔜 The model is much less inclined to agree with the user and flatter him.
🔜 GPT-5 was classified as a high-risk model for hazards in biology and chemistry — appropriate measures have been taken. At the same time, the model is 65% safer than its predecessors, and more than 9,000 hours were spent on testing.
More details are in the model's system map.
🔜 GPT-5 is an intelligent system — depending on the complexity of the request, the model chooses the depth of reasoning. All current models (from GPT-4o mini to o3 Pro) will be replaced by the corresponding branch of GPT-5.
🔜 Available to everyone today, including free users — if the limit is exceeded, they will be transferred to GPT-5 mini.
🔜 The best model for coding on the market. The Pro version scores 100% in the AIME 2025 mathematical benchmark.
🔜 Even the basic model hallucinates significantly less than GPT-4o and o3.
🔜 The model is much less inclined to agree with the user and flatter him.
🔜 GPT-5 was classified as a high-risk model for hazards in biology and chemistry — appropriate measures have been taken. At the same time, the model is 65% safer than its predecessors, and more than 9,000 hours were spent on testing.
More details are in the model's system map.
❤218👍211🎉211🔥192
This media is not supported in your browser
VIEW IN TELEGRAM
Midjourney has introduced HD Mode
Only available for Pro and Mega subscription plans.
Costs 3.2 times more credits.
Because… lots of pixels.
Only available for Pro and Mega subscription plans.
Costs 3.2 times more credits.
Because… lots of pixels.
❤229👍198🎉197🔥190
Media is too big
VIEW IN TELEGRAM
Minimax Speech 2.5
Minimax Text-to-Speech (TTS) Generator:
Compared to the Speech 02 version released in May, Speech 2.5 features three new improvements:
• Higher speech expressiveness in multiple languages,
• More realistic voice reproduction,
• Broad coverage of 40 languages.
https://www.minimax.io/audio
Minimax Text-to-Speech (TTS) Generator:
Compared to the Speech 02 version released in May, Speech 2.5 features three new improvements:
• Higher speech expressiveness in multiple languages,
• More realistic voice reproduction,
• Broad coverage of 40 languages.
https://www.minimax.io/audio
❤287🔥264👍258🎉257
Grok 4 is now free to use! 🎉
Cranks out and edits images like a pro.
Update: Free users get 5 requests every 12 hours.
Cranks out and edits images like a pro.
Update: Free users get 5 requests every 12 hours.
👍277🎉277❤273🔥270
This media is not supported in your browser
VIEW IN TELEGRAM
A cool product feature.
A long press on an image turns it into a video.
This is the iOS app for Grok. It’s coming to Android soon.
A long press on an image turns it into a video.
This is the iOS app for Grok. It’s coming to Android soon.
❤253🎉244👍234🔥229
This media is not supported in your browser
VIEW IN TELEGRAM
SkyWorks: they’ve released Matrix-3D, a 3D world generator that works through a combination of video generation and 3D reconstruction. It’s sort of a response to Hunyuan World 1 from Tencent, the Odyssey project, and Google’s recently announced Genie 3.
You enter a prompt or upload an image, and you get either a video panorama or a 3D scene to explore. The catch is, it seems you’ll have to move through it by setting a trajectory. World 1, judging by the demos, supported a gamepad.
Generation can be done at resolutions of 960 × 480 or 1440 × 720. On a single A800 with 40 GB VRAM, rendering 720p takes about an hour...
You enter a prompt or upload an image, and you get either a video panorama or a 3D scene to explore. The catch is, it seems you’ll have to move through it by setting a trajectory. World 1, judging by the demos, supported a gamepad.
Generation can be done at resolutions of 960 × 480 or 1440 × 720. On a single A800 with 40 GB VRAM, rendering 720p takes about an hour...
❤258🎉252🔥242👍224
This media is not supported in your browser
VIEW IN TELEGRAM
Suno Studio!
Now this is intriguing.
At least because of this:
Multi-track creation. Export to MIDI.
If this is real multitrack — where one track = one instrument — then it’s an absolute bomb that will just wipe out all competitors.
Now this is intriguing.
At least because of this:
Multi-track creation. Export to MIDI.
If this is real multitrack — where one track = one instrument — then it’s an absolute bomb that will just wipe out all competitors.
🎉79❤69👍61🔥56
Imagen 4 has been rolled out in Google AI Studio and via the API. And it’s damn good — especially Ultra.
AI Studio has limits.
And via API, the prices are: Fast ($0.02/image) Standard ($0.04/image) Ultra ($0.06/image)
AI Studio has limits.
And via API, the prices are: Fast ($0.02/image) Standard ($0.04/image) Ultra ($0.06/image)
👍692🎉677❤659🔥637
Higgsfield keeps dropping viral features.
Product-to-Video is basically Flux Context, but for video.
Something similar was done by Pika and Runway, but Higgs’ cherry-picks look really polished.
Product-to-Video is basically Flux Context, but for video.
Something similar was done by Pika and Runway, but Higgs’ cherry-picks look really polished.
👍409🔥408❤402🎉399
Qwen Edit vs Nano Banana vs Flux Kontext Pro & Flux Kontext Dev
Prompt: Turn the motorcycle pink and put it against the backdrop of a big city at night, glowing with huge neon signs.
Banano really delivers! 🚀
Prompt: Turn the motorcycle pink and put it against the backdrop of a big city at night, glowing with huge neon signs.
Banano really delivers! 🚀
👍632🎉594❤593🔥530
Virtual fitting room on VideoX-Fun / Wan2.1-I2V-14B
Qwen2.5-VL-7B-Instruct is used for clothing description.
And under the hood, there’s also OpenPose, DensePose, and more.
If anyone wanted to fine-tune WAN 2.1 for virtual try-on — here it is.
https://vivocameraresearch.github.io/magictryon/
Qwen2.5-VL-7B-Instruct is used for clothing description.
And under the hood, there’s also OpenPose, DensePose, and more.
If anyone wanted to fine-tune WAN 2.1 for virtual try-on — here it is.
https://vivocameraresearch.github.io/magictryon/
❤873🎉820🔥794👍773