There's how it works:
1. You upload the dataset with preconfigured format into HuggingFa\u0441e [1].
2. Choose your LLM (e.g. LLaMa 70B, Mistral 7B)
3. Place your submission into the queue
4. Wait for it to get trained.
5. Then you get your trained model there on HuggingFace.
Essentially, why would we want to do it?
1. We already have an experience with training big LLMs.
2. We could achieve near-perfect infrastructure performance for training.
3. Sometimes GPUs have just nothing to train.
Thus we thought it would be cool if we could utilize our GPU cluster 100%. And give back to Open Source community (already built an e2e distributed training framework [2]).
This is in an early stage, so you can expect some bugs.
Any thoughts, opinions, or ideas are quite welcome!
[1]: https://github.com/higgsfield-ai/higgsfield/blob/main/tutori...
[2]: https://github.com/higgsfield-ai/higgsfield"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Show HN: Higgsfield \u2013 Finetune LLMs for Free"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield.xyz"}},"_tags":["story","author_higgsfield","story_38246379","show_hn"],"author":"higgsfield","children":[38246462],"created_at":"2023-11-13T02:58:07Z","created_at_i":1699844287,"num_comments":1,"objectID":"38246379","points":5,"story_id":38246379,"story_text":"We have a massive GPU cluster and developed our own infrastructure to manage the cluster and train massive models.
There's how it works:
1. You upload the dataset with preconfigured format into HuggingFa\u0441e [1].
2. Choose your LLM (e.g. LLaMa 70B, Mistral 7B)
3. Place your submission into the queue
4. Wait for it to get trained.
5. Then you get your trained model there on HuggingFace.
Essentially, why would we want to do it?
1. We already have an experience with training big LLMs.
2. We could achieve near-perfect infrastructure performance for training.
3. Sometimes GPUs have just nothing to train.
Thus we thought it would be cool if we could utilize our GPU cluster 100%. And give back to Open Source community (already built an e2e distributed training framework [2]).
This is in an early stage, so you can expect some bugs.
Any thoughts, opinions, or ideas are quite welcome!
[1]: https://github.com/higgsfield-ai/higgsfield/blob/main/tutori...
[2]: https://github.com/higgsfield-ai/higgsfield","title":"Show HN: Higgsfield \u2013 Finetune LLMs for Free","updated_at":"2024-09-20T15:37:56Z","url":"https://higgsfield.xyz"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"azeemba"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Primordial black holes and the Higgs field in the early universe"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://theconversation.com/the-higgs-particle-could-have-ended-the-universe-by-now-heres-why-were-still-here-235694"}},"_tags":["story","author_azeemba","story_41146241"],"author":"azeemba","created_at":"2024-08-03T12:26:14Z","created_at_i":1722687974,"num_comments":0,"objectID":"41146241","points":5,"story_id":41146241,"title":"Primordial black holes and the Higgs field in the early universe","updated_at":"2024-09-20T17:34:02Z","url":"https://theconversation.com/the-higgs-particle-could-have-ended-the-universe-by-now-heres-why-were-still-here-235694"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jenthoven"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Stripe, Google, Canva, Cloudflare and Higgsfield Are Selling"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://www.saastr.com/how-stripe-google-canva-cloudflare-and-higgsfield-are-actually-selling-in-2026/"}},"_tags":["story","author_jenthoven","story_48865097"],"author":"jenthoven","created_at":"2026-07-10T20:52:44Z","created_at_i":1783716764,"num_comments":0,"objectID":"48865097","points":4,"story_id":48865097,"title":"Stripe, Google, Canva, Cloudflare and Higgsfield Are Selling","updated_at":"2026-07-11T00:52:50Z","url":"https://www.saastr.com/how-stripe-google-canva-cloudflare-and-higgsfield-are-actually-selling-in-2026/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"adithyaharish"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Looking for open source tools to build animated, interactive websites, things like landing pages, storytelling sites, or product pages with smooth transitions, motion, or scroll effects."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Ask HN: Open-source website builders for animated sites like Higgsfield?"}},"_tags":["story","author_adithyaharish","story_48802983","ask_hn"],"author":"adithyaharish","children":[48867121],"created_at":"2026-07-06T10:50:53Z","created_at_i":1783335053,"num_comments":0,"objectID":"48802983","points":4,"story_id":48802983,"story_text":"Looking for open source tools to build animated, interactive websites, things like landing pages, storytelling sites, or product pages with smooth transitions, motion, or scroll effects.","title":"Ask HN: Open-source website builders for animated sites like Higgsfield?","updated_at":"2026-07-11T00:26:51Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"arpanetus"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Higgsfield: Distributed LLM training and cluster management framework"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield-ai/higgsfield"}},"_tags":["story","author_arpanetus","story_37973626"],"author":"arpanetus","created_at":"2023-10-22T07:54:47Z","created_at_i":1697961287,"num_comments":0,"objectID":"37973626","points":4,"story_id":37973626,"title":"Higgsfield: Distributed LLM training and cluster management framework","updated_at":"2024-09-20T15:28:23Z","url":"https://github.com/higgsfield-ai/higgsfield"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"akhairov"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Higgsfield AI: Anyone Can Train Llama 70B or Mistral for Free"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield.ai"}},"_tags":["story","author_akhairov","story_38284761"],"author":"akhairov","children":[38284915,38284963],"created_at":"2023-11-16T01:20:13Z","created_at_i":1700097613,"num_comments":3,"objectID":"38284761","points":2,"story_id":38284761,"title":"Higgsfield AI: Anyone Can Train Llama 70B or Mistral for Free","updated_at":"2024-09-20T15:42:24Z","url":"https://higgsfield.ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sayonidroy"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"If you use AI to create images and videos, you most likely use Higgsfield or similar subscription tools.
I did too. But I had 3 main issues with it:
The markup is super high. So if the model provider charges x, higgsfield charges a markup of + 50% on top.
Credits I did not use would expire every month....EVEN THOUGH I PAID FOR IT
And I had to switch between Claude and Higgsfield constantly...copy pasting prompts and all
So I built a free open source repo of creative production skills for AI agents.
So basically, you just install this skill package in Claude (or any agent you use) and say "Create a summer sale UGC video for my product". And it handles the rest. You obviously stay in control all along.
And:
You can use your own API keys...which adds ZERO markup on the model provider costs
There is no subscription. You only pay for what you use and nothing EXPIRES
And no more switching between tools and no more copy pasting
We are releasing the skills in batches. The first batch that can research the market, create ugc videos, product videos, product photoshoots, etc. is already live here: https://github.com/SupercmoHQ/superCMO-skills"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Show HN: Open-source Claude skills to replace Higgsfield"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/SupercmoHQ/superCMO-skills"}},"_tags":["story","author_sayonidroy","story_50191723","show_hn"],"author":"sayonidroy","created_at":"2026-10-22T06:03:55Z","created_at_i":1792649035,"num_comments":0,"objectID":"50191723","points":2,"story_id":50191723,"story_text":"If you use AI to create images and videos, you most likely use Higgsfield or similar subscription tools.
I did too. But I had 3 main issues with it:
The markup is super high. So if the model provider charges x, higgsfield charges a markup of + 50% on top.
Credits I did not use would expire every month....EVEN THOUGH I PAID FOR IT
And I had to switch between Claude and Higgsfield constantly...copy pasting prompts and all
So I built a free open source repo of creative production skills for AI agents.
So basically, you just install this skill package in Claude (or any agent you use) and say "Create a summer sale UGC video for my product". And it handles the rest. You obviously stay in control all along.
And:
You can use your own API keys...which adds ZERO markup on the model provider costs
There is no subscription. You only pay for what you use and nothing EXPIRES
And no more switching between tools and no more copy pasting
We are releasing the skills in batches. The first batch that can research the market, create ugc videos, product videos, product photoshoots, etc. is already live here: https://github.com/SupercmoHQ/superCMO-skills","title":"Show HN: Open-source Claude skills to replace Higgsfield","updated_at":"2026-10-22T06:30:51Z","url":"https://github.com/SupercmoHQ/superCMO-skills"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"lliu56p"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Minicut AI: Higgsfield but for AI Drama and Film"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.minicut.ai"}},"_tags":["story","author_lliu56p","story_49862980"],"author":"lliu56p","children":[49862981,49863166],"created_at":"2026-09-27T03:05:08Z","created_at_i":1790478308,"num_comments":2,"objectID":"49862980","points":1,"story_id":49862980,"title":"Minicut AI: Higgsfield but for AI Drama and Film","updated_at":"2026-09-27T03:59:15Z","url":"https://www.minicut.ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"daly"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"I got into a discussion of UFOs. The question was how would they\nbe able to change directions and accelerate instantly. He was trying\nto guess what kind of propulstion they used. This is the reply I sent.
The more I think about it it becomes clear that physics is wrongly constructed.
The ability to manipulate space-time directly solves a lot of problems.\nTake, for example, the UFO problem I mentioned. Physics has the concept of\npropulsion, equal and opposite "forces". If instead you could directly manipulate\n"mass" as a variable then things change. The photon is massless which makes the\nuniverse "flat" and leads to travel at the speed of light. If you could shed mass\nat will you could easily approach the speed of light without "propulsion". Thus a\ncivilization that figures out how to manipulate "mass" would be able to travel\nanywhere at the speed of light. You don't accelerate mass, which is dumb,\nyou "shed mass" and thus move without propulsion. Indeed, if you could\nachieve "negative mass" (changing a gravity well into a gravity hill) you could\neasitly exceed the speed of light.
Einstein felt that things got more massive as they approach the speed of light\nand, in the limit, would have infinite mass. That's true under propulsion. It is not\ntrue if you can "shed mass". Thus Einstein missed the very thing that mattered.
Shedding mass likely involves directly manipulating the Higgs boson.
Dark energy and dark matter might be "negative masses".
I don't know any physics professors but I'm pretty sure I'm talking nonsense :-)"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"UFOs and the Higgs Field"}},"_tags":["story","author_daly","story_47031841","ask_hn"],"author":"daly","children":[47032071],"created_at":"2026-02-16T07:09:53Z","created_at_i":1771225793,"num_comments":1,"objectID":"47031841","points":1,"story_id":47031841,"story_text":"I got into a discussion of UFOs. The question was how would they\nbe able to change directions and accelerate instantly. He was trying\nto guess what kind of propulstion they used. This is the reply I sent.
The more I think about it it becomes clear that physics is wrongly constructed.
The ability to manipulate space-time directly solves a lot of problems.\nTake, for example, the UFO problem I mentioned. Physics has the concept of\npropulsion, equal and opposite "forces". If instead you could directly manipulate\n"mass" as a variable then things change. The photon is massless which makes the\nuniverse "flat" and leads to travel at the speed of light. If you could shed mass\nat will you could easily approach the speed of light without "propulsion". Thus a\ncivilization that figures out how to manipulate "mass" would be able to travel\nanywhere at the speed of light. You don't accelerate mass, which is dumb,\nyou "shed mass" and thus move without propulsion. Indeed, if you could\nachieve "negative mass" (changing a gravity well into a gravity hill) you could\neasitly exceed the speed of light.
Einstein felt that things got more massive as they approach the speed of light\nand, in the limit, would have infinite mass. That's true under propulsion. It is not\ntrue if you can "shed mass". Thus Einstein missed the very thing that mattered.
Shedding mass likely involves directly manipulating the Higgs boson.
Dark energy and dark matter might be "negative masses".
I don't know any physics professors but I'm pretty sure I'm talking nonsense :-)","title":"UFOs and the Higgs Field","updated_at":"2026-03-05T23:33:31Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"varshith17"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"comment down your answer"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Why Suddenly higgsfield made its core repo Opensource?"}},"_tags":["story","author_varshith17","story_49744855","ask_hn"],"author":"varshith17","children":[49744898],"created_at":"2026-09-17T18:44:57Z","created_at_i":1789670697,"num_comments":0,"objectID":"49744855","points":1,"story_id":49744855,"story_text":"comment down your answer","title":"Why Suddenly higgsfield made its core repo Opensource?","updated_at":"2026-09-17T18:49:47Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"memalign"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Higgsfield Visual Effects"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield.ai/collection/effects"}},"_tags":["story","author_memalign","story_43976306"],"author":"memalign","created_at":"2025-05-13T18:50:57Z","created_at_i":1747162257,"num_comments":0,"objectID":"43976306","points":1,"story_id":43976306,"title":"Higgsfield Visual Effects","updated_at":"2025-10-04T01:52:07Z","url":"https://higgsfield.ai/collection/effects"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"handfuloflight"},"title":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Higgsfield"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield.ai/"}},"_tags":["story","author_handfuloflight","story_43543627"],"author":"handfuloflight","created_at":"2025-04-01T06:53:36Z","created_at_i":1743490416,"num_comments":0,"objectID":"43543627","points":1,"story_id":43543627,"title":"Higgsfield","updated_at":"2025-06-30T10:35:58Z","url":"https://higgsfield.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"evo_9"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Researchers mimic relativity and the Higgs field in graphene-like material"},"url":{"matchLevel":"none","matchedWords":[],"value":"http://arstechnica.com/science/news/2012/03/tuning-molecular-graphene-to-make-exotic-quasiparticles.ars"}},"_tags":["story","author_evo_9","story_3709329"],"author":"evo_9","created_at":"2012-03-15T17:03:45Z","created_at_i":1331831025,"num_comments":0,"objectID":"3709329","points":1,"story_id":3709329,"title":"Researchers mimic relativity and the Higgs field in graphene-like material","updated_at":"2024-09-19T18:21:24Z","url":"http://arstechnica.com/science/news/2012/03/tuning-molecular-graphene-to-make-exotic-quasiparticles.ars"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"axiomdata316"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Higgsfield/RL-Adventure-2: PyTorch4 tutorial of: actor critic etc."},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/RL-Adventure-2"}},"_tags":["story","author_axiomdata316","story_17177010"],"author":"axiomdata316","created_at":"2018-05-29T06:56:05Z","created_at_i":1527576965,"num_comments":0,"objectID":"17177010","points":1,"story_id":17177010,"title":"Higgsfield/RL-Adventure-2: PyTorch4 tutorial of: actor critic etc.","updated_at":"2024-09-20T02:34:33Z","url":"https://github.com/higgsfield/RL-Adventure-2"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"platevoltage"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Just to preface, this isn't the typical "Am I being replaced by vibe coding" post.
I just had a talk with one of my first ongoing clients, who sort of helped me kickstart my freelancing "career". He is one of those people who has a broad understanding of the tech out there, and what its capabilities are, but is not a programmer. He wanted to show me, and get my opinion on a restaurant management app that he's been using Lovable to develop. Admittedly, a lot of it is very impressive, and he has obviously become very good at using these tools. Part of that scares me, but I've largely accepted the reality that my skillset has to expand beyond being a React code monkey (wish is not what I am, but you get the point). I've always been a curious generalist, so picking up new stuff comes easily, and I'll manage.
Anyways, here's the part that feels really gross. He's working on a feature that does some automated marketing using Higgsfield, so automated slop posting. Also a feature that consumes reviews from vendors, and he intends to use agents to post reviews as well.
I'm 43, I managed to convince my parents to get AOL when I was 12. I made a Geocities site at 14. I've been obsessed with computers and the internet for my entire life. I'm also an artist who had a somewhat successful run as a touring musician, much to the detriment of my career growth (not much money in Punk).
I feel the hobbies that I've been nurturing for my entire life are being destroyed by people who have finally found an entry point that requires the smallest amount of effort. We are literally watching the dead internet happen in real time, social media turning into a AI slop gallery, AI music topping the charts, and I just had a one-on-one conversation with someone that just doesn't care, because they get to feel creative for possibly the first time ever.
How are you all dealing with this?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: I just talked to an AI-obsessed client, and I need a shower afterwards"}},"_tags":["story","author_platevoltage","story_49826029","ask_hn"],"author":"platevoltage","children":[49826104,49826345,49826586,49827123,49827430,49827618,49828265,49828397,49829428,49830222,49830260,49831601,49832296,49832929,49833187,49833692,49833724,49834071,49835479,49835679,49839341,49840334,49841025,49843960,49844439,49845304,49845687,49846939,49918125],"created_at":"2026-09-24T03:48:41Z","created_at_i":1790221721,"num_comments":51,"objectID":"49826029","points":30,"story_id":49826029,"story_text":"Just to preface, this isn't the typical "Am I being replaced by vibe coding" post.
I just had a talk with one of my first ongoing clients, who sort of helped me kickstart my freelancing "career". He is one of those people who has a broad understanding of the tech out there, and what its capabilities are, but is not a programmer. He wanted to show me, and get my opinion on a restaurant management app that he's been using Lovable to develop. Admittedly, a lot of it is very impressive, and he has obviously become very good at using these tools. Part of that scares me, but I've largely accepted the reality that my skillset has to expand beyond being a React code monkey (wish is not what I am, but you get the point). I've always been a curious generalist, so picking up new stuff comes easily, and I'll manage.
Anyways, here's the part that feels really gross. He's working on a feature that does some automated marketing using Higgsfield, so automated slop posting. Also a feature that consumes reviews from vendors, and he intends to use agents to post reviews as well.
I'm 43, I managed to convince my parents to get AOL when I was 12. I made a Geocities site at 14. I've been obsessed with computers and the internet for my entire life. I'm also an artist who had a somewhat successful run as a touring musician, much to the detriment of my career growth (not much money in Punk).
I feel the hobbies that I've been nurturing for my entire life are being destroyed by people who have finally found an entry point that requires the smallest amount of effort. We are literally watching the dead internet happen in real time, social media turning into a AI slop gallery, AI music topping the charts, and I just had a one-on-one conversation with someone that just doesn't care, because they get to feel creative for possibly the first time ever.
How are you all dealing with this?","title":"Ask HN: I just talked to an AI-obsessed client, and I need a shower afterwards","updated_at":"2026-10-12T17:26:37Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Alisaqqt"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"ByteDance quietly shipped Seedance 2.0. The interesting part isn't the usual text-to-video upgrade \u2014 it's the reference/conditioning system.
What's different from the typical T2V model:
Accepts 4 input modalities simultaneously: text, images (up to 9), video clips (up to 3, \u226415s total), and audio (up to 3, \u226415s total). Mixed input cap is 12 files.\nReference-driven generation: you can use an image to lock composition/character appearance, a video clip to specify camera movement and motion dynamics, and an audio track to drive rhythm and tempo. Outputs include generated SFX/BGM.\nThe key claim is "audio-driven video" rather than "video with audio attached" \u2014 meaning motion is actually synced to the audio input's beat structure, not just overlaid.\nSupports video continuation/extension with shot-to-shot coherence, and editing operations (character swap, segment insertion/removal) on existing clips.\nOutput: 4\u201315s, selectable. Comes with built-in sound.\nWhy this matters technically:
Most current video models treat audio as a post-processing step. Seedance 2.0 appears to condition the diffusion process on audio features directly, which would explain the beat-sync behavior. The multi-reference @ tagging system (@image1 for composition, @video1 for motion, @audio1 for rhythm) suggests a mixture-of-conditions architecture rather than simple concatenation.
Haven't seen an official announcement yet. Docs are up on Dreamina (ByteDance's creative platform). Curious if anyone has more details on the architecture.
If you want to test them after launch, here are a few good platforms depending on your use case:\n- For developers (API): https://www.atlascloud.ai/\n- For creators: Higgsfield, ImagenArt
More info of the Seedance 2.0: https://www.reddit.com/r/SoraAi/comments/1qxdv5u/seedance_20_teaser_better_than_sora_2_true/\nSubreddit of Seedance 2.0 for discussion: https://www.reddit.com/r/Seedance_AI"},"title":{"matchLevel":"none","matchedWords":[],"value":"Seedance 2.0 preview: The best video model of 2026, outperforming Sora 2"}},"_tags":["story","author_Alisaqqt","story_46940720","ask_hn"],"author":"Alisaqqt","children":[46942097,46955967,46955970,46958022,46960318,47021461],"created_at":"2026-02-09T01:58:10Z","created_at_i":1770602290,"num_comments":7,"objectID":"46940720","points":7,"story_id":46940720,"story_text":"ByteDance quietly shipped Seedance 2.0. The interesting part isn't the usual text-to-video upgrade \u2014 it's the reference/conditioning system.
What's different from the typical T2V model:
Accepts 4 input modalities simultaneously: text, images (up to 9), video clips (up to 3, \u226415s total), and audio (up to 3, \u226415s total). Mixed input cap is 12 files.\nReference-driven generation: you can use an image to lock composition/character appearance, a video clip to specify camera movement and motion dynamics, and an audio track to drive rhythm and tempo. Outputs include generated SFX/BGM.\nThe key claim is "audio-driven video" rather than "video with audio attached" \u2014 meaning motion is actually synced to the audio input's beat structure, not just overlaid.\nSupports video continuation/extension with shot-to-shot coherence, and editing operations (character swap, segment insertion/removal) on existing clips.\nOutput: 4\u201315s, selectable. Comes with built-in sound.\nWhy this matters technically:
Most current video models treat audio as a post-processing step. Seedance 2.0 appears to condition the diffusion process on audio features directly, which would explain the beat-sync behavior. The multi-reference @ tagging system (@image1 for composition, @video1 for motion, @audio1 for rhythm) suggests a mixture-of-conditions architecture rather than simple concatenation.
Haven't seen an official announcement yet. Docs are up on Dreamina (ByteDance's creative platform). Curious if anyone has more details on the architecture.
If you want to test them after launch, here are a few good platforms depending on your use case:\n- For developers (API): https://www.atlascloud.ai/\n- For creators: Higgsfield, ImagenArt
More info of the Seedance 2.0: https://www.reddit.com/r/SoraAi/comments/1qxdv5u/seedance_20_teaser_better_than_sora_2_true/\nSubreddit of Seedance 2.0 for discussion: https://www.reddit.com/r/Seedance_AI","title":"Seedance 2.0 preview: The best video model of 2026, outperforming Sora 2","updated_at":"2026-03-05T23:32:51Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"JimsonYang"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"So I was prototyping with claude code and Godot engine, and I thought it was weird that this specific category of AI isn't as big as video gen creative suites. Like theres big names like runway and higgsfield but when it comes to video games there's no major company trying to make a one stop shop using AI to make video games. I'm sure it isn't because of a lack of trying, I'm sure developers thought about it as one of the first things to automate.
If I had to guess there's a couple reasons:
1)Quality-making a good game requires extreme attention to detail all of which is non coding related. The animations on idle characters, how the background characters look, the sound design and clever story. 90% of gaming is art
2) backlash from players. I think this is an important factor but not the dominant factor. For reasons like games like Slotbound have had amazing positive responses despite being ai generated as long as they are high quality. On the flip side reputation does matter for major studios so adoption can be slow.
3)No money in the industry. Most game studios aren't raining in money, therefore making video games would be an expensive and time consuming hobby that got slightly easier, but is still difficult. A normal prosumer simply doesn't have the time nor dedication to make an project that is going to take months of their time
These are my guesses, but I would love to hear responses from people in the industry or people who have attempted this idea or VCs who have their own thesis on the market.
p.s. this isn't a lazy attempt startup research, I'm seriously just curious"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Why isn't there a cursor for video games development"}},"_tags":["story","author_JimsonYang","story_49571091","ask_hn"],"author":"JimsonYang","children":[49571103,49572647,49573258,49579521,49583326,49609392],"created_at":"2026-09-04T22:45:49Z","created_at_i":1788561949,"num_comments":6,"objectID":"49571091","points":2,"story_id":49571091,"story_text":"So I was prototyping with claude code and Godot engine, and I thought it was weird that this specific category of AI isn't as big as video gen creative suites. Like theres big names like runway and higgsfield but when it comes to video games there's no major company trying to make a one stop shop using AI to make video games. I'm sure it isn't because of a lack of trying, I'm sure developers thought about it as one of the first things to automate.
If I had to guess there's a couple reasons:
1)Quality-making a good game requires extreme attention to detail all of which is non coding related. The animations on idle characters, how the background characters look, the sound design and clever story. 90% of gaming is art
2) backlash from players. I think this is an important factor but not the dominant factor. For reasons like games like Slotbound have had amazing positive responses despite being ai generated as long as they are high quality. On the flip side reputation does matter for major studios so adoption can be slow.
3)No money in the industry. Most game studios aren't raining in money, therefore making video games would be an expensive and time consuming hobby that got slightly easier, but is still difficult. A normal prosumer simply doesn't have the time nor dedication to make an project that is going to take months of their time
These are my guesses, but I would love to hear responses from people in the industry or people who have attempted this idea or VCs who have their own thesis on the market.
p.s. this isn't a lazy attempt startup research, I'm seriously just curious","title":"Ask HN: Why isn't there a cursor for video games development","updated_at":"2026-09-08T13:07:01Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Kumar963"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Open Source alternative to Higgsfield AI- No subscription. Free AI image generation and cinema studio with 20+ models. Self-hosted and customizable. BYOK (Bring Your Own key) support"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Open-Source Higgsfiled AI Alternative"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/Anil-matcha/Open-Higgsfield-AI"}},"_tags":["story","author_Kumar963","story_47673602","show_hn"],"author":"Kumar963","children":[47673658],"created_at":"2026-04-07T11:36:50Z","created_at_i":1775561810,"num_comments":2,"objectID":"47673602","points":2,"story_id":47673602,"story_text":"Open Source alternative to Higgsfield AI- No subscription. Free AI image generation and cinema studio with 20+ models. Self-hosted and customizable. BYOK (Bring Your Own key) support","title":"Show HN: Open-Source Higgsfiled AI Alternative","updated_at":"2026-04-07T12:12:10Z","url":"https://github.com/Anil-matcha/Open-Higgsfield-AI"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"joonjung"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"I think Suno opened up music. Kling and Runway opened up video - but there's one creative domain nobody has touched yet: dance.
So my team built an web app \u2014 give it a YouTube link to your music and it generates a 3D dance animation in under 2 minutes. The core is a diffusion-based music-to-motion model (mvnt-m4) trained on proprietary mocap/label data from professional choreographers.
I think dance is the missing piece of AI-generated content \u2014 just like how performance made K-pop a global phenomenon, and we believe AI dance will play the same role in the AI UGC era.
This is v0.1 \u2014 a fast, experimental playground. Dance quality is still improving (m4.1 in progress), and we're working on faster inference, and finger/facial motion generation. We're also preparing API integrations with platforms like Higgsfield.
I think our tech is quite validated through Epic MegaGrant but still very early in finding user validation. Would love honest feedback on the output quality and what you'd want to see next.
Also in product hunt: https://www.producthunt.com/products/mvntstudio
Demo vid:\nhttps://youtu.be/mjq2iAr96iM"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: mvntSTUDIO \u2013 AI model that generates dance choreography from music"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.mvnt.studio/"}},"_tags":["story","author_joonjung","story_47180954","show_hn"],"author":"joonjung","created_at":"2026-02-27T14:36:42Z","created_at_i":1772203002,"num_comments":0,"objectID":"47180954","points":2,"story_id":47180954,"story_text":"I think Suno opened up music. Kling and Runway opened up video - but there's one creative domain nobody has touched yet: dance.
So my team built an web app \u2014 give it a YouTube link to your music and it generates a 3D dance animation in under 2 minutes. The core is a diffusion-based music-to-motion model (mvnt-m4) trained on proprietary mocap/label data from professional choreographers.
I think dance is the missing piece of AI-generated content \u2014 just like how performance made K-pop a global phenomenon, and we believe AI dance will play the same role in the AI UGC era.
This is v0.1 \u2014 a fast, experimental playground. Dance quality is still improving (m4.1 in progress), and we're working on faster inference, and finger/facial motion generation. We're also preparing API integrations with platforms like Higgsfield.
I think our tech is quite validated through Epic MegaGrant but still very early in finding user validation. Would love honest feedback on the output quality and what you'd want to see next.
Also in product hunt: https://www.producthunt.com/products/mvntstudio
Demo vid:\nhttps://youtu.be/mjq2iAr96iM","title":"Show HN: mvntSTUDIO \u2013 AI model that generates dance choreography from music","updated_at":"2026-03-05T23:38:29Z","url":"https://www.mvnt.studio/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"phanahm"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"I got tired of all the expensive and complicated ways of making tiktok videos to promote my app (like using higgsfield or paying strangers $20 to act surprised on video), so I built somewhere you can paste any viral tiktok or Instagram reels URL to get your own version of the video + unlimited ai generations
If anyone wants to poke around and tell me what you think, you can find it at --> cloney.app. It's intuitive and there's no complicated setup. Just paste a tiktok URL.
PS try not to use it for porn
AI inference backend built on modal and hosted on vercel"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: I made an unlimited TikTok video generator to get views on the internet"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://cloney.app"}},"_tags":["story","author_phanahm","story_49692990","show_hn"],"author":"phanahm","children":[49693644],"created_at":"2026-09-14T07:06:21Z","created_at_i":1789369581,"num_comments":1,"objectID":"49692990","points":1,"story_id":49692990,"story_text":"I got tired of all the expensive and complicated ways of making tiktok videos to promote my app (like using higgsfield or paying strangers $20 to act surprised on video), so I built somewhere you can paste any viral tiktok or Instagram reels URL to get your own version of the video + unlimited ai generations
If anyone wants to poke around and tell me what you think, you can find it at --> cloney.app. It's intuitive and there's no complicated setup. Just paste a tiktok URL.
PS try not to use it for porn
AI inference backend built on modal and hosted on vercel","title":"Show HN: I made an unlimited TikTok video generator to get views on the internet","updated_at":"2026-09-14T08:31:08Z","url":"https://cloney.app"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zaaaaooo"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"If you've been using Nano Banana Pro for image generation, you already know it's been the strongest model in its class since launch. Now Nano Banana 2 is coming, and early preview samples are already circulating. Here's what we know so far and how to get access to it
What\u2019s New & Upgraded\n1. Pro Intelligence at Flash Speed and price: NB2 delivers quality close to Pro while keeping generation times in line with the Flash tier, standard generations clock in at under two seconds, with pricing significantly lower than Nano Banana Pro.\n2. Precision Text Rendering: Inheriting major upgrades from the Pro version, NB2 renders crisp, accurate text for marketing assets, UI/UX wireframes, and product labels. It also supports seamless translation and localization of text already existing within an image.\n3. Large-Scale Subject Consistency: Maintain identity across up to 5 characters and 14 distinct objects within a single workflow. This is a massive leap compared to the original Nano Banana.\n4. Real-World Grounding via Search The model leverages Google Search live to accurately render specific logos, landmarks, and real-world products. It can even transform handwritten notes or rough sketches into data-driven charts and infographics.
Use Cases\n1. Film & Creative Visuals: Fantasy concept design for pre-production, emphasizing the photorealistic grounding of surreal elements and logical physics.\n2. Marketing & E-Commerce Advertising: Rapid generation of product posters featuring precise typography and realistic materials.\n3. Documentary Photography & High-Fidelity Character Creation: Cinematic character settings or humanistic documentary projects. Focuses on skin texture, micro-details (e.g., hair, iris reflections), and authentic emotional delivery, breaking away from the "plastic" look of traditional AI generation.\n4. Illustration Design & Native Text Rendering: Educational materials, vintage posters, or graphic design assets. The core pain point is accurate spelling and seamless integration of typography into specific artistic styles.
Cost Check\nNano Banana 2\n- Official: $0.080\n- AtlasCloud.ai: $0.040\n- Fal AI: $0.080\n- Wavespeed: $0.080\n- Higgsfield: $0.057\nNano Banana Pro\n- Official: $0.139 - $0.240\n- AtlasCloud.ai: $0.063\n- Fal AI: $0.150\n- Wavespeed: $0.140\n- Higgsfield: $0.076\nSeedream 5.0 Lite\n- Official: $0.035\n- AtlasCloud.ai: $0.026\n- Fal AI: $0.035\n- Wavespeed: $0.035\n- Higgsfield: $0.038\nSeedream 4.5\n- Official: $0.040\n- AtlasCloud.ai: $0.038\n- Fal AI: $0.040\n- Wavespeed: $0.040\n- Higgsfield: $0.038
How to access Nano Banana 2\n- For creator: Gemini, Lovart.ai\uff0cHiggsfield AI, arena ai\uff0cVertex AI \n- For developer\uff1aAtlasCloud.ai, google ai studio
TL;DR\nNano Banana 2 is undeniably sending shockwaves through the AI image gen scene and it\u2019s poised to become one of the most popular tools. For everyday users and creators like us, this is truly the golden age:D"},"title":{"matchLevel":"none","matchedWords":[],"value":"Nano Banana 2 Is Really Coming! Here's How to Access It Early"}},"_tags":["story","author_zaaaaooo","story_47178063","ask_hn"],"author":"zaaaaooo","children":[47182510],"created_at":"2026-02-27T08:29:40Z","created_at_i":1772180980,"num_comments":1,"objectID":"47178063","points":1,"story_id":47178063,"story_text":"If you've been using Nano Banana Pro for image generation, you already know it's been the strongest model in its class since launch. Now Nano Banana 2 is coming, and early preview samples are already circulating. Here's what we know so far and how to get access to it
What\u2019s New & Upgraded\n1. Pro Intelligence at Flash Speed and price: NB2 delivers quality close to Pro while keeping generation times in line with the Flash tier, standard generations clock in at under two seconds, with pricing significantly lower than Nano Banana Pro.\n2. Precision Text Rendering: Inheriting major upgrades from the Pro version, NB2 renders crisp, accurate text for marketing assets, UI/UX wireframes, and product labels. It also supports seamless translation and localization of text already existing within an image.\n3. Large-Scale Subject Consistency: Maintain identity across up to 5 characters and 14 distinct objects within a single workflow. This is a massive leap compared to the original Nano Banana.\n4. Real-World Grounding via Search The model leverages Google Search live to accurately render specific logos, landmarks, and real-world products. It can even transform handwritten notes or rough sketches into data-driven charts and infographics.
Use Cases\n1. Film & Creative Visuals: Fantasy concept design for pre-production, emphasizing the photorealistic grounding of surreal elements and logical physics.\n2. Marketing & E-Commerce Advertising: Rapid generation of product posters featuring precise typography and realistic materials.\n3. Documentary Photography & High-Fidelity Character Creation: Cinematic character settings or humanistic documentary projects. Focuses on skin texture, micro-details (e.g., hair, iris reflections), and authentic emotional delivery, breaking away from the "plastic" look of traditional AI generation.\n4. Illustration Design & Native Text Rendering: Educational materials, vintage posters, or graphic design assets. The core pain point is accurate spelling and seamless integration of typography into specific artistic styles.
Cost Check\nNano Banana 2\n- Official: $0.080\n- AtlasCloud.ai: $0.040\n- Fal AI: $0.080\n- Wavespeed: $0.080\n- Higgsfield: $0.057\nNano Banana Pro\n- Official: $0.139 - $0.240\n- AtlasCloud.ai: $0.063\n- Fal AI: $0.150\n- Wavespeed: $0.140\n- Higgsfield: $0.076\nSeedream 5.0 Lite\n- Official: $0.035\n- AtlasCloud.ai: $0.026\n- Fal AI: $0.035\n- Wavespeed: $0.035\n- Higgsfield: $0.038\nSeedream 4.5\n- Official: $0.040\n- AtlasCloud.ai: $0.038\n- Fal AI: $0.040\n- Wavespeed: $0.040\n- Higgsfield: $0.038
How to access Nano Banana 2\n- For creator: Gemini, Lovart.ai\uff0cHiggsfield AI, arena ai\uff0cVertex AI \n- For developer\uff1aAtlasCloud.ai, google ai studio
TL;DR\nNano Banana 2 is undeniably sending shockwaves through the AI image gen scene and it\u2019s poised to become one of the most popular tools. For everyday users and creators like us, this is truly the golden age:D","title":"Nano Banana 2 Is Really Coming! Here's How to Access It Early","updated_at":"2026-03-05T23:38:23Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"johnpolacek"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"I wrote a screenplay a couple years ago and decided to see if I could actually \u201cfilm it\u201d with AI video tools. Deviant is the result, an 84 minute sci-fi thriller about a tech mogul's clone that finds an exploit in its own safety code.\nThe process: Character and set reference images in Midjourney and Nano Banana. Scene by scene prompts written from the script (I have an agent skill now for this), generated in Google Veo for most of the film, until it stopped working reliably, then I shelved it. Now with Seedance 2.5 available in the US on Higgsfield, I completed the final act. The last 20 minutes look noticeably better than the first 60.\nEdited just in iMovie, audio with ElevenLabs.\nI decided to release it as is rather than regenerate everything. Biggest problems were getting the actors to \u201cact\u201d, shot composition, character consistency, voice matching, and getting more than two characters to interact coherently in one frame. Happy to go into detail on any of it."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Deviant, a feature-length sci-fi thriller about AI, made with AI"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://deviantmovie.com/"}},"_tags":["story","author_johnpolacek","story_49673745","show_hn"],"author":"johnpolacek","children":[49693680],"created_at":"2026-09-12T16:02:08Z","created_at_i":1789228928,"num_comments":0,"objectID":"49673745","points":1,"story_id":49673745,"story_text":"I wrote a screenplay a couple years ago and decided to see if I could actually \u201cfilm it\u201d with AI video tools. Deviant is the result, an 84 minute sci-fi thriller about a tech mogul's clone that finds an exploit in its own safety code.\nThe process: Character and set reference images in Midjourney and Nano Banana. Scene by scene prompts written from the script (I have an agent skill now for this), generated in Google Veo for most of the film, until it stopped working reliably, then I shelved it. Now with Seedance 2.5 available in the US on Higgsfield, I completed the final act. The last 20 minutes look noticeably better than the first 60.\nEdited just in iMovie, audio with ElevenLabs.\nI decided to release it as is rather than regenerate everything. Biggest problems were getting the actors to \u201cact\u201d, shot composition, character consistency, voice matching, and getting more than two characters to interact coherently in one frame. Happy to go into detail on any of it.","title":"Show HN: Deviant, a feature-length sci-fi thriller about AI, made with AI","updated_at":"2026-09-14T08:35:19Z","url":"https://deviantmovie.com/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"naxtsass"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Hi HN,
Like many of you, I've been tracking the "Sora-level" video models coming out of ByteDance and Kuaishou. While the tech is impressive, the accessibility often sucks\u2014many are locked behind mobile apps (like Higgsfield) or fragmented invite-only portals.
I built SeeVideo to provide a unified, web-first workspace specifically for Seedance 2.0 and Kling 3.0.
Why I built this:
The Web Gap: Many users are searching for "Higgsfield Online" only to find it's mobile-only. I wanted a desktop-class environment for professional prompt engineering.
Model Nuances: Seedance 2.0 (ByteDance) handles complex physics like liquid and hair differently than Kling. I\u2019ve included a "Prompt Transformer" to help adapt generic prompts into the specific "dialects" these models prefer.
Side-by-Side Benchmarking: It\u2019s hard to judge "cinematic quality" without a direct comparison. This tool lets you see how they stack up on the same prompt.
Technical Observations:\nFrom running ~1000 generations, Seedance 2.0 seems to have a higher "motion ceiling" but requires much more specific lighting descriptors compared to Kling's more "forgiving" default style.
I\u2019d love to get the community's feedback on the UI and hear about your experience with the temporal consistency of these latest Chinese diffusion models.
Check it out here: https://seevideo.dance/"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: SeeVideo A web-first workspace to benchmark Seedance 2.0 vs. Kling 3.0"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://seevideo.dance/"}},"_tags":["story","author_naxtsass","story_47230843","show_hn"],"author":"naxtsass","created_at":"2026-03-03T11:15:45Z","created_at_i":1772536545,"num_comments":0,"objectID":"47230843","points":1,"story_id":47230843,"story_text":"Hi HN,
Like many of you, I've been tracking the "Sora-level" video models coming out of ByteDance and Kuaishou. While the tech is impressive, the accessibility often sucks\u2014many are locked behind mobile apps (like Higgsfield) or fragmented invite-only portals.
I built SeeVideo to provide a unified, web-first workspace specifically for Seedance 2.0 and Kling 3.0.
Why I built this:
The Web Gap: Many users are searching for "Higgsfield Online" only to find it's mobile-only. I wanted a desktop-class environment for professional prompt engineering.
Model Nuances: Seedance 2.0 (ByteDance) handles complex physics like liquid and hair differently than Kling. I\u2019ve included a "Prompt Transformer" to help adapt generic prompts into the specific "dialects" these models prefer.
Side-by-Side Benchmarking: It\u2019s hard to judge "cinematic quality" without a direct comparison. This tool lets you see how they stack up on the same prompt.
Technical Observations:\nFrom running ~1000 generations, Seedance 2.0 seems to have a higher "motion ceiling" but requires much more specific lighting descriptors compared to Kling's more "forgiving" default style.
I\u2019d love to get the community's feedback on the UI and hear about your experience with the temporal consistency of these latest Chinese diffusion models.
Check it out here: https://seevideo.dance/","title":"Show HN: SeeVideo A web-first workspace to benchmark Seedance 2.0 vs. Kling 3.0","updated_at":"2026-03-05T23:40:28Z","url":"https://seevideo.dance/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"fcpguru"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"Idiogram excels at text rendering
https://ideogram.ai/
Nano Banana - Photoshop-like capabilities for free
https://nanobanana.ai/
Sea Dance offers multi-shot storytelling
https://seed.bytedance.com/en/seedance
Runway's ALF feature allows precise video editing for under $1 per video
https://runwayml.com/research/introducing-runway-aleph
Higsfield provides 60+ camera
https://higgsfield.ai/
Invideo creates complete videos with narration,
https://invideo.io/
Wavespeed provides access to multiple generators
https://wavespeed.ai/
Free tools like Google Vids and Nano Banana compete with expensive professional software
https://workspace.google.com/products/vids/
from:
https://andrewarrow.dev/podpapyrus/summaries/nano-banana-runway-aleph-more-the-ultimate-ai-image-video.html"},"title":{"matchLevel":"none","matchedWords":[],"value":"Tell HN: Latest AI Video Tools"}},"_tags":["story","author_fcpguru","story_45102290","ask_hn"],"author":"fcpguru","created_at":"2025-09-02T12:34:25Z","created_at_i":1756816465,"num_comments":0,"objectID":"45102290","points":1,"story_id":45102290,"story_text":"Idiogram excels at text rendering
https://ideogram.ai/
Nano Banana - Photoshop-like capabilities for free
https://nanobanana.ai/
Sea Dance offers multi-shot storytelling
https://seed.bytedance.com/en/seedance
Runway's ALF feature allows precise video editing for under $1 per video
https://runwayml.com/research/introducing-runway-aleph
Higsfield provides 60+ camera
https://higgsfield.ai/
Invideo creates complete videos with narration,
https://invideo.io/
Wavespeed provides access to multiple generators
https://wavespeed.ai/
Free tools like Google Vids and Nano Banana compete with expensive professional software
https://workspace.google.com/products/vids/
from:
https://andrewarrow.dev/podpapyrus/summaries/nano-banana-runway-aleph-more-the-ultimate-ai-image-video.html","title":"Tell HN: Latest AI Video Tools","updated_at":"2026-03-05T22:36:47Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"codentropy"},"title":{"matchLevel":"none","matchedWords":[],"value":"Reinforcement Learning: From Zero to State of the Art with Pytorch 4"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/RL-Adventure-2"}},"_tags":["story","author_codentropy","story_17261063"],"author":"codentropy","children":[17261356,17261803,17261872,17263625,17263813,17264081],"created_at":"2018-06-07T22:51:13Z","created_at_i":1528411873,"num_comments":14,"objectID":"17261063","points":238,"story_id":17261063,"title":"Reinforcement Learning: From Zero to State of the Art with Pytorch 4","updated_at":"2024-09-20T02:37:35Z","url":"https://github.com/higgsfield/RL-Adventure-2"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Capsule Network tutorial with clean readable code in Pytorch"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/Capsule-Network-Tutorial"}},"_tags":["story","author_higgsfield","story_16105750"],"author":"higgsfield","children":[16108406,16128714],"created_at":"2018-01-09T12:48:33Z","created_at_i":1515502113,"num_comments":2,"objectID":"16105750","points":76,"story_id":16105750,"title":"Capsule Network tutorial with clean readable code in Pytorch","updated_at":"2024-09-20T01:48:49Z","url":"https://github.com/higgsfield/Capsule-Network-Tutorial"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Combinatorial optimization with reinforcement learning"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/np-hard-deep-reinforcement-learning"}},"_tags":["story","author_higgsfield","story_16006572"],"author":"higgsfield","children":[16007512],"created_at":"2017-12-26T01:42:37Z","created_at_i":1514252557,"num_comments":2,"objectID":"16006572","points":28,"story_id":16006572,"title":"Combinatorial optimization with reinforcement learning","updated_at":"2024-09-20T01:47:41Z","url":"https://github.com/higgsfield/np-hard-deep-reinforcement-learning"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zdotbirch"},"title":{"matchLevel":"none","matchedWords":[],"value":"We Hardened an AI Security Platform Against 16 Critical Vulnerabilities"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://aegis-security.higgsfield.app"}},"_tags":["story","author_zdotbirch","story_49095787"],"author":"zdotbirch","children":[49096417],"created_at":"2026-07-29T10:56:00Z","created_at_i":1785322560,"num_comments":2,"objectID":"49095787","points":9,"story_id":49095787,"title":"We Hardened an AI Security Platform Against 16 Critical Vulnerabilities","updated_at":"2026-07-30T01:49:11Z","url":"https://aegis-security.higgsfield.app"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: DeepMind's Imagination Augmented Agents step-by-step implementation"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/Imagination-Augmented-Agents"}},"_tags":["story","author_higgsfield","story_16715004","show_hn"],"author":"higgsfield","created_at":"2018-03-30T10:26:15Z","created_at_i":1522405575,"num_comments":0,"objectID":"16715004","points":7,"story_id":16715004,"title":"Show HN: DeepMind's Imagination Augmented Agents step-by-step implementation","updated_at":"2024-09-20T02:11:40Z","url":"https://github.com/higgsfield/Imagination-Augmented-Agents"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"frays"},"title":{"matchLevel":"none","matchedWords":[],"value":"First ever AI feature film premieres at the Cannes Film Festival"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://www.cgmagonline.com/news/hell-grind-made-only-with-higgsfield-ai/"}},"_tags":["story","author_frays","story_48231869"],"author":"frays","created_at":"2026-05-22T04:14:37Z","created_at_i":1779423277,"num_comments":0,"objectID":"48231869","points":4,"story_id":48231869,"title":"First ever AI feature film premieres at the Cannes Film Festival","updated_at":"2026-05-22T07:07:53Z","url":"https://www.cgmagonline.com/news/hell-grind-made-only-with-higgsfield-ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thelastgallon"},"title":{"matchLevel":"none","matchedWords":[],"value":"The Cully Hill Boys \u2013 110 minute AI film"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield.ai/original-series/cully-hill-boys/full-film"}},"_tags":["story","author_thelastgallon","story_49483809"],"author":"thelastgallon","children":[49483810,49484677],"created_at":"2026-08-28T20:27:28Z","created_at_i":1787948848,"num_comments":2,"objectID":"49483809","points":3,"story_id":49483809,"title":"The Cully Hill Boys \u2013 110 minute AI film","updated_at":"2026-08-31T18:36:23Z","url":"https://higgsfield.ai/original-series/cully-hill-boys/full-film"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"RL-Adventure-2: PyTorch tutorial of a2c/ppo/acer/ddpg/td3/soft a2c/gail/her"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/RL-Adventure-2"}},"_tags":["story","author_higgsfield","story_17165118"],"author":"higgsfield","created_at":"2018-05-27T02:02:54Z","created_at_i":1527386574,"num_comments":0,"objectID":"17165118","points":3,"story_id":17165118,"title":"RL-Adventure-2: PyTorch tutorial of a2c/ppo/acer/ddpg/td3/soft a2c/gail/her","updated_at":"2024-09-20T02:33:18Z","url":"https://github.com/higgsfield/RL-Adventure-2"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Practical Reinforcement Learning with clean readable code in Pytorch"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/RL-Adventure"}},"_tags":["story","author_higgsfield","story_16668003"],"author":"higgsfield","created_at":"2018-03-24T18:24:48Z","created_at_i":1521915888,"num_comments":0,"objectID":"16668003","points":3,"story_id":16668003,"title":"Practical Reinforcement Learning with clean readable code in Pytorch","updated_at":"2024-09-20T02:16:44Z","url":"https://github.com/higgsfield/RL-Adventure"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"arpanetus"},"title":{"matchLevel":"none","matchedWords":[],"value":"A Comment Generator for Your Hacker News Posts"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield-ai.github.io/"}},"_tags":["story","author_arpanetus","story_38299886"],"author":"arpanetus","created_at":"2023-11-17T05:18:18Z","created_at_i":1700198298,"num_comments":0,"objectID":"38299886","points":2,"story_id":38299886,"title":"A Comment Generator for Your Hacker News Posts","updated_at":"2024-09-20T15:43:59Z","url":"https://higgsfield-ai.github.io/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"codentropy"},"title":{"matchLevel":"none","matchedWords":[],"value":"Reinforcement Learning Tutorial: From Zero to State of the Art"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/RL-Adventure-2"}},"_tags":["story","author_codentropy","story_17225620"],"author":"codentropy","created_at":"2018-06-04T08:21:24Z","created_at_i":1528100484,"num_comments":0,"objectID":"17225620","points":2,"story_id":17225620,"title":"Reinforcement Learning Tutorial: From Zero to State of the Art","updated_at":"2024-09-20T02:33:56Z","url":"https://github.com/higgsfield/RL-Adventure-2"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Interaction Networks for Learning about Objects, Relations and Physics. Pytorch"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/interaction_network_pytorch"}},"_tags":["story","author_higgsfield","story_16111089"],"author":"higgsfield","created_at":"2018-01-09T23:17:59Z","created_at_i":1515539879,"num_comments":0,"objectID":"16111089","points":2,"story_id":16111089,"title":"Interaction Networks for Learning about Objects, Relations and Physics. Pytorch","updated_at":"2024-09-20T01:49:31Z","url":"https://github.com/higgsfield/interaction_network_pytorch"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: DeepMind's Interaction Network in Pytorch"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/interaction_network_pytorch"}},"_tags":["story","author_higgsfield","story_16013743","show_hn"],"author":"higgsfield","created_at":"2017-12-27T09:13:07Z","created_at_i":1514365987,"num_comments":0,"objectID":"16013743","points":2,"story_id":16013743,"title":"Show HN: DeepMind's Interaction Network in Pytorch","updated_at":"2024-09-20T01:48:21Z","url":"https://github.com/higgsfield/interaction_network_pytorch"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"How to design API of Machine Learning library"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://higgsfield.substack.com/p/how-to-design-api-of-machine-learning"}},"_tags":["story","author_higgsfield","story_37916185"],"author":"higgsfield","created_at":"2023-10-17T14:59:51Z","created_at_i":1697554791,"num_comments":0,"objectID":"37916185","points":1,"story_id":37916185,"title":"How to design API of Machine Learning library","updated_at":"2024-09-20T15:21:34Z","url":"https://higgsfield.substack.com/p/how-to-design-api-of-machine-learning"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"amashrabov"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Multi node training of Llama 70B without crying"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["higgsfield"],"value":"https://github.com/higgsfield/higgsfield"}},"_tags":["story","author_amashrabov","story_37877459","show_hn"],"author":"amashrabov","created_at":"2023-10-14T01:53:00Z","created_at_i":1697248380,"num_comments":0,"objectID":"37877459","points":1,"story_id":37877459,"title":"Show HN: Multi node training of Llama 70B without crying","updated_at":"2024-09-20T15:27:51Z","url":"https://github.com/higgsfield/higgsfield"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"A computational model of the Moth Olfactory Network learns to read MNIST [pdf]"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://openreview.net/pdf?id=HyYuqoCUz"}},"_tags":["story","author_higgsfield","story_16363479"],"author":"higgsfield","children":[16366588,16367035,16368173,16369517,16370323],"created_at":"2018-02-12T23:48:03Z","created_at_i":1518479283,"num_comments":14,"objectID":"16363479","points":71,"story_id":16363479,"title":"A computational model of the Moth Olfactory Network learns to read MNIST [pdf]","updated_at":"2024-09-20T02:05:08Z","url":"https://openreview.net/pdf?id=HyYuqoCUz"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"AI Chatbots can guess your personal information from what you type"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.wired.com/story/ai-chatbots-can-guess-your-personal-information/"}},"_tags":["story","author_higgsfield","story_37942017"],"author":"higgsfield","children":[37952968],"created_at":"2023-10-19T12:37:30Z","created_at_i":1697719050,"num_comments":1,"objectID":"37942017","points":9,"story_id":37942017,"title":"AI Chatbots can guess your personal information from what you type","updated_at":"2024-09-20T15:24:50Z","url":"https://www.wired.com/story/ai-chatbots-can-guess-your-personal-information/"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"51-qubit quantum computer"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/1707.04344"}},"_tags":["story","author_higgsfield","story_14801182"],"author":"higgsfield","created_at":"2017-07-19T00:23:23Z","created_at_i":1500423803,"num_comments":0,"objectID":"14801182","points":3,"story_id":14801182,"title":"51-qubit quantum computer","updated_at":"2024-09-20T01:03:27Z","url":"https://arxiv.org/abs/1707.04344"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Conditional Probability Theory (For Scientists and Engineers)"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://betanalpha.github.io/assets/case_studies/conditional_probability_theory.html"}},"_tags":["story","author_higgsfield","story_18247315"],"author":"higgsfield","created_at":"2018-10-18T11:01:06Z","created_at_i":1539860466,"num_comments":0,"objectID":"18247315","points":2,"story_id":18247315,"title":"Conditional Probability Theory (For Scientists and Engineers)","updated_at":"2024-09-20T03:10:47Z","url":"https://betanalpha.github.io/assets/case_studies/conditional_probability_theory.html"},{"_highlightResult":{"author":{"fullyHighlighted":true,"matchLevel":"full","matchedWords":["higgsfield"],"value":"higgsfield"},"title":{"matchLevel":"none","matchedWords":[],"value":"Russian YouTuber faces 5 years in prison for playing Pok\u00e9mon Go in a church"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.theguardian.com/technology/2016/sep/05/pokemon-go-russian-youtuber-ruslan-sokolovsky-five-years-jail-church"}},"_tags":["story","author_higgsfield","story_12686726"],"author":"higgsfield","created_at":"2016-10-11T17:50:22Z","created_at_i":1476208222,"num_comments":0,"objectID":"12686726","points":1,"story_id":12686726,"title":"Russian YouTuber faces 5 years in prison for playing Pok\u00e9mon Go in a church","updated_at":"2024-09-19T23:51:24Z","url":"https://www.theguardian.com/technology/2016/sep/05/pokemon-go-russian-youtuber-ruslan-sokolovsky-five-years-jail-church"}],"hitsPerPage":50,"nbHits":54,"nbPages":2,"page":0,"params":"query=Higgsfield&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":21,"processingTimingsMS":{"_request":{"roundTrip":17},"afterFetch":{"format":{"highlighting":1,"total":2},"merge":{"total":1},"total":1},"fetch":{"query":17,"scanning":1,"total":19},"total":21},"query":"Higgsfield","serverTimeMS":24}