Viewing enriched Twitter post
Excited to share Halcyon, our latest research on multimodal pretraining! Building strong text-to-image generative models is simpler than we think! Blog: https://t.co/HOrwIJIHTf Models: https://t.co/xBPJkEAOPU Paper: https://t.co/8kROIz1c7L
{
"media": [
{
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2110775401756958947/media_0.jpg",
"media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2110775401756958947/media_0.jpg",
"type": "photo",
"filename": "media_0.jpg"
},
{
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2110775401756958947/media_1.jpg",
"media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2110775401756958947/media_1.jpg",
"type": "photo",
"filename": "media_1.jpg"
}
],
"processed_at": "2026-10-22T07:07:11.786720",
"pipeline_version": "2.0"
}{
"type": "tweet",
"id": "2110775401756958947",
"url": "https://x.com/keshigeyan/status/2110775401756958947",
"twitterUrl": "https://twitter.com/keshigeyan/status/2110775401756958947",
"text": "Excited to share Halcyon, our latest research on multimodal pretraining!\n\nBuilding strong text-to-image generative models is simpler than we think!\n\nBlog: https://t.co/HOrwIJIHTf\nModels: https://t.co/xBPJkEAOPU\nPaper: https://t.co/8kROIz1c7L",
"source": "",
"retweetCount": 28,
"replyCount": 4,
"likeCount": 74,
"quoteCount": 2,
"viewCount": 11871,
"createdAt": "Thu Oct 15 16:50:46 +0000 2026",
"lang": "en",
"bookmarkCount": 24,
"isReply": false,
"inReplyToId": null,
"conversationId": "2110775401756958947",
"displayTextRange": [
0,
241
],
"inReplyToUserId": null,
"inReplyToUsername": null,
"author": {
"type": "user",
"userName": "keshigeyan",
"url": "https://x.com/keshigeyan",
"twitterUrl": "https://twitter.com/keshigeyan",
"id": "1095671366949335040",
"name": "Keshigeyan Chandrasegaran",
"isVerified": false,
"isBlueVerified": true,
"verifiedType": null,
"profilePicture": "https://pbs.twimg.com/profile_images/1857746255625256960/b51xdElE_normal.jpg",
"coverPicture": "https://pbs.twimg.com/profile_banners/1095671366949335040/1731200671",
"description": "",
"location": "Stanford, CA",
"followers": 1131,
"following": 388,
"status": "",
"canDm": false,
"canMediaTag": true,
"createdAt": "Wed Feb 13 13:09:45 +0000 2019",
"entities": {
"description": {
"urls": []
},
"url": {}
},
"fastFollowersCount": 0,
"favouritesCount": 1383,
"hasCustomTimelines": true,
"isTranslator": false,
"mediaCount": 26,
"statusesCount": 403,
"withheldInCountries": [],
"affiliatesHighlightedLabel": {},
"possiblySensitive": false,
"pinnedTweetIds": [
"2060398262591668315"
],
"profile_bio": {
"description": "CS PhD student @Stanford. Research @StanfordAILab, @StanfordSVL & @RadicalNumerics. Prev: research @LiquidAI_, @sutdsg (Temasek Labs), undergrad @sutdsg.",
"entities": {
"description": {
"user_mentions": [
{
"id_str": "",
"indices": [
15,
24
],
"name": "",
"screen_name": "Stanford"
},
{
"id_str": "",
"indices": [
35,
49
],
"name": "",
"screen_name": "StanfordAILab"
},
{
"id_str": "",
"indices": [
51,
63
],
"name": "",
"screen_name": "StanfordSVL"
},
{
"id_str": "",
"indices": [
66,
82
],
"name": "",
"screen_name": "RadicalNumerics"
},
{
"id_str": "",
"indices": [
99,
109
],
"name": "",
"screen_name": "LiquidAI_"
},
{
"id_str": "",
"indices": [
111,
118
],
"name": "",
"screen_name": "sutdsg"
},
{
"id_str": "",
"indices": [
145,
152
],
"name": "",
"screen_name": "sutdsg"
}
]
},
"url": {
"urls": [
{
"display_url": "cs.stanford.edu/~keshik/",
"expanded_url": "http://cs.stanford.edu/~keshik/",
"indices": [
0,
23
],
"url": "https://t.co/f6ZaZKjBM0"
}
]
}
}
},
"isAutomated": false,
"automatedBy": null
},
"extendedEntities": {},
"card": {
"binding_values": [
{
"key": "photo_image_full_size_large",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 419,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=800x419",
"width": 800
}
}
},
{
"key": "thumbnail_image",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 150,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=280x150",
"width": 267
}
}
},
{
"key": "description",
"value": {
"string_value": "Halcyon is a family of early-fusion multimodal autoregressive models pretrained from scratch on image-text pairs, with no internet-scale language pretraining."
}
},
{
"key": "domain",
"value": {
"string_value": "www.radicalnumerics.ai"
}
},
{
"key": "thumbnail_image_large",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 320,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=800x320_1",
"width": 569
}
}
},
{
"key": "summary_photo_image_small",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 202,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=386x202",
"width": 386
}
}
},
{
"key": "thumbnail_image_original",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 941,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=orig",
"width": 1672
}
}
},
{
"key": "photo_image_full_size_small",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 202,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=386x202",
"width": 386
}
}
},
{
"key": "summary_photo_image_large",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 419,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=800x419",
"width": 800
}
}
},
{
"key": "thumbnail_image_small",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 81,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=144x144",
"width": 144
}
}
},
{
"key": "thumbnail_image_x_large",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 941,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=png&name=2048x2048_2_exp",
"width": 1672
}
}
},
{
"key": "photo_image_full_size_original",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 941,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=orig",
"width": 1672
}
}
},
{
"key": "photo_image_full_size_alt_text",
"value": {
"string_value": "Text-to-image generative modeling without language pretraining — Radical Numerics"
}
},
{
"key": "vanity_url",
"value": {
"scribe_key": "vanity_url",
"string_value": "radicalnumerics.ai"
}
},
{
"key": "photo_image_full_size",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 314,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=600x314",
"width": 600
}
}
},
{
"key": "summary_photo_image_alt_text",
"value": {
"string_value": "Text-to-image generative modeling without language pretraining — Radical Numerics"
}
},
{
"key": "thumbnail_image_color",
"value": {
"image_color_value": {
"palette": [
{
"percentage": 92.4,
"rgb": {
"blue": 193,
"green": 215,
"red": 237
}
},
{
"percentage": 2.78,
"rgb": {
"blue": 134,
"green": 118,
"red": 117
}
},
{
"percentage": 2.3,
"rgb": {
"blue": 97,
"green": 113,
"red": 147
}
},
{
"percentage": 1.88,
"rgb": {
"blue": 118,
"green": 159,
"red": 213
}
},
{
"percentage": 0.22,
"rgb": {
"blue": 179,
"green": 164,
"red": 192
}
}
]
}
}
},
{
"key": "title",
"value": {
"string_value": "Text-to-image generative modeling without language pretraining — Radical Numerics"
}
},
{
"key": "summary_photo_image_color",
"value": {
"image_color_value": {
"palette": [
{
"percentage": 92.4,
"rgb": {
"blue": 193,
"green": 215,
"red": 237
}
},
{
"percentage": 2.78,
"rgb": {
"blue": 134,
"green": 118,
"red": 117
}
},
{
"percentage": 2.3,
"rgb": {
"blue": 97,
"green": 113,
"red": 147
}
},
{
"percentage": 1.88,
"rgb": {
"blue": 118,
"green": 159,
"red": 213
}
},
{
"percentage": 0.22,
"rgb": {
"blue": 179,
"green": 164,
"red": 192
}
}
]
}
}
},
{
"key": "summary_photo_image_x_large",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 941,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=png&name=2048x2048_2_exp",
"width": 1672
}
}
},
{
"key": "summary_photo_image",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 314,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=600x314",
"width": 600
}
}
},
{
"key": "photo_image_full_size_color",
"value": {
"image_color_value": {
"palette": [
{
"percentage": 92.4,
"rgb": {
"blue": 193,
"green": 215,
"red": 237
}
},
{
"percentage": 2.78,
"rgb": {
"blue": 134,
"green": 118,
"red": 117
}
},
{
"percentage": 2.3,
"rgb": {
"blue": 97,
"green": 113,
"red": 147
}
},
{
"percentage": 1.88,
"rgb": {
"blue": 118,
"green": 159,
"red": 213
}
},
{
"percentage": 0.22,
"rgb": {
"blue": 179,
"green": 164,
"red": 192
}
}
]
}
}
},
{
"key": "photo_image_full_size_x_large",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 941,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=png&name=2048x2048_2_exp",
"width": 1672
}
}
},
{
"key": "card_url",
"value": {
"scribe_key": "card_url",
"string_value": "https://t.co/HOrwIJIHTf"
}
},
{
"key": "summary_photo_image_original",
"value": {
"image_value": {
"alt": "Text-to-image generative modeling without language pretraining — Radical Numerics",
"height": 941,
"url": "https://pbs.twimg.com/card_img/2110767509809057799/Low72iKT?format=jpg&name=orig",
"width": 1672
}
}
}
],
"card_platform": {
"platform": {
"audience": {
"name": "production"
},
"device": {
"name": "iPhone",
"version": "13"
}
}
},
"name": "summary_large_image",
"url": "https://t.co/HOrwIJIHTf",
"user_refs_results": []
},
"place": {},
"entities": {
"urls": [
{
"display_url": "radicalnumerics.ai/blog/halcyon",
"expanded_url": "https://www.radicalnumerics.ai/blog/halcyon",
"indices": [
155,
178
],
"url": "https://t.co/HOrwIJIHTf"
},
{
"display_url": "huggingface.co/collections/ra…",
"expanded_url": "https://huggingface.co/collections/radicalnumerics/halcyon",
"indices": [
187,
210
],
"url": "https://t.co/xBPJkEAOPU"
},
{
"display_url": "radicalnumerics.ai/assets/halcyon…",
"expanded_url": "https://www.radicalnumerics.ai/assets/halcyon_paper.pdf",
"indices": [
218,
241
],
"url": "https://t.co/8kROIz1c7L"
}
]
},
"quoted_tweet": {
"type": "tweet",
"id": "2110766682977521742",
"url": "https://x.com/RadicalNumerics/status/2110766682977521742",
"twitterUrl": "https://twitter.com/RadicalNumerics/status/2110766682977521742",
"text": "1/8 Does text-to-image generation need language pretraining?\n\nToday, we share Halcyon, a family of early-fusion, base models trained from scratch. Halcyon outperforms strong late-fusion baselines with up to 275× less pretraining compute. Models and recipes are open.\n\nHalcyon is part of a research program at Radical Numerics on the science of multimodal pretraining and scaling.",
"source": "",
"retweetCount": 30,
"replyCount": 4,
"likeCount": 175,
"quoteCount": 3,
"viewCount": 68469,
"createdAt": "Thu Oct 15 16:16:08 +0000 2026",
"lang": "en",
"bookmarkCount": 120,
"isReply": false,
"inReplyToId": null,
"conversationId": "2110766682977521742",
"displayTextRange": [
0,
278
],
"inReplyToUserId": null,
"inReplyToUsername": null,
"author": {
"type": "user",
"userName": "RadicalNumerics",
"url": "https://x.com/RadicalNumerics",
"twitterUrl": "https://twitter.com/RadicalNumerics",
"id": "1922396702675501057",
"name": "Radical Numerics",
"isVerified": false,
"isBlueVerified": true,
"verifiedType": "Business",
"profilePicture": "https://pbs.twimg.com/profile_images/1959989699117977601/AOAc1z9b_normal.jpg",
"coverPicture": "",
"description": "",
"location": "San Francisco & Tokyo",
"followers": 4266,
"following": 4,
"status": "",
"canDm": false,
"canMediaTag": true,
"createdAt": "Tue May 13 21:01:26 +0000 2025",
"entities": {
"description": {
"urls": []
},
"url": {}
},
"fastFollowersCount": 0,
"favouritesCount": 48,
"hasCustomTimelines": true,
"isTranslator": false,
"mediaCount": 34,
"statusesCount": 51,
"withheldInCountries": [],
"affiliatesHighlightedLabel": {},
"possiblySensitive": false,
"pinnedTweetIds": [],
"profile_bio": {
"description": "Systems, scaling and architecture for general biological intelligence",
"entities": {
"description": {},
"url": {
"urls": [
{
"display_url": "radicalnumerics.ai",
"expanded_url": "https://www.radicalnumerics.ai/",
"indices": [
0,
23
],
"url": "https://t.co/m2GhCEu2pv"
}
]
}
}
},
"isAutomated": false,
"automatedBy": null
},
"extendedEntities": {
"media": [
{
"additional_media_info": {
"monetizable": false
},
"allow_download_status": {
"allow_download": true
},
"display_url": "pic.x.com/8pnQTUqnzu",
"expanded_url": "https://x.com/RadicalNumerics/status/2110766682977521742/video/1",
"ext_master_playlist_only": [],
"ext_media_availability": {
"status": "Available"
},
"ext_playlists": [],
"id_str": "2110765504961146880",
"indices": [
279,
302
],
"media_key": "13_2110765504961146880",
"media_results": {
"id": "QXBpTWVkaWFSZXN1bHRzOgwABAoAAR1K8gsfW3AAAAA=",
"result": {
"__typename": "ApiMedia",
"id": "QXBpTWVkaWE6DAAECgABHUryCx9bcAAAAA==",
"media_key": "13_2110765504961146880"
}
},
"media_url_https": "https://pbs.twimg.com/amplify_video_thumb/2110765504961146880/img/hmLfs3Bj2_AfswTB.jpg",
"original_info": {
"focus_rects": [],
"height": 1080,
"width": 1920
},
"sizes": {
"large": {
"h": 1080,
"w": 1920
}
},
"type": "video",
"url": "https://t.co/8pnQTUqnzu",
"video_info": {
"aspect_ratio": [
16,
9
],
"duration_millis": 10500,
"variants": [
{
"content_type": "application/x-mpegURL",
"url": "https://video.twimg.com/amplify_video/2110765504961146880/pl/H130ksXKOl8h1d-I.m3u8?tag=29"
},
{
"bitrate": 256000,
"content_type": "video/mp4",
"url": "https://video.twimg.com/amplify_video/2110765504961146880/vid/avc1/480x270/ADoZz6zcTL3Xgk-Z.mp4?tag=29"
},
{
"bitrate": 832000,
"content_type": "video/mp4",
"url": "https://video.twimg.com/amplify_video/2110765504961146880/vid/avc1/640x360/7xFiy-K3Qj8CUbm0.mp4?tag=29"
},
{
"bitrate": 2176000,
"content_type": "video/mp4",
"url": "https://video.twimg.com/amplify_video/2110765504961146880/vid/avc1/1280x720/6ytD0tfQKxktcvtx.mp4?tag=29"
},
{
"bitrate": 10368000,
"content_type": "video/mp4",
"url": "https://video.twimg.com/amplify_video/2110765504961146880/vid/avc1/1920x1080/o2zYibODgKxbRAF7.mp4?tag=29"
}
]
}
}
]
},
"card": null,
"place": {},
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"quoted_tweet": null,
"retweeted_tweet": null,
"isLimitedReply": false,
"communityInfo": null,
"article": null
},
"retweeted_tweet": null,
"isLimitedReply": false,
"communityInfo": null,
"article": null
}