{"paper":{"title":"Phenaki: Variable Length Video Generation From Open Domain Textual Description","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Phenaki generates arbitrarily long videos from sequences of text prompts describing evolving scenes.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Dumitru Erhan, Han Zhang, Hernan Moraldo, Julius Kunze, Mohammad Babaeizadeh, Mohammad Taghi Saffar, Pieter-Jan Kindermans, Ruben Villegas, Santiago Castro","submitted_at":"2022-10-05T17:18:28Z","abstract_excerpt":"We present Phenaki, a model capable of realistic video synthesis, given a sequence of textual prompts. Generating videos from text is particularly challenging due to the computational cost, limited quantities of high quality text-video data and variable length of videos. To address these issues, we introduce a new model for learning video representation which compresses the video to a small representation of discrete tokens. This tokenizer uses causal attention in time, which allows it to work with variable-length videos. To generate video tokens from text we are using a bidirectional masked t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Phenaki can generate arbitrary long videos conditioned on a sequence of prompts (i.e. time variable text or a story) in open domain. To the best of our knowledge, this is the first time a paper studies generating videos from time variable prompts.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That joint training on a large corpus of image-text pairs as well as a smaller number of video-text examples can result in generalization beyond what is available in the video datasets.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Phenaki generates arbitrary-length videos from sequences of text prompts by tokenizing videos with causal temporal attention and generating tokens with a text-conditioned masked transformer, trained jointly on images and videos.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Phenaki generates arbitrarily long videos from sequences of text prompts describing evolving scenes.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f745509c0bad2ab37e58c3943ae7856003a85695ccdb6ca9dd4f69f0e7c644cd"},"source":{"id":"2210.02399","kind":"arxiv","version":1},"verdict":{"id":"0ff5b6dc-2ee4-45f2-8b5a-f6cac5d36245","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T01:37:51.649559Z","strongest_claim":"Phenaki can generate arbitrary long videos conditioned on a sequence of prompts (i.e. time variable text or a story) in open domain. To the best of our knowledge, this is the first time a paper studies generating videos from time variable prompts.","one_line_summary":"Phenaki generates arbitrary-length videos from sequences of text prompts by tokenizing videos with causal temporal attention and generating tokens with a text-conditioned masked transformer, trained jointly on images and videos.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That joint training on a large corpus of image-text pairs as well as a smaller number of video-text examples can result in generalization beyond what is available in the video datasets.","pith_extraction_headline":"Phenaki generates arbitrarily long videos from sequences of text prompts describing evolving scenes."},"references":{"count":61,"sample":[{"doi":"","year":2021,"title":"Vivit: A video vision transformer","work_id":"8c555c6f-f480-49d7-89f7-435aedd756bd","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Stochastic variational video prediction","work_id":"7ebdee21-b63c-4ceb-a897-a0e2326c0f2a","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Fitvid: Overfitting in pixel-level video prediction","work_id":"98b75ffa-1d61-4641-a59f-5967267b7d2c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Frozen in time: A joint video and image encoder for end-to-end retrieval","work_id":"1ae0008f-2042-4c35-a953-fb26d78a25c2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Con- ditional gan with discriminative filter generation for text-to-video synthesis","work_id":"14854921-8b97-4939-9d4b-2c8723e1596c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":61,"snapshot_sha256":"f24d49bcd3d44a312b074908ad4dc8bbd828765188fcb3fc1759a0257e414eab","internal_anchors":12},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c42790fbbfe7d935441820feabe9affea9f1cdb898f97273afde1a8c768d8968"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}