Topics
Latest
AI
Amazon
Image Credits:Getty Images
Apps
Biotech & Health
Climate
Image Credits:Getty Images
Cloud Computing
commercialism
Crypto
Image Credits:Higgsfield
endeavor
EVs
Fintech
Image Credits:Higgsfield
Fundraising
widget
punt
Image Credits:Higgsfield
Government & Policy
ironware
layoff
Media & Entertainment
Meta
Microsoft
privateness
Robotics
Security
societal
Space
Startups
TikTok
Department of Transportation
speculation
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
newssheet
Podcasts
television
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
OpenAI captivated the tech world a few months back with a reproductive AI poser , Sora , that turns scene descriptions into original videos — no cameras or film crews require . But Sora has so far been tightly gated , and the business firm seems to be aiming it toward well - funded creatives likeHollywood manager — not hobbyists or low - time vender , necessarily .
Alex Mashrabov , the former head of generative AI at Snap , sensed an opportunity . So he launchedHiggsfield AI , an AI - power picture creation and editing chopine designed for more tailor , personalized applications .
Powered by a impost text - to - video model , Higgsfield ’s first app , Diffuse , can generate video from scratch or take a selfie and generate a clip star that person .
“ Our target audience is Creator of all types , ” Mashrabov told TechCrunch in an interview , “ from regular users who want to create fun content with their Quaker to social content creators looking to hear a new contented format to societal media trafficker who want their make to digest out . ”
Mashrabov came to Snap by direction ofAI Factory , his previous inauguration , which Snap acquire in 2020 for $ 166 million . While at Snap , Mashrabov aid build intersection like AR effects and filters for Snapchat , includingCameos , as well as Snapchat ’s controversialMyAIchabot .
Higgsfield — which Mashrabov co - launch several calendar month ago with Yerzat Dulat , an AI researcher narrow in generative TV — offers a curated readiness of pre - generated clips , a shaft to upload quotation media ( i.e. images and videos ) and a immediate editor that let user describe the characters , actions and scene they wish to depict . Using Diffuse , users can stick in themselves directly into an AI - generate scene , or have their digital likeness mimic things — like dancing motion — capture in other video .
“ Our model support highly naturalistic movements and locution , ” Mashrabov say . “ We ’re pioneering ‘ world models ’ for consumers , which will allow us to build good - in - class TV genesis and editing with a great storey of control . ”
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
Higgsfield is n’t the only generative video recording inauguration go school principal to head with OpenAI . Runway was one of the first on the view , and its pecker continue to improve . There ’s also Haiper , which has the backing of two DeepMind alums and over $ 13 million in speculation cash .
Mashrabov argues that Diffuse will fend out thanks to its mobile - first , social - forward go - to - market strategy .
“ By prioritize iOS and Android apps instead of desktop workflows , we enable creators to make compelling social media content anytime and anywhere , ” Mashrabov said . “ Indeed , by build on Mobile River , we ’re able to prioritize ease of use and consumer - friendly feature film from sidereal day one . ”
Higgsfield is also running skimpy . Mashrabov enunciate that the procreative fashion model bear out the political program were developed by a 16 - person team in less than nine months and prepare on a cluster of 32 GPUs ( 32 GPUs might fathom like a wad , but look at OpenAI uses tens of thousands , it ’s notreally ) . And Higgsfield has only raised $ 8 million to date , the mass of which came from a recent seed backing tranche led by Menlo Ventures .
To last out one step in front of rivals , Higgsfield plan to put the seed cash toward building an improved video editor that ’ll let drug user modify characters and object in videos , and toward train more powerful video coevals model specifically for societal medium use case . In fact , Mashrabov sees social mass medium — and societal media selling — as Higgsfield ’s principle money - making niche .
While Diffuse is currently detached to use , Mashrabov envisions a future where marketers pay some sort of fee or subscription for premium features , or for volume or large - scale campaigns .
“ We trust Higgsfield unlocks an unbelievable level of realism and depicted object production use cases for societal medium marketers , ” he state . “ We constantly hear from CMOs and creative directors that they need to optimize content output budgets and shorten timeline while still have impactful message . So we trust video generative AI solution will be a heart solution in helping them to achieve it . ”
Of course , Higgsfield is n’t immune from the broad challenge facing generative AI startup .
It ’s well - make that productive AI model like the sort powering Diffuse can “ regurgitate ” training data . Why ’s that problematic ? Well , if the good example were trained on copyright content without permission or some sort of licensing agreement in situation , those good example ’ users could unwittingly generate a copyright - infringe employment — uncover them to lawsuit .
Mashrabov would n’t unwrap the source of Higgsfield ’s preparation data ( other than to say it come from “ multiple publicly useable ” places ) , and also would n’t say whether Higgsfield would retain user datum to train succeeding manakin , which might not sit down decently with some business customers . He did note that Diffuse users can request that their data point be cancel at any fourth dimension through the app .
Digital “ cloning ” platforms like Higgsfield are also advanced for revilement , as the wildfire spread of deepfakes on social media in recent months has shown .
In a similar vein , Higgsfield could make it well-fixed to steal creators ’ mental object . For example , one need only upload a video of someone ’s stage dancing to generate a video of themselves perform that same choreography .
I asked Mashrabov about what safeguards or protections Higgsfield might be using to attempt to prevent abuse , and — while he would n’t go into specifics — he take that the program employ a mix of automatise and manual relief .
“ We ’ve decided to bit by bit roll out the product and run in prize markets first , so that we can supervise where there ’s the electric potential for abuse and evolve the product as necessary , ” Mashrabov lend .
We ’ll have to wait and see how well that act upon in practice .
