Runway¶

Development line: organization:runway · thread runway-video-ai-development
Last event: 2024-09-19 · 3 dated since 2023-03-20 · Researched: 2026-09-04 · confidence: medium

What it is¶

Runway is a cloud platform for filmmakers, creative teams and developers, competing with Adobe Firefly/Premiere on the creative side and Replicate on the API side.

The platform provides three surfaces: - Creative: generates and edits video, images and audio. - Dev: provides APIs, MCP and multi-model routing. - Robotics: provides simulation tooling.

Current model limits and measures: - Gen-4.5: current Runway video model; web generations are 2–10 seconds, 720p, 24/25 fps and 12 credits per second. - Aleph 2.0: edits 2–30-second clips. - Act-Two: transfers a recorded performance to a character for up to 30 seconds.

Use Runway as an integrated shot-production system, not as a one-prompt finished-film generator.

Development line¶

• 2023-03-20 — Runway documented Gen-2 on its research site. It generated clips from text, text plus image, or an image, while retaining video-to-video and stylization workflows.
• 2023-08-11 — A longer video-duration capability was reported for Runway. On 2023-08-11, a community link reported a Runway capability for videos lasting 18 seconds. This represents a material duration-related product step if confirmed, but no first-party announcement establishes the rollout scope.
• 2024-09-19 — Runway published news of a partnership with Lionsgate. On 2024-09-19, Runway's news URL identified a partnership with Lionsgate. This is a material organizational-development milestone, but absent article contents leave the partnership terms unverified.

What changed¶

• 2022-08-11 — An early text-guided video-editing preview pointed toward Gen-1: retain a source clip’s structure while changing content or style with text or an image. The exact social post is no longer retrievable, so this establishes lineage rather than a verified release.
• 2022-10-18 — Runway was becoming a suite rather than one generator. [Found 2026-09-04] Its first-party archive dates AI Magic Tools to 2022-10-12 and Erase and Replace to 2022-10-17; the generic landing page does not identify which change prompted this date.
• 2022-11-03 — The exact change cannot be verified because no source URL survives. [Found 2026-09-04] Runway’s archive dates Frame Interpolation, which generated video from still images, to 2022-11-02; temporal proximity does not prove it was the same item.
• 2023-01-11 — AI Training introduced custom image-model training. [Found 2026-09-04] Runway’s archive dates the feature to 2022-12-05, so 2023-01-11 appears to document adoption rather than a new version.
• 2023-03-20 — Gen-2 removed the need for an existing source video in key modes: it generated clips from text, text plus image, or an image, while retaining video-to-video and stylization workflows.
• 2023-08-11 — A contemporary user report documented Gen-2 extensions up to 18 seconds with better continuity than repeatedly animating the previous last frame. No matching first-party release note was recovered.
• 2023-08-31 — Motion Slider added direct control over how much movement Gen-2 introduced. Runway’s official tutorial followed on 2023-09-12.
• 2023-09-05 — No first-party product named “Runway 3D Video” can be verified from this date. The surviving landing page supports treating it as a 2.5D/parallax-style workflow or demonstration, not a separate model release.
• 2023-09-09 — Director Mode added camera direction and intensity controls to Gen-2 in the browser, with iOS announced to follow.
• 2023-11-21 — Motion Brush, Gen-2 style presets, fractional Director Mode controls and an improved image model reached users. Runway’s consolidated first-party announcement is dated 2023-11-23.
• 2024-06-17/2024-07-01 — [Found 2026-09-04] Gen-3 Alpha replaced Gen-2 as Runway’s next model generation, improving fidelity, consistency, motion and temporal control.
• 2024-09-18/2024-09-19 — Runway and Lionsgate announced a custom model trained on Lionsgate’s proprietary catalog for pre-production and post-production experimentation; the first-party release is dated one day earlier than the recorded event.
• 2024-10-25 — [Found 2026-09-04] Act-One introduced performance-driven character animation from one driving video and one character image.
• 2025-03-31/2025-04-01 — [Found 2026-09-04] Gen-4 introduced cross-shot consistency for characters, locations and objects; Gen-4 Turbo followed as the faster, cheaper iteration model.
• 2025-04-30 — [Found 2026-09-04] Gen-4 References made reusable character, object and location consistency possible from one or several images without a custom-training cycle.
• 2025-07-17/2025-07-18 — Act-Two replaced Act-One’s performance-capture path with better head, face, body and hand tracking. The official changelog is dated one day earlier than the recorded event.
• 2025-07-25/2025-07-31 — [Found 2026-09-04] Aleph added prompt-driven changes to existing footage: add, remove or transform objects, alter lighting and style, or generate another view.
• 2025-10-24 — [Found 2026-09-04] Workflows added reusable node graphs that chain models, media types and intermediate processing.
• 2025-12-01/2025-12-11 — [Found 2026-09-04] Gen-4.5 became Runway’s current video-generation model; image-to-video followed on 2026-01-21.
• 2026-02-20 — [Found 2026-09-04] Runway became a multi-vendor model platform, adding third-party image and video models alongside its own.
• 2026-03-09 — [Found 2026-09-04] Runway Characters added real-time conversational avatars through the web demo and API.
• 2026-05-13/2026-05-21 — [Found 2026-09-04] Runway Agent added conversational multi-shot production, while Aleph 2.0 and Edit Studio expanded prompt-based editing to footage up to 30 seconds.
• 2026-07-21 — [Found 2026-09-04] Model Router let Dev users select models automatically by cost, latency or quality constraints.
• 2026-08-10/2026-08-24 — [Found 2026-09-04] Gen-4.5 and Aleph 2.0 gained ProRes and PNG-sequence delivery on eligible plans; Wan 3.0 added 2–30-second video with audio and multiple media references.
• 2026-09-03 — [Found 2026-09-04] GWM Worlds 2 appeared as a research preview for indefinitely running, interactive 720p video at 24 fps with 48 kHz audio. It remains research, with fidelity, fast-camera and long-term-memory limits.

How to use this¶

As of 2025-07-18, do not change production workflows from this timeline alone. Use it to locate the Gen-2, longer-duration video, and Lionsgate-partnership milestones, then verify current first-party Runway documentation before relying on any capability, availability, or partnership implication.

1. Open the Runway web app and create or enter a workspace. — https://app.runwayml.com/
2. Choose the surface: Tools for explicit model settings, Apps for task-specific workflows, Agent for conversational multi-shot work, or Workflows for reusable node pipelines. Each creates a Session that keeps outputs together. — https://help.runwayml.com/hc/en-us/articles/37425232841875-Getting-Started-with-Generative-Video
3. For a new shot, choose Gen-4.5 Text to Video for an unanchored idea or Image to Video when composition and identity must start from a specific frame. Set aspect ratio, 2–10-second duration and 24 or 25 fps, then generate and iterate. — https://help.runwayml.com/hc/en-us/articles/46974685288467-Creating-with-Gen-4-5
4. Build reusable characters, products or locations with Gen-4 Image References; name references, invoke them with @ and use up to three in one generation. — https://help.runwayml.com/hc/en-us/articles/40042718905875-Creating-with-Gen-4-Image-References
5. For character acting, open Apps > Performance Capture with Act-Two, upload a driving-performance video plus a character image or video, choose gesture handling and generate. — https://help.runwayml.com/hc/en-us/articles/42311337895827-Performance-Capture-with-Act-Two
6. For existing footage, open Edit Studio, upload the clip, edit a representative keyframe, preview the change and let Aleph 2.0 propagate it through the selected shot or range. — https://help.runwayml.com/hc/en-us/articles/51683104370451-Creating-with-Edit-Studio
7. Assemble and trim accepted shots in Agent or Studio. For editorial or VFX delivery, choose MP4, ProRes or PNG sequence before generation; ProRes and PNG sequence require an eligible plan and cannot be retrofitted to an existing generation. — https://help.runwayml.com/hc/en-us/articles/54396547993491-Exporting-Videos-in-ProRes-and-PNG-Sequence-Formats
8. For automation, create a Runway Dev account and API key, use the SDK or REST endpoint, submit the generation as an asynchronous task and wait for its output or explicit failure. — https://docs.dev.runwayml.com/guides/using-the-api/

Best practices¶

• Choose the control mode from the material you already have: text-to-video for exploration, image-to-video for a locked first frame, and video-to-video/Edit Studio when original motion and framing matter most. — https://runway.com/resources/how-ai-video-generation-works
• Start with the smallest clear prompt, change one element per iteration and describe what should happen in positive, literal language. JSON formatting itself adds no accuracy. — https://help.runwayml.com/hc/en-us/articles/47313698911891-Introduction-to-Prompting
• For image-to-video, let the image supply subject, composition, lighting and style; use text mainly for subject, environment and camera motion. Remove blur, malformed anatomy and conflicting implied-motion cues before animating. — https://help.runwayml.com/hc/en-us/articles/48324313115155-Image-to-Video-Prompting-Guide
• Describe camera work with established shot, angle, direction and speed terms. Give each clip one directed beat and control cuts in the timeline instead of asking one generation for an entire sequence. — https://runway.com/resources/ai-camera-prompts
• For consistent characters, start with an evenly lit, neutral reference; save and name it, then add only the references required for the shot. — https://help.runwayml.com/hc/en-us/articles/40042718905875-Creating-with-Gen-4-Image-References
• For Act-Two, keep the performer visible at least from the waist up, preserve clear eyes and mouth, avoid cuts and match the character-video duration to the performance. Use a character image when gesture transfer matters; use character video when existing camera and environmental motion must remain. — https://help.runwayml.com/hc/en-us/articles/42311337895827-Performance-Capture-with-Act-Two
• For Aleph 2.0, state one action verb and the requested transformation, preview the keyframe before propagation and limit the edited range when only part of the clip needs work. — https://help.runwayml.com/hc/en-us/articles/52150503729171-Aleph-2-0-Prompting-Guide
• Keep original exports and their C2PA metadata, and disclose synthetic footage where provenance matters. Human visual inspection is not a dependable authenticity check. — https://runway.com/research/theturingreel

Superseded by this¶

• 2025-05-11 — Gen-2 was fully deprecated. Its Motion Brush, Motion Slider, Director Mode and 18-second-extension instructions are no longer usable; Runway directs users to Gen-4.5 and prompt-based motion control.
• 2025-04-30 — Gen-4 References superseded most day-to-day 2022 AI Training guidance for recurring characters, locations and objects; references work without running a custom-training job.
• 2025-07-17 — Act-Two superseded Act-One as Runway’s supported performance-capture workflow.
• 2025-12-11 — Gen-4.5 superseded Gen-4 as the quality-first Runway video model. Gen-4 and Gen-4 Turbo remain older, cheaper or faster alternatives where available.
• 2026-05-21 — Aleph 2.0 in Edit Studio superseded the original 2025 Aleph workflow and increased the supported edit window to 30 seconds.
• 2026-02-20 — Guidance that treats Runway as access to only Runway-trained models became incomplete; Creative and Dev now expose models from multiple vendors as well as Runway’s own.

Still unknown¶

• The 2022-08-11 source at https://twitter.com/i/status/1557517982095626241 could not be retrieved; the later Gen-1 research page verifies the technical direction but not that exact demonstration.
• The exact 2022-10-18, 2022-11-03 and 2023-01-11 triggers cannot be bound conclusively: the surviving links are a generic landing page or absent, while the first-party archive only establishes nearby releases.
• The 2023-08-11 18-second limit is supported by a contemporary community report, not a recovered first-party release note.
• The 2023-09-05 “3D video” label does not map to a verifiable first-party product. It may describe a parallax-style result made with image-to-video controls rather than a separate Runway capability.
• The official dates differ slightly from three recorded dates: Motion Brush’s consolidated announcement is 2023-11-23, Lionsgate’s release is 2024-09-18 and Act-Two’s changelog entry is 2025-07-17.
• Runway is one company, not two unrelated subjects, but “Runway” now ambiguously refers to Creative, Dev, Robotics or the company’s research previews. GWM Worlds 2 should not be presented as generally available Creative functionality.
• Plan access, credit prices, regional availability and the third-party model catalog are live product state and may change after the observation timestamp.

Sources¶

Agent brief¶

• Subject: organization:runway, thread runway-video-ai-development, 3 dated events 2023-03-20 → 2024-09-19.
• Practical note: As of 2025-07-18, practitioners should not change production workflows from this timeline alone. Use it to locate the Gen-2, longer-duration video, and Lionsgate-partnership milestones, then verify current first-party Runway documentation before relying on any capability, availability, or partnership implication.
• Confidence: medium. Dated supersedes above are the authority for what is obsolete.