正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Cherry Blossom Mind + With Krea , I’m training visual models that echo and unfold like petals in dialogue. Each output feels like a fragment of a larger poetic system.

Karoline Georges

7,942 subscribers

26,266 次观看 • 1 年前 •via X (Twitter)

艺术科学技术

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Wow… the more I use #klingai 2.6, the more it blows my mind 🤯🎥 Every output feels like a visual level-up. It’s seriously next level.

Wow… the more I use #klingai 2.6, the more it blows my mind 🤯🎥 Every output feels like a visual level-up. It’s seriously next level.

Pierrick Chevallier | IA

18,979 次观看 • 8 个月前

For 3.11Labs, we created a visual identity that reflects the studio’s balance of structure and experimentation. The system supports their AI creations with a distinct visual style while revealing more about the technology behind each piece.

For 3.11Labs, we created a visual identity that reflects the studio’s balance of structure and experimentation. The system supports their AI creations with a distinct visual style while revealing more about the technology behind each piece.

Holographik

12,163 次观看 • 1 年前

I feel like each parent of Kaylee, Maddie, Xana, and Ethan Each should get 5 minutes in a locked room with that demon #BryanKohberger…

I feel like each parent of Kaylee, Maddie, Xana, and Ethan Each should get 5 minutes in a locked room with that demon #BryanKohberger…

chrissy paradis

92,440 次观看 • 1 年前

Kling 3.0 is now available in ComfyUI via Partner Nodes! You can now generate multiple shots in a single run with precise duration control for each. A new level of consistency, multilingual dialogue, and up to 15 seconds of cinematic output. Fully composable with ComfyUI’s ecosystem of nodes and extensions!

Kling 3.0 is now available in ComfyUI via Partner Nodes! You can now generate multiple shots in a single run with precise duration control for each. A new level of consistency, multilingual dialogue, and up to 15 seconds of cinematic output. Fully composable with ComfyUI’s ecosystem of nodes and extensions!

ComfyUI

14,650 次观看 • 6 个月前

"Do not miss the once in a lifetime chance to see the largest collection of 329 ‘Taihaku’ cherry blossom trees in the entire world all in their two week blossom window this spring at Alnwick Garden. Such a rare sight and no where else in the world has this beauty at such a scale but right here in Northumberland you can visit the cherry tree orchard and take it all in. The view is breathtaking and something you have to see with your own eyes. In 1926 Captain Collingwood Ingram was invited to Japan to give a lecture to the Sakurakai (cherry society). Whilst there he was shown a picture in an 18th Century book of a large, white cherry blossom that had become extinct. He recognised the same blossom on a cherry tree which had been imported to a Sussex garden in 1899 and was able to take cuttings from it and reintroduce the lost Taihaku to Japan in 1932.All of the Taihaku in cultivation today are off-spring of that cherry tree in Sussex!" -Emma.

"Do not miss the once in a lifetime chance to see the largest collection of 329 ‘Taihaku’ cherry blossom trees in the entire world all in their two week blossom window this spring at Alnwick Garden. Such a rare sight and no where else in the world has this beauty at such a scale but right here in Northumberland you can visit the cherry tree orchard and take it all in. The view is breathtaking and something you have to see with your own eyes. In 1926 Captain Collingwood Ingram was invited to Japan to give a lecture to the Sakurakai (cherry society). Whilst there he was shown a picture in an 18th Century book of a large, white cherry blossom that had become extinct. He recognised the same blossom on a cherry tree which had been imported to a Sussex garden in 1899 and was able to take cuttings from it and reintroduce the lost Taihaku to Japan in 1932.All of the Taihaku in cultivation today are off-spring of that cherry tree in Sussex!" -Emma.

North East Tweets

13,071 次观看 • 2 年前

im ctfu not a moa wiggling kai’s favorite snack in front of him like they’re calling a puppy😭 and the cherry on top is that he’s actually reacting like a puppy

im ctfu not a moa wiggling kai’s favorite snack in front of him like they’re calling a puppy😭 and the cherry on top is that he’s actually reacting like a puppy

mochikai☆⁺⸝⋆

56,176 次观看 • 4 个月前

it's crazy what a 1.5B model can do these days! "VibeThinker-1.5B is a 1.5-billion parameter dense language model. With a total training cost of only $7,800 USD, it achieves reasoning performance comparable to larger models like GPT OSS-20B Medium." runs perfectly on device!

it's crazy what a 1.5B model can do these days! "VibeThinker-1.5B is a 1.5-billion parameter dense language model. With a total training cost of only $7,800 USD, it achieves reasoning performance comparable to larger models like GPT OSS-20B Medium." runs perfectly on device!

Maziyar PANAHI

202,387 次观看 • 9 个月前

Most video tools can generate clips. Very few can maintain identity. That has been the real bottleneck in AI video creation. Kling O1 changes that. For the first time, creators can carry a character, style, and visual language across scenes without constant fixes. You can reference past clips, assets, or images and the output stays consistently on-model. No visual drift. No rework loops. No “this doesn’t look like the last shot” moments. It feels less like prompting a tool and more like working with a creative collaborator that remembers context. The impact is practical, not theoretical: → Faster production cycles → Lower iteration costs → Noticeably higher output quality This is what mature AI tooling looks like. Not louder features. Not bigger claims. Just reliability where it actually matters. Consistency is no longer the problem.

Most video tools can generate clips. Very few can maintain identity. That has been the real bottleneck in AI video creation. Kling O1 changes that. For the first time, creators can carry a character, style, and visual language across scenes without constant fixes. You can reference past clips, assets, or images and the output stays consistently on-model. No visual drift. No rework loops. No “this doesn’t look like the last shot” moments. It feels less like prompting a tool and more like working with a creative collaborator that remembers context. The impact is practical, not theoretical: → Faster production cycles → Lower iteration costs → Noticeably higher output quality This is what mature AI tooling looks like. Not louder features. Not bigger claims. Just reliability where it actually matters. Consistency is no longer the problem.

Darshal Jaitwar

141,038 次观看 • 7 个月前

Playing with scale in Simulon is so magical and a lot of fun. It feels like being a kid again!

Playing with scale in Simulon is so magical and a lot of fun. It feels like being a kid again!

Divesh Naidoo

166,321 次观看 • 2 年前

What once required hours of production for a UGC ad can now be done in just minutes. Pollo AI Seedance 2.0 4K created this entire commercial with: • Ultra-realistic skin detail • Smooth, natural facial expressions • Perfect product consistency throughout • Text that remains stable and fixed • High-end 4K visual output We’re watching a new phase of AI advertising unfold.

What once required hours of production for a UGC ad can now be done in just minutes. Pollo AI Seedance 2.0 4K created this entire commercial with: • Ultra-realistic skin detail • Smooth, natural facial expressions • Perfect product consistency throughout • Text that remains stable and fixed • High-end 4K visual output We’re watching a new phase of AI advertising unfold.

Soulful Ai

35,992 次观看 • 2 个月前

I'm in that phase where every night feels like a battle with my own thoughts

I'm in that phase where every night feels like a battle with my own thoughts

Real Post Folder

21,384 次观看 • 11 个月前

Both the face and the mind have become like that of a buffalo. #Trump

Both the face and the mind have become like that of a buffalo. #Trump

رضوانہ رضا

14,438 次观看 • 2 个月前

Ah yes, I sure love the one that horribly mischaracterizes Shadow, and then all I’m left with is unfunny face humor (not meme face before you act like a dumbass). I don’t like Tyson Hesse’s visual humor style in these shorts at all…

Ah yes, I sure love the one that horribly mischaracterizes Shadow, and then all I’m left with is unfunny face humor (not meme face before you act like a dumbass). I don’t like Tyson Hesse’s visual humor style in these shorts at all…

Shamaboy

33,404 次观看 • 1 年前

We reviewed STARBITES! It's a sexy, fun, tactical JRPG bursting with action~ There's something truly special about STARBITES. It feels like a charming and fun long-lost PS2 classic with standout characters and a combat system that will keep you hooked~

We reviewed STARBITES! It's a sexy, fun, tactical JRPG bursting with action~ There's something truly special about STARBITES. It feels like a charming and fun long-lost PS2 classic with standout characters and a combat system that will keep you hooked~

Niche Gamer

25,263 次观看 • 3 个月前

Most recent diffusion language model research (that I’ve seen) seems to be using masking as the noising process. It looks like, however, most closed-source models (Google Gemini Diffusion and possibly Inception Labs’ Mercury) use a different noising process, where instead of masking tokens, they replace them with different tokens (either with a random token or a semantically similar token). I wondered how they were getting such high throughput with the latter noising process, since I believed that optimizing inference with KVCache approximation would be more difficult (for various reasons). I visualized this noising process with tiny-diffusion and compared it to normal unmasking, and was very surprised to see how fast the generation “settles” into a reasonable output, and then only slightly refines afterwards, requiring much fewer steps in total. Unmasking (where tokens are never remasked, the typical implementation) is inherently limited in generation speed by the fact that an increase in tokens decoded per step leads to more errors due to the mismatch between individual and marginal token probability distributions we sample from. The token replacement noising process seems to have a much different set of characteristics. Because we sample each token per step, every token makes “progress” towards the final output each iteration (in addition to *potentially* giving other tokens more information in future steps). Generally, masking has outperformed other noising processes, which is probably why most research focused on it (using smaller models). But the paper referred to in the retweet shows that random replacement as a noising process may scale better as model size increases. Big labs might have noticed these results much earlier (due to having drastically more training resources and being able to test larger models), which may explain the discrepancy in the choice of noising process. I’m gonna test this with larger models, since tiny-diffusion only has 10M parameters.

Most recent diffusion language model research (that I’ve seen) seems to be using masking as the noising process. It looks like, however, most closed-source models (Google Gemini Diffusion and possibly Inception Labs’ Mercury) use a different noising process, where instead of masking tokens, they replace them with different tokens (either with a random token or a semantically similar token). I wondered how they were getting such high throughput with the latter noising process, since I believed that optimizing inference with KVCache approximation would be more difficult (for various reasons). I visualized this noising process with tiny-diffusion and compared it to normal unmasking, and was very surprised to see how fast the generation “settles” into a reasonable output, and then only slightly refines afterwards, requiring much fewer steps in total. Unmasking (where tokens are never remasked, the typical implementation) is inherently limited in generation speed by the fact that an increase in tokens decoded per step leads to more errors due to the mismatch between individual and marginal token probability distributions we sample from. The token replacement noising process seems to have a much different set of characteristics. Because we sample each token per step, every token makes “progress” towards the final output each iteration (in addition to potentially giving other tokens more information in future steps). Generally, masking has outperformed other noising processes, which is probably why most research focused on it (using smaller models). But the paper referred to in the retweet shows that random replacement as a noising process may scale better as model size increases. Big labs might have noticed these results much earlier (due to having drastically more training resources and being able to test larger models), which may explain the discrepancy in the choice of noising process. I’m gonna test this with larger models, since tiny-diffusion only has 10M parameters.

nathan (in sf)

40,440 次观看 • 7 个月前

Introducing Claude Sonnet 5, our most agentic Sonnet yet. It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models.

Introducing Claude Sonnet 5, our most agentic Sonnet yet. It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models.

Claude

11,852,582 次观看 • 2 个月前

I’ve been obsessively training a Wan 2.2 LoRA for morphing transitions for a couple of months and it feels like I’m getting closer now It’s been deceptively hard training something so conceptual, since you don’t want the model to learn any specific movements or design elements

I’ve been obsessively training a Wan 2.2 LoRA for morphing transitions for a couple of months and it feels like I’m getting closer now It’s been deceptively hard training something so conceptual, since you don’t want the model to learn any specific movements or design elements

Ingi Erlingsson 🪄

31,158 次观看 • 9 个月前