Multimodal general intelligence for work and creation. Bridging digital and physical worlds.
Hong Kong
Joined July 2018
- Tweets924
- Following32
- Followers5.7K
- Likes601
SenseTime retweeted
SenseNova U1.5 is here. Exciting to see such a big progress since U1.
๐ฏ Highly controllable: Complex conditioned generation and precise editing
๐ผ๏ธ High resolution: Stunning 4K output with fine-grained details
โก Highly cost-effective: An 8B-MoT model that rivals the latest commercial models
Smaller model. Powerful performance. Limitless creativity.
๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ๐ญ.๐ฑ ๐๐ถ๐๐ฒ โ ๐ฆ๐ต๐ฎ๐ฟ๐ฝ๐ฒ๐ฟ. ๐ ๐ผ๐ฟ๐ฒ ๐๐ผ๐ป๐๐ฟ๐ผ๐น๐น๐ฎ๐ฏ๐น๐ฒ. ๐ ๐ผ๐ฟ๐ฒ ๐๐ผ๐๐-๐๐ณ๐ณ๐ถ๐ฐ๐ถ๐ฒ๐ป๐. An ๐ผ๐ฝ๐ฒ๐ป-๐๐ผ๐๐ฟ๐ฐ๐ฒ, lightweight native unified multimodal model for visual understanding, generation & editing.
๐๐๐ถ๐น๐ ๐ณ๐ผ๐ฟ ๐ฟ๐ฒ๐ฎ๐น-๐๐ผ๐ฟ๐น๐ฑ ๐๐ถ๐๐๐ฎ๐น ๐ฎ๐ฟ๐๐ถ๐ณ๐ฎ๐ฐ๐ ๐๐ผ๐ฟ๐ธ๐ณ๐น๐ผ๐๐:
๐น๐ก๐ฎ๐๐ถ๐๐ฒ ๐ฐ๐ ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป with better detail, composition & realism
๐น๐๐ผ๐บ๐ฝ๐น๐ฒ๐
๐ถ๐ป๐๐๐ฟ๐๐ฐ๐๐ถ๐ผ๐ป ๐ณ๐ผ๐น๐น๐ผ๐๐ถ๐ป๐ด across subjects, counts, text, layouts & styles
๐น๐๐ป๐ต๐ฎ๐ป๐ฐ๐ฒ๐ฑ ๐๐ต๐ถ๐ป๐ฒ๐๐ฒ/๐๐ป๐ด๐น๐ถ๐๐ต ๐๐ฒ๐
๐ ๐ฟ๐ฒ๐ป๐ฑ๐ฒ๐ฟ๐ถ๐ป๐ด & ๐บ๐๐น๐๐ถ-๐๐ฒ๐
๐ ๐น๐ฎ๐๐ผ๐๐
๐น๐ฃ๐ฟ๐ฒ๐ฐ๐ถ๐๐ฒ ๐ถ๐บ๐ฎ๐ด๐ฒ ๐ฒ๐ฑ๐ถ๐๐ถ๐ป๐ด & ๐ฐ๐ผ๐ป๐๐ฟ๐ผ๐น with visual markers, bounding boxes & multi-image references
Just ๐ด๐ parameters โ outperforming same-size models in instruction following and editing preservation, while ๐ฐ๐ผ๐บ๐ฝ๐ฒ๐๐ถ๐ป๐ด ๐๐ถ๐๐ต ๐น๐ฎ๐ฟ๐ด๐ฒ ๐ฐ๐ผ๐บ๐บ๐ฒ๐ฟ๐ฐ๐ถ๐ฎ๐น ๐บ๐ผ๐ฑ๐ฒ๐น๐ ๐ถ๐ป ๐๐ฒ๐
๐ ๐ฟ๐ฒ๐ป๐ฑ๐ฒ๐ฟ๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐
๐น๐ฎ๐๐ผ๐๐๐.
๐ ๏ธGitHub: github.com/OpenSenseNova/Senโฆ
๐คHF: huggingface.co/collections/sโฆ
๐พDiscord: discord.com/invite/BuTXPHmQuโฆ
๐จSenseNova Studio: unify.light-ai.top/
SenseTime retweeted
The era of specialized vision models is officially over.
SenseNova-Vision is here, a single 7B open-source model that handles detection, segmentation, depth, and multi-view geometry through simple natural language prompts.
It treats vision as a native foundation-model capability, breaking down research silos and outperforming specialist systems like Vision Banana across the board.
Everything is open: weights, code, and a massive 50M-example dataset.
Go explore the repo and show some love by dropping a star!
๐ GitHub: github.com/OpenSenseNova/Senโฆ
๐ค HF: huggingface.co/sensenova/Senโฆ
๐ Paper: arxiv.org/abs/2607.06560
@SenseTime_AI
SenseTime retweeted
Created this running SenseNova-U1-8B-MoT-Infographic-V3 Q6_K GGUF on an Nvidia L1 GPU. Great for creating comics and infographics
SenseTime retweeted
Looks like another high-quality model release happened in China yesterday that no one in my timeline is talking about. U1 Pro from SenseTime seems to give GPT image 2 and Seed Dream a run for their money. Feature details in reply.
๐จ๐ป๐๐ฒ๐ถ๐น๐ถ๐ป๐ด ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฃ๐ฟ๐ผ
SenseTimeโs flagship native multimodal foundation model. Built on the ๐ก๐๐ข-๐จ๐ป๐ถ๐ณ๐ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ, U1 Pro unifies ๐๐ป๐ฑ๐ฒ๐ฟ๐๐๐ฎ๐ป๐ฑ๐ถ๐ป๐ด, ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป, ๐ฎ๐ป๐ฑ ๐ฎ๐ฐ๐๐ถ๐ผ๐ป with ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐ถ๐๐ถ๐ผ๐ป-๐น๐ฎ๐ป๐ด๐๐ฎ๐ด๐ฒ ๐๐ต๐ถ๐ป๐ธ๐ถ๐ป๐ด. The unification of these capabilities lays a ๐ฐ๐ฟ๐๐ฐ๐ถ๐ฎ๐น ๐ณ๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ณ๐ผ๐ฟ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐
๐๐ถ๐๐๐ฎ๐น ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป.
Over years, AI image creation has evolved from visuals that look right, to details that feel real. U1 Pro takes the next step by ๐ฏ๐ฟ๐ถ๐ป๐ด๐ถ๐ป๐ด ๐ฝ๐ฟ๐ผ๐ณ๐ฒ๐๐๐ถ๐ผ๐ป๐ฎ๐น ๐ฑ๐ฒ๐๐ถ๐ด๐ป ๐ฎ๐ฒ๐๐๐ต๐ฒ๐๐ถ๐ฐ๐ ๐๐ผ ๐๐ต๐ฒ ๐ถ๐บ๐ฎ๐ด๐ฒ as a whole.
Key upgrades over SenseNova U1 Lite:
โจ Native ๐น๐ผ๐ป๐ด-๐ต๐ผ๐ฟ๐ถ๐๐ผ๐ป ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐ฟ๐ฒ๐ฎ๐๐ผ๐ป๐ถ๐ป๐ด across dozens of rounds in an ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐๐ผ๐ผ๐ฝ
โจ Enhanced capabilities in ๐ถ๐ป๐๐ฒ๐ด๐ฟ๐ฎ๐๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ฎ๐ป๐ฎ๐น๐๐๐ถ๐ป๐ด ๐๐ฒ๐
๐๐๐ฎ๐น ๐ฎ๐ป๐ฑ ๐๐ถ๐๐๐ฎ๐น ๐ถ๐ป๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ถ๐ผ๐ป
โจ ๐ฅ๐ฒ๐น๐ถ๐ฎ๐ฏ๐น๐ฒ ๐ฐ๐ผ๐ป๐๐ฟ๐ผ๐น ๐ผ๐๐ฒ๐ฟ ๐๐ฒ๐
๐ ๐ฎ๐ป๐ฑ ๐ถ๐บ๐ฎ๐ด๐ฒ ๐ฑ๐ฒ๐๐ฎ๐ถ๐น๐, with a very low error rate in rendered texts
โจ Up to 8๐ ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ฟ๐ฒ๐๐ผ๐น๐๐๐ถ๐ผ๐ป output
โจ Enables ๐๐น๐๐ฟ๐ฎ-๐๐ถ๐ฑ๐ฒ and ๐๐น๐๐ฟ๐ฎ-๐๐ฎ๐น๐น image creation
From ๐ฏ๐๐๐ถ๐ป๐ฒ๐๐ ๐ฝ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป๐ and ๐ฏ๐ฟ๐ฎ๐ป๐ฑ ๐บ๐ฎ๐ฟ๐ธ๐ฒ๐๐ถ๐ป๐ด to ๐ฒ๐ฑ๐๐ฐ๐ฎ๐๐ถ๐ผ๐ป, ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ถ๐ป๐ด, ๐ณ๐ถ๐น๐บ, ๐ฎ๐ป๐ถ๐บ๐ฎ๐๐ถ๐ผ๐ป, and ๐ฑ๐ถ๐ด๐ถ๐๐ฎ๐น ๐ฐ๐ผ๐ป๐๐ฒ๐ป๐ ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป, U1 Pro empowers creators across industries.
๐ป Preview (invite-only) available now.
๐
Official launch with API + pricing: August 2026.
โจNewly Addedโจ ๐๐ฒ๐ฎ๐ฟ๐ป ๐บ๐ผ๐ฟ๐ฒ ๐ฎ๐ฏ๐ผ๐๐ ๐ผ๐๐ฟ ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฆ๐ฒ๐ฟ๐ถ๐ฒ๐:
github.com/OpenSenseNova/Senโฆ
SenseTime retweeted
SenseNova U1 Pro is SenseTime's flagship multimodal model โ trained at significantly larger scale than U1 Lite, with delivery-ready image creation that rivals top international models.
๐จ๐ป๐๐ฒ๐ถ๐น๐ถ๐ป๐ด ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฃ๐ฟ๐ผ
SenseTimeโs flagship native multimodal foundation model. Built on the ๐ก๐๐ข-๐จ๐ป๐ถ๐ณ๐ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ, U1 Pro unifies ๐๐ป๐ฑ๐ฒ๐ฟ๐๐๐ฎ๐ป๐ฑ๐ถ๐ป๐ด, ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป, ๐ฎ๐ป๐ฑ ๐ฎ๐ฐ๐๐ถ๐ผ๐ป with ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐ถ๐๐ถ๐ผ๐ป-๐น๐ฎ๐ป๐ด๐๐ฎ๐ด๐ฒ ๐๐ต๐ถ๐ป๐ธ๐ถ๐ป๐ด. The unification of these capabilities lays a ๐ฐ๐ฟ๐๐ฐ๐ถ๐ฎ๐น ๐ณ๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ณ๐ผ๐ฟ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐
๐๐ถ๐๐๐ฎ๐น ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป.
Over years, AI image creation has evolved from visuals that look right, to details that feel real. U1 Pro takes the next step by ๐ฏ๐ฟ๐ถ๐ป๐ด๐ถ๐ป๐ด ๐ฝ๐ฟ๐ผ๐ณ๐ฒ๐๐๐ถ๐ผ๐ป๐ฎ๐น ๐ฑ๐ฒ๐๐ถ๐ด๐ป ๐ฎ๐ฒ๐๐๐ต๐ฒ๐๐ถ๐ฐ๐ ๐๐ผ ๐๐ต๐ฒ ๐ถ๐บ๐ฎ๐ด๐ฒ as a whole.
Key upgrades over SenseNova U1 Lite:
โจ Native ๐น๐ผ๐ป๐ด-๐ต๐ผ๐ฟ๐ถ๐๐ผ๐ป ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐ฟ๐ฒ๐ฎ๐๐ผ๐ป๐ถ๐ป๐ด across dozens of rounds in an ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐๐ผ๐ผ๐ฝ
โจ Enhanced capabilities in ๐ถ๐ป๐๐ฒ๐ด๐ฟ๐ฎ๐๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ฎ๐ป๐ฎ๐น๐๐๐ถ๐ป๐ด ๐๐ฒ๐
๐๐๐ฎ๐น ๐ฎ๐ป๐ฑ ๐๐ถ๐๐๐ฎ๐น ๐ถ๐ป๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ถ๐ผ๐ป
โจ ๐ฅ๐ฒ๐น๐ถ๐ฎ๐ฏ๐น๐ฒ ๐ฐ๐ผ๐ป๐๐ฟ๐ผ๐น ๐ผ๐๐ฒ๐ฟ ๐๐ฒ๐
๐ ๐ฎ๐ป๐ฑ ๐ถ๐บ๐ฎ๐ด๐ฒ ๐ฑ๐ฒ๐๐ฎ๐ถ๐น๐, with a very low error rate in rendered texts
โจ Up to 8๐ ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ฟ๐ฒ๐๐ผ๐น๐๐๐ถ๐ผ๐ป output
โจ Enables ๐๐น๐๐ฟ๐ฎ-๐๐ถ๐ฑ๐ฒ and ๐๐น๐๐ฟ๐ฎ-๐๐ฎ๐น๐น image creation
From ๐ฏ๐๐๐ถ๐ป๐ฒ๐๐ ๐ฝ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป๐ and ๐ฏ๐ฟ๐ฎ๐ป๐ฑ ๐บ๐ฎ๐ฟ๐ธ๐ฒ๐๐ถ๐ป๐ด to ๐ฒ๐ฑ๐๐ฐ๐ฎ๐๐ถ๐ผ๐ป, ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ถ๐ป๐ด, ๐ณ๐ถ๐น๐บ, ๐ฎ๐ป๐ถ๐บ๐ฎ๐๐ถ๐ผ๐ป, and ๐ฑ๐ถ๐ด๐ถ๐๐ฎ๐น ๐ฐ๐ผ๐ป๐๐ฒ๐ป๐ ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป, U1 Pro empowers creators across industries.
๐ป Preview (invite-only) available now.
๐
Official launch with API + pricing: August 2026.
โจNewly Addedโจ ๐๐ฒ๐ฎ๐ฟ๐ป ๐บ๐ผ๐ฟ๐ฒ ๐ฎ๐ฏ๐ผ๐๐ ๐ผ๐๐ฟ ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฆ๐ฒ๐ฟ๐ถ๐ฒ๐:
github.com/OpenSenseNova/Senโฆ
SenseTime retweeted
Corporate infographics, commercial posters, educational illustrations, animation storyboards. SenseNova U1 Pro's delivery-grade quality makes it applicable across all of these.
๐จ๐ป๐๐ฒ๐ถ๐น๐ถ๐ป๐ด ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฃ๐ฟ๐ผ
SenseTimeโs flagship native multimodal foundation model. Built on the ๐ก๐๐ข-๐จ๐ป๐ถ๐ณ๐ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ, U1 Pro unifies ๐๐ป๐ฑ๐ฒ๐ฟ๐๐๐ฎ๐ป๐ฑ๐ถ๐ป๐ด, ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป, ๐ฎ๐ป๐ฑ ๐ฎ๐ฐ๐๐ถ๐ผ๐ป with ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐ถ๐๐ถ๐ผ๐ป-๐น๐ฎ๐ป๐ด๐๐ฎ๐ด๐ฒ ๐๐ต๐ถ๐ป๐ธ๐ถ๐ป๐ด. The unification of these capabilities lays a ๐ฐ๐ฟ๐๐ฐ๐ถ๐ฎ๐น ๐ณ๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ณ๐ผ๐ฟ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐
๐๐ถ๐๐๐ฎ๐น ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป.
Over years, AI image creation has evolved from visuals that look right, to details that feel real. U1 Pro takes the next step by ๐ฏ๐ฟ๐ถ๐ป๐ด๐ถ๐ป๐ด ๐ฝ๐ฟ๐ผ๐ณ๐ฒ๐๐๐ถ๐ผ๐ป๐ฎ๐น ๐ฑ๐ฒ๐๐ถ๐ด๐ป ๐ฎ๐ฒ๐๐๐ต๐ฒ๐๐ถ๐ฐ๐ ๐๐ผ ๐๐ต๐ฒ ๐ถ๐บ๐ฎ๐ด๐ฒ as a whole.
Key upgrades over SenseNova U1 Lite:
โจ Native ๐น๐ผ๐ป๐ด-๐ต๐ผ๐ฟ๐ถ๐๐ผ๐ป ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐ฟ๐ฒ๐ฎ๐๐ผ๐ป๐ถ๐ป๐ด across dozens of rounds in an ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐๐ผ๐ผ๐ฝ
โจ Enhanced capabilities in ๐ถ๐ป๐๐ฒ๐ด๐ฟ๐ฎ๐๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ฎ๐ป๐ฎ๐น๐๐๐ถ๐ป๐ด ๐๐ฒ๐
๐๐๐ฎ๐น ๐ฎ๐ป๐ฑ ๐๐ถ๐๐๐ฎ๐น ๐ถ๐ป๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ถ๐ผ๐ป
โจ ๐ฅ๐ฒ๐น๐ถ๐ฎ๐ฏ๐น๐ฒ ๐ฐ๐ผ๐ป๐๐ฟ๐ผ๐น ๐ผ๐๐ฒ๐ฟ ๐๐ฒ๐
๐ ๐ฎ๐ป๐ฑ ๐ถ๐บ๐ฎ๐ด๐ฒ ๐ฑ๐ฒ๐๐ฎ๐ถ๐น๐, with a very low error rate in rendered texts
โจ Up to 8๐ ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ฟ๐ฒ๐๐ผ๐น๐๐๐ถ๐ผ๐ป output
โจ Enables ๐๐น๐๐ฟ๐ฎ-๐๐ถ๐ฑ๐ฒ and ๐๐น๐๐ฟ๐ฎ-๐๐ฎ๐น๐น image creation
From ๐ฏ๐๐๐ถ๐ป๐ฒ๐๐ ๐ฝ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป๐ and ๐ฏ๐ฟ๐ฎ๐ป๐ฑ ๐บ๐ฎ๐ฟ๐ธ๐ฒ๐๐ถ๐ป๐ด to ๐ฒ๐ฑ๐๐ฐ๐ฎ๐๐ถ๐ผ๐ป, ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ถ๐ป๐ด, ๐ณ๐ถ๐น๐บ, ๐ฎ๐ป๐ถ๐บ๐ฎ๐๐ถ๐ผ๐ป, and ๐ฑ๐ถ๐ด๐ถ๐๐ฎ๐น ๐ฐ๐ผ๐ป๐๐ฒ๐ป๐ ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป, U1 Pro empowers creators across industries.
๐ป Preview (invite-only) available now.
๐
Official launch with API + pricing: August 2026.
โจNewly Addedโจ ๐๐ฒ๐ฎ๐ฟ๐ป ๐บ๐ผ๐ฟ๐ฒ ๐ฎ๐ฏ๐ผ๐๐ ๐ผ๐๐ฟ ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฆ๐ฒ๐ฟ๐ถ๐ฒ๐:
github.com/OpenSenseNova/Senโฆ
SenseTime retweeted
SenseNova U1 Pro natively supports long-horizon interleaved reasoning across dozens of rounds in an Agentic Generation Loop. That's a fundamentally different operating mode.
๐จ๐ป๐๐ฒ๐ถ๐น๐ถ๐ป๐ด ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฃ๐ฟ๐ผ
SenseTimeโs flagship native multimodal foundation model. Built on the ๐ก๐๐ข-๐จ๐ป๐ถ๐ณ๐ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ, U1 Pro unifies ๐๐ป๐ฑ๐ฒ๐ฟ๐๐๐ฎ๐ป๐ฑ๐ถ๐ป๐ด, ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป, ๐ฎ๐ป๐ฑ ๐ฎ๐ฐ๐๐ถ๐ผ๐ป with ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐ถ๐๐ถ๐ผ๐ป-๐น๐ฎ๐ป๐ด๐๐ฎ๐ด๐ฒ ๐๐ต๐ถ๐ป๐ธ๐ถ๐ป๐ด. The unification of these capabilities lays a ๐ฐ๐ฟ๐๐ฐ๐ถ๐ฎ๐น ๐ณ๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ณ๐ผ๐ฟ ๐ฐ๐ผ๐บ๐ฝ๐น๐ฒ๐
๐๐ถ๐๐๐ฎ๐น ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป.
Over years, AI image creation has evolved from visuals that look right, to details that feel real. U1 Pro takes the next step by ๐ฏ๐ฟ๐ถ๐ป๐ด๐ถ๐ป๐ด ๐ฝ๐ฟ๐ผ๐ณ๐ฒ๐๐๐ถ๐ผ๐ป๐ฎ๐น ๐ฑ๐ฒ๐๐ถ๐ด๐ป ๐ฎ๐ฒ๐๐๐ต๐ฒ๐๐ถ๐ฐ๐ ๐๐ผ ๐๐ต๐ฒ ๐ถ๐บ๐ฎ๐ด๐ฒ as a whole.
Key upgrades over SenseNova U1 Lite:
โจ Native ๐น๐ผ๐ป๐ด-๐ต๐ผ๐ฟ๐ถ๐๐ผ๐ป ๐ถ๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐ฟ๐ฒ๐ฎ๐๐ผ๐ป๐ถ๐ป๐ด across dozens of rounds in an ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐๐ผ๐ผ๐ฝ
โจ Enhanced capabilities in ๐ถ๐ป๐๐ฒ๐ด๐ฟ๐ฎ๐๐ถ๐ป๐ด ๐ฎ๐ป๐ฑ ๐ฎ๐ป๐ฎ๐น๐๐๐ถ๐ป๐ด ๐๐ฒ๐
๐๐๐ฎ๐น ๐ฎ๐ป๐ฑ ๐๐ถ๐๐๐ฎ๐น ๐ถ๐ป๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ถ๐ผ๐ป
โจ ๐ฅ๐ฒ๐น๐ถ๐ฎ๐ฏ๐น๐ฒ ๐ฐ๐ผ๐ป๐๐ฟ๐ผ๐น ๐ผ๐๐ฒ๐ฟ ๐๐ฒ๐
๐ ๐ฎ๐ป๐ฑ ๐ถ๐บ๐ฎ๐ด๐ฒ ๐ฑ๐ฒ๐๐ฎ๐ถ๐น๐, with a very low error rate in rendered texts
โจ Up to 8๐ ๐ป๐ฎ๐๐ถ๐๐ฒ ๐ฟ๐ฒ๐๐ผ๐น๐๐๐ถ๐ผ๐ป output
โจ Enables ๐๐น๐๐ฟ๐ฎ-๐๐ถ๐ฑ๐ฒ and ๐๐น๐๐ฟ๐ฎ-๐๐ฎ๐น๐น image creation
From ๐ฏ๐๐๐ถ๐ป๐ฒ๐๐ ๐ฝ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป๐ and ๐ฏ๐ฟ๐ฎ๐ป๐ฑ ๐บ๐ฎ๐ฟ๐ธ๐ฒ๐๐ถ๐ป๐ด to ๐ฒ๐ฑ๐๐ฐ๐ฎ๐๐ถ๐ผ๐ป, ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ถ๐ป๐ด, ๐ณ๐ถ๐น๐บ, ๐ฎ๐ป๐ถ๐บ๐ฎ๐๐ถ๐ผ๐ป, and ๐ฑ๐ถ๐ด๐ถ๐๐ฎ๐น ๐ฐ๐ผ๐ป๐๐ฒ๐ป๐ ๐ฐ๐ฟ๐ฒ๐ฎ๐๐ถ๐ผ๐ป, U1 Pro empowers creators across industries.
๐ป Preview (invite-only) available now.
๐
Official launch with API + pricing: August 2026.
โจNewly Addedโจ ๐๐ฒ๐ฎ๐ฟ๐ป ๐บ๐ผ๐ฟ๐ฒ ๐ฎ๐ฏ๐ผ๐๐ ๐ผ๐๐ฟ ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐ฆ๐ฒ๐ฟ๐ถ๐ฒ๐:
github.com/OpenSenseNova/Senโฆ
SenseTime retweeted
An important step towards emergent intelligence from visual learning.
SenseNova-Vision just dropped, which unifies a wide range of vision tasks into a single generative model.
Not only achieving competitive performance on all these tasks, but more importantly showing the capability of doing unseen and complex vision tasks.
๐ข๐ฝ๐ฒ๐ป-๐๐ผ๐๐ฟ๐ฐ๐ถ๐ป๐ด ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ-๐ฉ๐ถ๐๐ถ๐ผ๐ป-7๐-๐ ๐ผ๐ง โ ๐ผ๐ป๐ฒ ๐บ๐ผ๐ฑ๐ฒ๐น, ๐ฎ๐น๐น ๐บ๐ฎ๐ท๐ผ๐ฟ ๐๐ถ๐๐ถ๐ผ๐ป ๐๐ฎ๐๐ธ๐.
Reimagines CV as multimodal generation. Steerable by language or visual prompts. ๐ก๐ผ ๐๐ฎ๐๐ธ-๐๐ฝ๐ฒ๐ฐ๐ถ๐ณ๐ถ๐ฐ ๐ต๐ฒ๐ฎ๐ฑ๐ โ ๐๐๐ถ๐น๐น ๐ฆ๐ข๐ง๐, ๐๐๐ฟ๐ฝ๐ฎ๐๐๐ถ๐ป๐ด ๐๐ผ๐ผ๐ด๐น๐ฒ ๐๐ฒ๐ฒ๐ฝ๐ ๐ถ๐ป๐ฑ'๐ ๐ฉ๐ถ๐๐ถ๐ผ๐ป ๐๐ฎ๐ป๐ฎ๐ป๐ฎ ๐ผ๐ป ๐๐ฒ๐ด๐บ๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป ๐ฎ๐ป๐ฑ ๐ฑ๐ฒ๐ป๐๐ฒ ๐ด๐ฒ๐ผ๐บ๐ฒ๐๐ฟ๐.
๐ ๐๐. ๐ฉ๐ถ๐๐ถ๐ผ๐ป ๐๐ฎ๐ป๐ฎ๐ป๐ฎ (๐๐ผ๐ผ๐ด๐น๐ฒ ๐๐ฒ๐ฒ๐ฝ๐ ๐ถ๐ป๐ฑ):
๐นReferring segmentation (RefCOCOg cIoU): 80.3 vs 73.8
๐นSemantic segmentation (Cityscapes mIoU): 71.2 vs 69.9
๐นDepth estimation (NYUv2 ฮด1): 98.1 vs 94.8
๐นSurface normal (NYUv2 mean error): 14.4 vs 17.8
Built on a decade of SenseTime's CV leadership โ #1 in China's Vision AI market for 10 consecutive years, and named a Tech Innovator in global GenAI CV by Gartner. ๐ฉ๐ถ๐๐ถ๐ผ๐ป, ๐ป๐ผ๐ ๐ป๐ฎ๐๐ถ๐๐ฒ ๐๐ผ ๐ณ๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐บ๐ผ๐ฑ๐ฒ๐น๐. New tasks, defined by instruction โ not by new model heads.
SenseTime retweeted
SenseNova U1 now has an infographic enhanced model and an arXiv-style paper rendering from open source which is genuinely significant. Dense two column layouts, small text, citations.
Run it yourself:
huggingface.co/sensenova/Senโฆ
๐ค Paid partnership
๐๐ผ๐ ๐ผ๐๐ฟ ๐๐ฝ๐ด๐ฟ๐ฎ๐ฑ๐ฒ๐ฑ ๐ถ๐ป๐ณ๐ผ๐ด๐ฟ๐ฎ๐ฝ๐ต๐ถ๐ฐ ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐บ๐ผ๐ฑ๐ฒ๐น โ ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ-๐จ๐ญ-๐ด๐-๐ ๐ผ๐ง-๐๐ป๐ณ๐ผ๐ด๐ฟ๐ฎ๐ฝ๐ต๐ถ๐ฐ โ ๐ฑ๐ฒ๐น๐ถ๐๐ฒ๐ฟ๐ ๐ฒ๐๐ฒ๐ป ๐๐๐ฟ๐ผ๐ป๐ด๐ฒ๐ฟ ๐ฐ๐ฎ๐ฝ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ถ๐ฒ๐ ๐ช
โข ๐ง๐ฒ๐
๐ ๐ฎ๐ฐ๐ฐ๐๐ฟ๐ฎ๐ฐ๐ & ๐ฟ๐ฒ๐ฎ๐ฑ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ enhanced โ reduced repetition, avoided unnatural enlargement, and stronger support for small fonts
โข ๐๐ฎ๐๐ผ๐๐ ๐ฐ๐ผ๐ป๐๐ถ๐๐๐ฒ๐ป๐ฐ๐ & ๐ฟ๐ฎ๐๐ถ๐ผ๐ป๐ฎ๐น๐ถ๐๐ improved, with more stable backgrounds
โข ๐๐ต๐ฎ๐ฟ๐ & ๐ฑ๐ถ๐ฎ๐ด๐ฟ๐ฎ๐บ ๐พ๐๐ฎ๐น๐ถ๐๐ elevated
โข ๐๐ฐ๐ฎ๐ฑ๐ฒ๐บ๐ถ๐ฐ ๐ฟ๐ฒ๐ป๐ฑ๐ฒ๐ฟ๐ถ๐ป๐ด supported
Try it out:
๐ฅฐ huggingface.co/sensenova/Senโฆ
๐ผ Showcases: github.com/OpenSenseNova/Senโฆ
๐พ Discord: discord.gg/BuTXPHmQub
@huggingface @github
SenseTime retweeted
๐ฏ Generate nearly NanoBanana2 quality infographics at home for free.
An open-source Infographics LLM recently dropped to run locally. ๐ป
๐ SenseNova-U1-8B-MoT-Infographic (Apache 2.0) just landed on ModelScope today.
This 8B native unified MoT model (NEO-unify architecture) was fine-tuned specifically for dense visual communication:
โ
BizGenEval Hard/Easy: 46.6 / 65.4
โ
IGenBench Q-ACC: 69.5 (+18.2 over base)
โ
IGenBench I-ACC: 17.0 (+12.8 over base)
It beats GPT-Image-1.5 and Qwen-Image-2.0 on multiple infographic metrics while staying competitive with top commercial models.
๐ผ๏ธ 100+ styles โข complex layouts โข dense small text โข charts โข arXiv-style pages โข posters โข resumes
๐ป Runs locally on ~10-12 GB VRAM (Q4 GGUF + balanced mode - this is an estimate)
๐ Also works great with LightLLM + LightX2V
๐ข๐ข ๐๐ป๐๐ฟ๐ผ๐ฑ๐๐ฐ๐ถ๐ป๐ด ๐ฎ๐ป ๐ฒ๐ป๐ต๐ฎ๐ป๐ฐ๐ฒ๐ฑ ๐บ๐ผ๐ฑ๐ฒ๐น ๐ณ๐ผ๐ฟ ๐ถ๐ป๐ณ๐ผ๐ด๐ฟ๐ฎ๐ฝ๐ต๐ถ๐ฐ ๐ด๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป!
๐ Posters, charts, recipe cards, postcards โ even arXiv-style pages โ all from ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ-๐จ๐ญ-๐ด๐-๐ ๐ผ๐ง-๐๐ป๐ณ๐ผ๐ด๐ฟ๐ฎ๐ฝ๐ต๐ถ๐ฐ.
๐ +6.8 / +18.2 on BizGenEval (hard) / IGenBench (Q-ACC) over base U1, plus 100+ diverse showcases.
๐ค huggingface.co/sensenova/Senโฆ
๐ผ๏ธShowcases: github.com/OpenSenseNova/Senโฆ
Try it out โ we'd love to see what you build!
@huggingface
SenseTime retweeted
6/n SenseNova U1-8B-MoT is around 2x faster than Qwen-Image-2.0 / Seedream-4.5 while staying in the same rough quality band
Much faster generation, while staying close to Qwen-Image-2.0 and Seedream-4.5 on quality.
SenseTime retweeted
Meet SenseNova U1: @SenseTime_AI groundbreaking native unified multimodal model!
Built on the NEO-unify architecture, it fuses language & vision into one monolithic framework for seamless understanding & generation.
Experience the future of AI today!
SenseTime retweeted
While everyone is scaling up VE/VAE, SenseNova U1 goes VEโfree + MoT.
Full tech report now public.
And SenseNova-U1-A3B-MoT (38B-A3B MoE) weights are now open-sourced.
๐ฅ New week, New ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ-๐จ๐ญ Drop โ and this one goes Deep!๐ฅ
๐ ๐ง๐ต๐ฒ ๐ณ๐๐น๐น ๐ง๐ฒ๐ฐ๐ต๐ป๐ถ๐ฐ๐ฎ๐น ๐ฅ๐ฒ๐ฝ๐ผ๐ฟ๐ ๐ถ๐ ๐ข๐จ๐ง โ the most detailed disclosure yet of how to build a frontier Native Multimodal Model.
Inside:
โจ Near-lossless visual interface (no VEs, no VAEs)
โจ Native Multimodal Unified Modeling
โจ Joint AR + pixel-space flow matching training
โจ Native Mixture-of-Transformers backbone
โจ 6-stage training recipe + RL post-training + distillation
If you work on NMM, this is the playbook.
๐ค One more thing: ๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ-๐จ๐ญ-๐๐ฏ๐-๐ ๐ผ๐ง (๐ฏ๐ด๐-๐๐ฏ๐ ๐ ๐ผ๐) ๐๐ฒ๐ถ๐ด๐ต๐๐ ๐ฎ๐ฟ๐ฒ ๐ป๐ผ๐ ๐ผ๐ฝ๐ฒ๐ป-๐๐ผ๐๐ฟ๐ฐ๐ฒ๐ฑ โ a RARE native unified model on an MoE backbone (Only 3B active! Lightning Fastโก)
๐ Tech Report: arxiv.org/abs/2605.12500
๐ค Daily Papers (Vote & Discuss): huggingface.co/papers/2605.1โฆ
๐ค Models: huggingface.co/collections/sโฆ
๐ป Code: github.com/OpenSenseNova/Senโฆ
๐ฎ Demo: unify.light-ai.top
๐พ Discord: discord.com/invite/BuTXPHmQuโฆ
SenseTime retweeted
Most multimodal models still treat text and images like separate problems.
One part reads. Another part reasons. Another part generates.
SenseNova U1 is trying something different: one unified model where language and vision work inside the same architecture.
SenseTime retweeted
open source nano-banana now on @huggingface?
SenseNova U1 is an open source unified image-text understanding & generation model (like Nano Banana/GPT-Image) by @SenseTime_AI
8B parameter unified generation model, of course, I built a Spaces demo for it ๐
SenseTime retweeted
SenseNova U1็จComfyUIใใผใๅ
ฌ้ใ
8ในใใใLoRAๅฏพๅฟใงๆจ่ซๆฐๅๆธใ็ๆๅน็ใฎๅไธใ
ใทในใใ RAM๏ผๆจๅฅจ36GBไปฅไธ๏ผใๅคงใใๆถ่ฒปใใ่จญ่จใ
VRAM16GB่ถ
ใฏprefetch_count=0ใงๅฎๅฎๅใ
#SenseNova #ComfyUI
URLใฏใชใโฌ๏ธ
SenseTime retweeted
Unified multimodal pipelines like this bring content creation closer to real workflows. Take a look and star github.com/OpenSenseNova/Senโฆ
๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐๐ถ๐๐ฒ ๐ฆ๐ฒ๐ฟ๐ถ๐ฒ๐ ๐ถ๐ ๐ป๐ผ๐ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ๐๐ฟ๐ฐ๐ฒ!
Built on the ๐ก๐๐ข-๐๐ป๐ถ๐ณ๐ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ, it natively unifies multimodal understanding and generation, delivering:
โข๐ฆ๐ข๐ง๐ ๐๐ณ๐ณ๐ถ๐ฐ๐ถ๐ฒ๐ป๐ฐ๐ ๐๐บ๐ผ๐ป๐ด ๐ข๐ฝ๐ฒ๐ป-๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐ ๐ผ๐ฑ๐ฒ๐น๐: Compact models (8B & A3B) delivering commercial-grade performance and exceptional cost efficiency. Leading performance among open-source models across a wide range of understanding, reasoning, and generation benchmarks.
โข๐ก๐ฎ๐๐ถ๐๐ฒ ๐๐บ๐ฎ๐ด๐ฒโ๐ง๐ฒ๐
๐ ๐๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป: Generate coherent interleaved text and images in a single flow using one model; ideal for practical applications like guides, where visuals turn complex information into intuitive insights.
โข๐๐ถ๐ด๐ต-๐๐ฒ๐ป๐๐ถ๐๐ ๐๐ป๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ถ๐ผ๐ป ๐ฅ๐ฒ๐ป๐ฑ๐ฒ๐ฟ๐ถ๐ป๐ด: Strong capabilities in dense visual communication, generating richly structured layouts for knowledge illustrations, posters, PPTs, comics and other information-rich formats.
๐๐๐ด๐ด๐ถ๐ป๐ด ๐๐ฎ๐ฐ๐ฒ: huggingface.co/collections/sโฆ
๐๐ถ๐๐๐๐ฏ: github.com/OpenSenseNova/Senโฆ
๐๐ถ๐๐ฐ๐ผ๐ฟ๐ฑ: discord.gg/cxkwXWjpย
@huggingface @github
SenseTime retweeted
This honestly feels like an open source GPT-Image 2 alternative
Except it actually handles structured outputs like posters and slides better
Explore it and give it a star github.com/OpenSenseNova/Senโฆ
๐ฆ๐ฒ๐ป๐๐ฒ๐ก๐ผ๐๐ฎ ๐จ1 ๐๐ถ๐๐ฒ ๐ฆ๐ฒ๐ฟ๐ถ๐ฒ๐ ๐ถ๐ ๐ป๐ผ๐ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ๐๐ฟ๐ฐ๐ฒ!
Built on the ๐ก๐๐ข-๐๐ป๐ถ๐ณ๐ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ, it natively unifies multimodal understanding and generation, delivering:
โข๐ฆ๐ข๐ง๐ ๐๐ณ๐ณ๐ถ๐ฐ๐ถ๐ฒ๐ป๐ฐ๐ ๐๐บ๐ผ๐ป๐ด ๐ข๐ฝ๐ฒ๐ป-๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐ ๐ผ๐ฑ๐ฒ๐น๐: Compact models (8B & A3B) delivering commercial-grade performance and exceptional cost efficiency. Leading performance among open-source models across a wide range of understanding, reasoning, and generation benchmarks.
โข๐ก๐ฎ๐๐ถ๐๐ฒ ๐๐บ๐ฎ๐ด๐ฒโ๐ง๐ฒ๐
๐ ๐๐ป๐๐ฒ๐ฟ๐น๐ฒ๐ฎ๐๐ฒ๐ฑ ๐๐ฒ๐ป๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป: Generate coherent interleaved text and images in a single flow using one model; ideal for practical applications like guides, where visuals turn complex information into intuitive insights.
โข๐๐ถ๐ด๐ต-๐๐ฒ๐ป๐๐ถ๐๐ ๐๐ป๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ถ๐ผ๐ป ๐ฅ๐ฒ๐ป๐ฑ๐ฒ๐ฟ๐ถ๐ป๐ด: Strong capabilities in dense visual communication, generating richly structured layouts for knowledge illustrations, posters, PPTs, comics and other information-rich formats.
๐๐๐ด๐ด๐ถ๐ป๐ด ๐๐ฎ๐ฐ๐ฒ: huggingface.co/collections/sโฆ
๐๐ถ๐๐๐๐ฏ: github.com/OpenSenseNova/Senโฆ
๐๐ถ๐๐ฐ๐ผ๐ฟ๐ฑ: discord.gg/cxkwXWjpย
@huggingface @github