Seedream 5.0 PRO
If you’re tired of Banana’s recent glitches, here’s a new toy.
Seedream 5.0 Pro is focused on editing tools: select an area with a point, box, lasso, color mark, or sketch, then add/remove objects, change colors by HEX, replace materials, or turn sketches into finished designs.
The underrated part: layers — PNG with transparency. It can split an image into 10+ independent layers: text, background, characters, decorations, and even restore hidden background areas.
Also: lots of focus on infographics and text. Weirdly, native resolution seems capped at 2K.
Available via API on BytePlus and aggregators.
Price-wise, it looks about half the cost of Banana — but check for yourself.
I also made a Seedream 5.0 Pro vs GPT-Image-2 comparison.
The one with the bigger white car is Seedream. Both look great.
If you’re tired of Banana’s recent glitches, here’s a new toy.
Seedream 5.0 Pro is focused on editing tools: select an area with a point, box, lasso, color mark, or sketch, then add/remove objects, change colors by HEX, replace materials, or turn sketches into finished designs.
The underrated part: layers — PNG with transparency. It can split an image into 10+ independent layers: text, background, characters, decorations, and even restore hidden background areas.
Also: lots of focus on infographics and text. Weirdly, native resolution seems capped at 2K.
Available via API on BytePlus and aggregators.
Price-wise, it looks about half the cost of Banana — but check for yourself.
I also made a Seedream 5.0 Pro vs GPT-Image-2 comparison.
The one with the bigger white car is Seedream. Both look great.
🔥259👍251🎉240❤238
This is WILD.
GPT-5.6 by @OpenAI just dropped and the model war got messy fast.
Fable 5 still looks like the raw intelligence king, sitting at 60 on Artificial Analysis.
But GPT-5.6 Sol is right behind at 59, takes #1 in coding agents with 80, and does it at roughly one-third the cost.
Meanwhile Grok 4.5 @grok is the chaos pick. Not the smartest, but cheap, fast, efficient, and suddenly close enough to make everyone nervous.
So the race is no longer just “who is smartest?”
It’s Fable for raw brain. But too expensive!
GPT-5.6 for work and coding.
Grok 4.5 for price pressure and speed.
AI labs are not fighting for benchmarks anymore.
They’re fighting for who becomes the default worker.
GPT-5.6 by @OpenAI just dropped and the model war got messy fast.
Fable 5 still looks like the raw intelligence king, sitting at 60 on Artificial Analysis.
But GPT-5.6 Sol is right behind at 59, takes #1 in coding agents with 80, and does it at roughly one-third the cost.
Meanwhile Grok 4.5 @grok is the chaos pick. Not the smartest, but cheap, fast, efficient, and suddenly close enough to make everyone nervous.
So the race is no longer just “who is smartest?”
It’s Fable for raw brain. But too expensive!
GPT-5.6 for work and coding.
Grok 4.5 for price pressure and speed.
AI labs are not fighting for benchmarks anymore.
They’re fighting for who becomes the default worker.
🎉258❤238🔥238👍237
A few rumors about Seedance 2.5 pricing.
I’m deliberately calling them rumors because I can’t verify the information from the video. The Jimeng website only has a Seedance 2.5 banner, but no pricing details.
Roughly speaking, on a per-second basis, Seedance 2.5 appears to be about 1.5 times more expensive than 2.0 for 720p. That’s what the first row indicates.
There’s no data yet for 1080p or 4K.
For now, we’re looking at roughly $5 for 30 seconds of 720p video.
However, this is not API pricing. It’s based on some fairly complicated calculations involving subscription plans and credits.
Overall, I wouldn’t trust these figures completely, but I would use them as a rough estimate — around a 1.5x price increase.
I’m deliberately calling them rumors because I can’t verify the information from the video. The Jimeng website only has a Seedance 2.5 banner, but no pricing details.
Roughly speaking, on a per-second basis, Seedance 2.5 appears to be about 1.5 times more expensive than 2.0 for 720p. That’s what the first row indicates.
There’s no data yet for 1080p or 4K.
For now, we’re looking at roughly $5 for 30 seconds of 720p video.
However, this is not API pricing. It’s based on some fairly complicated calculations involving subscription plans and credits.
Overall, I wouldn’t trust these figures completely, but I would use them as a rough estimate — around a 1.5x price increase.
🎉181👍172🔥171❤146
Prompts for a Frozen Camera in Seedance 2.0
A single continuous handheld handycam shot in slow motion, drifting and weaving organically through a completely frozen party scene. Every person is suspended mid-motion, like a living diorama. The camera gently circles and floats through the space with subtle handheld shake, creating a natural, immersive perspective. The environment remains perfectly still while only the camera moves. One seamless, uninterrupted slow-motion take.
A single continuous handheld handycam shot in slow motion, drifting and weaving organically through a completely frozen party scene. Every person is suspended mid-motion, like a living diorama. The camera gently circles and floats through the space with subtle handheld shake, creating a natural, immersive perspective. The environment remains perfectly still while only the camera moves. One seamless, uninterrupted slow-motion take.
👍130🔥96🎉96❤94
Lucy 2.5 – Real-Time Video Compositing and Editing
The idea itself isn’t exactly new—we’ve already seen similar approaches in LivePortrait, InfiniteTalk, and, of course, Kling’s Motion Control.
You upload an image of a character, take a video of yourself (or a more talented actor) performing in front of the camera, and your character instantly mimics those movements. In the 3D world, this is called retargeting.
But here’s the exciting part:
Everything happens in real time.
Just load an image, turn on your webcam, and start streaming at 24–30 FPS with minimal latency. The developers claim near-zero latency, although in real-world conditions, once internet transmission is factored in, I’d expect around 1–2 seconds of delay.
The idea itself isn’t exactly new—we’ve already seen similar approaches in LivePortrait, InfiniteTalk, and, of course, Kling’s Motion Control.
You upload an image of a character, take a video of yourself (or a more talented actor) performing in front of the camera, and your character instantly mimics those movements. In the 3D world, this is called retargeting.
But here’s the exciting part:
Everything happens in real time.
Just load an image, turn on your webcam, and start streaming at 24–30 FPS with minimal latency. The developers claim near-zero latency, although in real-world conditions, once internet transmission is factored in, I’d expect around 1–2 seconds of delay.
👍306🎉296❤288🔥278
Alibaba Announces Qwen3.8 — a New Open-Weight Rival to Kimi K3
Alibaba has teased Qwen3.8, its new flagship model with 2.4 trillion parameters. It will launch soon as an open-weight model, but architecture, context length, and benchmark results have not yet been revealed.
The announcement appears to challenge Moonshot AI’s Kimi K3, a 2.8-trillion-parameter MoE model with native image and video understanding and a 1M-token context window.
Qwen3.8-Max-Preview is already available through:
• QwenCloud Token Plan
• Qoder
• QoderWork
It is not yet available in the free Qwen Studio chat.
Alibaba has teased Qwen3.8, its new flagship model with 2.4 trillion parameters. It will launch soon as an open-weight model, but architecture, context length, and benchmark results have not yet been revealed.
The announcement appears to challenge Moonshot AI’s Kimi K3, a 2.8-trillion-parameter MoE model with native image and video understanding and a 1M-token context window.
Qwen3.8-Max-Preview is already available through:
• QwenCloud Token Plan
• Qoder
• QoderWork
It is not yet available in the free Qwen Studio chat.
❤132👍117🎉96🔥94
Higgsfield has launched a new Community section.
There are quite a few insightful resources there.
You can browse completed projects (videos) and see not only the prompt, but also all the assets used, the total number of generations, the number of credits spent, and in some cases even the failed generations and how many there were. Many creators also provide detailed breakdowns of their entire workflow.
It’s a really useful way to get a sense of the resources and effort behind different projects.
They’ve also significantly expanded their new Academy section.
From a community-building and ecosystem perspective, they’re doing a great job.
https://higgsfield.ai/community
There are quite a few insightful resources there.
You can browse completed projects (videos) and see not only the prompt, but also all the assets used, the total number of generations, the number of credits spent, and in some cases even the failed generations and how many there were. Many creators also provide detailed breakdowns of their entire workflow.
It’s a really useful way to get a sense of the resources and effort behind different projects.
They’ve also significantly expanded their new Academy section.
From a community-building and ecosystem perspective, they’re doing a great job.
https://higgsfield.ai/community
🔥247🎉244❤243👍226
FLUX 3 Image + Video 👀
Black Forest Labs is turning FLUX into a unified multimodal model for images, video, audio, and text.
Key features:
• Videos up to 20 seconds with native audio
• Text-to-video, image-to-video, and video-to-video
• Video continuation with sound
• Character consistency across scenes
• Image generation and editing
FLUX 3 Video is already in limited Early Access. FLUX 3 Image, API access, open weights, and robotics tools are planned next.
No release date or pricing yet.
Black Forest Labs is turning FLUX into a unified multimodal model for images, video, audio, and text.
Key features:
• Videos up to 20 seconds with native audio
• Text-to-video, image-to-video, and video-to-video
• Video continuation with sound
• Character consistency across scenes
• Image generation and editing
FLUX 3 Video is already in limited Early Access. FLUX 3 Image, API access, open weights, and robotics tools are planned next.
No release date or pricing yet.
❤179🎉171🔥158👍151
This media is not supported in your browser
VIEW IN TELEGRAM
Flux 3 is pretty good!
One prompt. It edits everything by itself. 20 seconds. Great voiceover—and thankfully, no lip sync is needed here.
We were waiting for Sora 3, but got Flux the Third instead.
One prompt. It edits everything by itself. 20 seconds. Great voiceover—and thankfully, no lip sync is needed here.
We were waiting for Sora 3, but got Flux the Third instead.
🔥146👍130❤126🎉109
MiniMax H3
MiniMax has started rolling out Hailuo H3, the next generation of its video model. It's more than an upgrade to Hailuo 2.3, combining video generation, directing, and audio in one system. H3 supports text-to-video, image-to-video, multi-scene generation, character consistency from reference images, and prompt-based editing.
The only catch is that the official MiniMax website still lists Hailuo 2.3 as the public model, suggesting H3 is being released through partners and selected users.
H3 generates 5–15 second videos in native 2K at 24 FPS with synchronized audio. It accepts text prompts, a starting or ending frame, up to nine images, three video references, and three audio files. Multi-shot scenes and consistent characters are supported.
The motion quality looks impressive.
Official pricing hasn't been announced, but one statement claims:
"It uses the same amount of consumption as Seedance 2.0's 480p at 2K resolution."
If true, 2K generation in H3 costs about the same as 480p in Seedance 2.0.
MiniMax has started rolling out Hailuo H3, the next generation of its video model. It's more than an upgrade to Hailuo 2.3, combining video generation, directing, and audio in one system. H3 supports text-to-video, image-to-video, multi-scene generation, character consistency from reference images, and prompt-based editing.
The only catch is that the official MiniMax website still lists Hailuo 2.3 as the public model, suggesting H3 is being released through partners and selected users.
H3 generates 5–15 second videos in native 2K at 24 FPS with synchronized audio. It accepts text prompts, a starting or ending frame, up to nine images, three video references, and three audio files. Multi-shot scenes and consistent characters are supported.
The motion quality looks impressive.
Official pricing hasn't been announced, but one statement claims:
"It uses the same amount of consumption as Seedance 2.0's 480p at 2K resolution."
If true, 2K generation in H3 costs about the same as 480p in Seedance 2.0.
❤449🔥438🎉438👍422
Try New RIFTFALL Demo!
Sixty minutes. No level locks, no "first three maps then buy it." The whole game is unlocked — twenty arenas, six weapons, every boss that wants you dead. The only thing counting down is you.
Grab the Breacher. Grab the Vulcan, the Arc Rail, the Flamer. Kill a dragon. Kill a tripod the size of a cathedral. Ride a train. Get eaten by something in the dark and try again, faster.
When the clock hits zero, it stops. What you did with the hour is on you.
Hell was a cakewalk. This is the main course.
https://store.steampowered.com/app/4965490/RIFTFALL/
Sixty minutes. No level locks, no "first three maps then buy it." The whole game is unlocked — twenty arenas, six weapons, every boss that wants you dead. The only thing counting down is you.
Grab the Breacher. Grab the Vulcan, the Arc Rail, the Flamer. Kill a dragon. Kill a tripod the size of a cathedral. Ride a train. Get eaten by something in the dark and try again, faster.
When the clock hits zero, it stops. What you did with the hour is on you.
Hell was a cakewalk. This is the main course.
https://store.steampowered.com/app/4965490/RIFTFALL/
🔥100👍98🎉91❤90
This media is not supported in your browser
VIEW IN TELEGRAM
Grok Video 1.5 vs. Seedance 2.5.
👍235🎉235🔥216❤202
Media is too big
VIEW IN TELEGRAM
Guys! AI Video wars are on!
FLUX 3 Video is here, and this could be Seedance 2.5’s biggest rival yet. It generates up to 20-second videos in native 1080p with audio, multilingual dialogue, text-to-video, image-to-video, multiple frames, and video continuation. I made this Riftfall trailer with it, and honestly, the cinematic quality looks wild. FLUX 3 is available now through the API and open weights coming soon. Could this beat Seedance 2.5?
FLUX 3 Video is here, and this could be Seedance 2.5’s biggest rival yet. It generates up to 20-second videos in native 1080p with audio, multilingual dialogue, text-to-video, image-to-video, multiple frames, and video continuation. I made this Riftfall trailer with it, and honestly, the cinematic quality looks wild. FLUX 3 is available now through the API and open weights coming soon. Could this beat Seedance 2.5?
👍188❤181🔥169🎉161
Wan 3.0 is here! 🚀
In short: Alibaba has officially launched the public beta of Wan 3.0. The new model can generate videos with audio up to 30 seconds long in 480p, 720p, and 1080p.
New features also include automatic duration selection, video continuation, and improved consistency of characters, objects, and scenes across longer clips.
https://article.9466.com/news/aKweEQlm
API pricing:
480p — around $0.04 per second
720p — around $0.08 per second
1080p — around $0.17 per second
That means a maximum 30-second video costs approximately $1.25 at 480p, $2.50 at 720p, and around $5 at 1080p.
But there’s a catch: when using a video reference, Alibaba charges not only for the generated output but also for the duration of the uploaded reference video.
The formula is: input video + generated video.
For example, a 10-second reference combined with a 30-second 1080p output would cost approximately $6.70.
In short: Alibaba has officially launched the public beta of Wan 3.0. The new model can generate videos with audio up to 30 seconds long in 480p, 720p, and 1080p.
New features also include automatic duration selection, video continuation, and improved consistency of characters, objects, and scenes across longer clips.
https://article.9466.com/news/aKweEQlm
API pricing:
480p — around $0.04 per second
720p — around $0.08 per second
1080p — around $0.17 per second
That means a maximum 30-second video costs approximately $1.25 at 480p, $2.50 at 720p, and around $5 at 1080p.
But there’s a catch: when using a video reference, Alibaba charges not only for the generated output but also for the duration of the uploaded reference video.
The formula is: input video + generated video.
For example, a 10-second reference combined with a 30-second 1080p output would cost approximately $6.70.
👍665🎉612❤588🔥578