Full papers
Submissions must be in PDF format, anonymized, and limited to four content pages (including figures and tables), with unlimited references and supplementary materials. Camera-ready versions may extend to five content pages.
Computer Vision/CORE Unranked
F2S@ICML 2026
ICML 2026 Workshop - From Frames to Stories (F2S): Toward Reliable, Controllable and Trustworthy Long-Horizon Video Generation
The F2S@ICML 2026 workshop focuses on advancing reliable, controllable, and trustworthy long-horizon video generation, addressing challenges such as identity drift, narrative consistency, and user control over minutes-long video sequences. It brings together researchers to explore model design, memory mechanisms, interactive systems, and evaluation benchmarks for multi-shot video generation.
Key deadlines
Verified September 9, 2026Full paper
May 11, 2026
AoE
Workshop timeline
Submission and decisions
Full paperKey deadline
May 11, 2026 · AoE
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Research areas in scope
01
Models, Memory & Long-Context Generation
Long-horizon and multi-shot video generation modelsLong-context architectures, memory mechanisms, retrieval, and state trackingConsistency maintenance across identities, scenes, events, and narrative structureHierarchical generation, planning, and inference-time refinement for long videos
02
Control, Editing & Interactive Systems
Controllable video generation with text, keyframes, layouts, trajectories, sketches, audio, or other multimodal signalsStoryboarding, shot planning, structured prompting, and script-guided generationVideo editing, localized revision, iterative refinement, and human-in-the-loop creation toolsInteractive, agentic, or creator-facing systems for long-form video generation
03
Benchmarks, Data & Trustworthy Evaluation
Datasets and annotations for long-form, story-level, or multi-shot video generationBenchmarks and protocols for coherence, consistency, controllability, and narrative fidelityAutomatic metrics, VLM-based evaluation, and scalable human evaluation for long-horizon videoReliability, robustness, safety, provenance, and reproducibility in generative video systems
Policies worth checking twice
- All submissions must be anonymized.
- Submissions are limited to four content pages; camera-ready versions may be up to five pages.
- References and supplementary materials must be included in the same PDF file.
- Maximum file size for submissions is 50MB.
- Submissions violating ICML style (e.g., margins, font sizes) or page limits may be rejected without review.
- Dual submissions are allowed: ongoing, under-review, or recently accepted (without proceedings) papers are welcome.
- The workshop is non-archival and does not have official proceedings; submissions can be concurrently or subsequently submitted elsewhere.
- Submissions and reviews are not public; only accepted papers will be made public.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026