Runway Aleph
Last reviewed
May 16, 2026
Sources
41 citations
Review status
Source-backed
Revision
v1 ยท 3,993 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 16, 2026
Sources
41 citations
Review status
Source-backed
Revision
v1 ยท 3,993 words
Add missing citations, update stale details, or suggest a clearer explanation.
Runway Aleph is an in-context video editing model developed by Runway, the New York based generative AI company founded by Cristobal Valenzuela, Alejandro Matamala, and Anastasis Germanidis. Aleph was introduced on July 25, 2025 through a Runway Research blog post titled "Introducing Runway Aleph," and is positioned by the company as "a state-of-the-art in-context video model, setting a new frontier for multi-task visual generation." The model accepts an existing video clip plus a natural-language instruction and produces an edited version of that clip, with the ability to add or remove objects, transform style and lighting, change weather or time of day, and synthesize entirely new camera angles from the original footage.
Unlike Runway's earlier text-to-video and image-to-video flagships such as Runway Gen-4, which generate fresh footage from scratch, Aleph is designed around the source clip itself. It treats the input video as the context for every edit, preserving identities, motion, and scene geometry while rewriting selected aspects of the picture. That design intent is reflected in Aleph's branded API name, runway-gen4-aleph, which signals that the model is built on top of the Gen-4 architecture family while specializing in video-to-video transformation. Aleph rolled out first to Runway's Enterprise and Creative Partners and within days became broadly available to paid subscribers on the Standard, Pro, Unlimited, and Enterprise tiers of runwayml.com, along with API access through the Runway developer portal and resellers including Fal.ai, WaveSpeed AI, Runware, Segmind, and Eachlabs.
Aleph competes with a small but expanding category of video-to-video editing models, most notably Luma Labs' Modify Video feature inside the Luma Dream Machine, Pika Labs' Pikadditions and Pikaffects features inside Pika 2.5, and Alibaba's Wan 2.5 family. Pure text-to-video models such as OpenAI's Sora 2 and Google DeepMind's Veo 3 sit in a different bracket because they create video from scratch. Runway's pitch is that the future of generative video is editing existing footage rather than hallucinating it whole, and that Aleph is the first production-grade tool aimed at professional film and television workflows from that angle.
Runway was founded in 2018 by three students at New York University's Tisch School of the Arts and has spent most of its history building generative tools for filmmakers and visual artists. The company released Gen-1, a video-to-video model that applied a style or reference image to an input clip, in February 2023, followed by Gen-2 in March 2023, Runway Gen-3 Alpha in June 2024, and Runway Gen-4 in March 2025. Each release pushed video output toward higher resolution, longer duration, and stronger character consistency, but all of them were primarily focused on generating new footage from text or image prompts rather than editing existing video at the shot level.
Aleph is the first Runway model explicitly framed as an editor rather than a generator. The launch post describes its design as performing "a wide range of edits on an input video such as adding, removing, and transforming objects, generating any angle of a scene, and modifying style and lighting." The model arrived during a period of intense activity in the video-editing AI category. Within a single 12-month window, Black Forest Labs shipped the Flux.1 Kontext image-editing model in May 2025, Luma Labs launched Modify Video, Pika Labs shipped Pika 2.5 with in-scene insertion, and Alibaba Cloud released Wan 2.5 with native audio. Aleph was positioned as the first video-to-video tool with general-purpose editing capabilities aimed at a professional workflow.
The launch followed a wave of Hollywood-facing deals. In September 2024 Runway announced a partnership with Lionsgate, the studio behind franchises such as John Wick and The Hunger Games, to train a custom Gen-4 derivative on Lionsgate's owned catalog. Runway has also signed with AMC Networks and counts production work from films such as Everything Everywhere All At Once among its case studies.
Runway introduced Aleph on July 25, 2025 through a Runway Research blog post and a launch campaign by Cristobal Valenzuela. Valenzuela posted demonstration clips on X with the caption "godmode with Runway Aleph," and on August 1, 2025 followed up with a tongue-in-cheek announcement of an Aleph Programming Interface framed as a Jorge Luis Borges parable, describing the model as "that impossible interface through which programmers might summon any video that ever was or could be." Borges' 1945 short story "The Aleph" gives the model its name; in the story, the Aleph is a single point in space that contains all other points.
Access opened first to Runway Enterprise customers and Creative Partners, then expanded to the company's full paid user base within days. CineD reported the initial rollout as gated behind higher tiers, but by early August 2025 Aleph appeared on the standard runwayml.com web app for any user with an active paid subscription. Runway's documentation site published a Creating with Aleph guide and an Aleph Prompting Guide alongside the launch, with worked examples for object removal, scene relighting, weather changes, and angle reframing. The API endpoint runway-gen4-aleph went live around the same time, with model ID runway:2@1 on partner platforms.
Aleph performs in-context video editing, meaning that the input video clip is itself the conditioning signal that constrains the output. The model parses scene structure, lighting, depth, and motion across the entire clip rather than processing frames individually, and rewrites selected aspects of the picture in response to a natural-language instruction. Runway groups the supported tasks into a small set of categories on its prompting guide, summarized below.
| Capability | What it does |
|---|---|
| Object insertion | Adds people, animals, props, or environmental elements to the existing scene with matched lighting, perspective, and shadow |
| Object removal | Deletes people, props, or unwanted artifacts from the scene and synthesizes the area that was previously hidden behind them |
| Object transformation | Swaps one element for another, retextures surfaces, ages or de-ages characters, or restyles clothing and hair |
| Scene angle generation | Synthesizes a new viewpoint of the recorded scene using reconstructed 3D geometry, including reverse shots, close-ups, wide shots, and aerial perspectives |
| Style transfer | Applies an aesthetic such as anime, watercolor, or a specific film stock to the entire clip while preserving motion |
| Relighting | Replaces the on-set lighting with a new lighting setup such as golden hour, blue hour, harsh noon, or moody interior |
| Environment change | Replaces the surrounding environment, including weather, time of day, season, or a complete location swap |
| Green screen generation | Isolates a foreground subject and outputs a clean key matte suitable for compositing |
| Motion transfer | Drives a still image with the motion captured in a reference video clip |
| Shot continuation | Extends a clip past its original final frame while preserving continuity of the scene |
The model accepts video uploads in standard formats including MP4, WEBM, and MOV, and outputs renders that Runway lists as supporting 720p, 1080p, and 4K depending on the export pipeline used. Through the Runware partner endpoint, Aleph supports common aspect ratios at the input stage including 1280 by 720 (16:9), 720 by 1280 (9:16), 1104 by 832 (4:3), 832 by 1104 (3:4), 960 by 960 (1:1), 1584 by 672 (21:9), 848 by 480, and 640 by 480. Generation length per pass is short, with Runway help documentation and third-party reviewers describing a five-second maximum per single Aleph generation, and a typical render time in the one to three minute range depending on edit complexity. Longer pieces are assembled by chaining Aleph generations together, often using Aleph's own shot continuation feature to keep visual coherence across cuts.
The most-discussed feature in launch coverage was scene angle generation. CineD framed it as a workflow change that "could revolutionize coverage acquisition," because a director can shoot a single master plate on set and then generate coverage including close-ups, over-the-shoulder shots, and reverse angles entirely in post. The capability rests on the model reconstructing the implicit 3D geometry of the scene from the source clip and then re-rendering it from a different virtual camera position. Object removal and object insertion sit on the same backbone; the model has to understand what is in front of what before it can repaint hidden areas correctly.
The environment-change feature is similarly tied to scene understanding. Aleph can turn a sunny morning into a rainy evening, drop snow into a previously dry scene, or shift the time of day from noon to golden hour, with shadows, reflections, and ambient color all adjusting to the new lighting. Reviewers at Filmora, Curious Refuge, and The Decoder have flagged lighting and environment edits as the most production-relevant features for advertising and broadcast, since those edits previously required either reshoots or expensive VFX passes.
Runway has not published a technical paper for Aleph and the company's launch materials do not disclose parameter count, training data, or specific architecture. Public technical disclosure is limited to two points. First, the model is part of the Gen-4 architecture family, exposed on Runway's developer portal as runway-gen4-aleph and described by Runway as "a state-of-the-art in-context video model." Second, the model is engineered for full-video context understanding rather than frame-by-frame processing, which is what allows it to preserve visual coherence across a single clip during edits.
Third-party partner platforms have added some specifications, though those should be read as their best public understanding rather than confirmed Runway disclosures. The Cliprise model guide refers to Aleph as a 14 billion parameter model built on a video diffusion-transformer backbone, while Runware lists it as model runway:2@1. Neither figure is sourced to a Runway publication. Anastasis Germanidis, Runway's chief technology officer, has spoken in a Semafor profile from August 2025 about the company's direction of building video models that approach "simulating reality," with applications extending into robotics and synthetic data. The architectural specifics for Aleph itself remain undisclosed as of mid 2026.
Aleph is available to end users through three primary surfaces. The first is the runwayml.com web application, where any user on a paid Runway subscription can upload a clip, type a natural-language instruction, and run an Aleph edit job. The second is the Runway developer API, where Aleph is exposed under the runway-gen4-aleph model name and billed on a per-credit basis. The third is partner platforms including Fal.ai, WaveSpeed AI, Runware, Segmind, and Eachlabs, each of which resells access to the same underlying Runway endpoint at its own pricing.
On the consumer plans, Aleph is included starting at the Standard tier. Runway charges $15 per month for the Standard plan billed monthly, or about $12 per month billed annually, and that plan ships with 625 monthly credits and access to Aleph alongside Gen-4.5 text-to-video, Gen-4 image-to-video, Runway Act-Two motion capture, and the rest of Runway's model lineup. The Pro plan at $28 per month with annual billing adds 2,250 monthly credits, and the Unlimited plan at $76 per month adds unlimited generations in the lower-cost Explore Mode. Enterprise pricing is custom. The free tier does not include Aleph.
| Model | API credit cost | At $0.01 per credit |
|---|---|---|
| Gen-4 Aleph | 15 credits per second of generated video | $0.15 per second |
| Gen-4.5 | 12 credits per second | $0.12 per second |
| Gen-4 Turbo | 5 credits per second | $0.05 per second |
| Gen-3 Turbo | 5 credits per second | $0.05 per second |
| Act-Two | 5 credits per second | $0.05 per second |
| Veo 3 (resold) | 40 credits per second | $0.40 per second |
At the Runway developer rate, a typical Aleph generation of five seconds costs 75 credits, or about $0.75. A 30-second sequence assembled from six Aleph passes would cost roughly $4.50 of API credit. That places Aleph between Runway's own faster generators such as Gen-4 Turbo and the more expensive resold endpoints such as Google's Veo 3. Runway's API documentation also notes a tiered access model in which organizations can apply for higher concurrency limits and custom rate caps from the developer portal.
Aleph sits in a small but growing category of in-context video editing models. The closest direct comparison is Luma Labs' Modify Video feature inside Luma Dream Machine, which shipped in 2025 with three preset strength modes: Adhere, Flex, and Reimagine. Pika 2.5 from Pika Labs ships Pikadditions for in-scene object insertion and Pikaffects for stylized transformations, and Alibaba's Wan 2.5 family inherits some video editing capacity through the Wan VACE lineage. Pure text-to-video generators such as Sora 2 and Veo 3 are not direct competitors in the editing category, although they are often grouped with Aleph in industry coverage because they share the same general video AI market.
| Dimension | Runway Aleph | Luma Dream Machine Modify | Pika 2.5 | Wan 2.5 |
|---|---|---|---|---|
| Primary task | In-context video editing | In-context video editing | Object insertion plus stylized effects | Multimodal generation with editing variants |
| Provider | Runway | Luma Labs | Pika Labs | Alibaba Cloud Tongyi Lab |
| Camera angle synthesis | Yes, full 3D scene reframing | No native angle synthesis | No native angle synthesis | Not in the base Wan 2.5 endpoint |
| Object insertion | Yes | Yes, via Modify Reimagine | Yes, via Pikadditions | Limited, via Wan VACE editing pipeline |
| Object removal | Yes | Yes | Limited | Limited |
| Relighting | Yes | Partial | Limited | Partial |
| Style transfer | Yes | Yes, via Modify presets | Yes, via Pikaffects | Yes |
| Maximum clip length per pass | About 5 seconds | About 10 seconds | About 10 to 15 seconds (Pikaframes extends to 25 seconds) | 10 seconds |
| Resolution | Up to 4K via export pipeline | Up to 1080p | Cinematic short-form output | 1080p, 4K preview |
| Starting consumer price | $12 per month annual, $15 monthly | Included in Dream Machine paid tiers | Included in Pika paid tiers | About $0.10 per second on Alibaba Cloud DashScope |
Reviewers at Filmora, Atlabs AI, and Pixflow describe Aleph as the leading product in its category for camera angle synthesis, since no other tool advertises the ability to generate a new viewpoint of an existing scene by reconstructing 3D geometry. Luma Modify is the closest match on overall edit breadth and is rated as easier for non-technical teams because of its preset strength modes. Pika 2.5 leans toward consumer and social use cases, especially with its Pikaffects library of stylized effects such as melt, explode, and cake-ify. Wan 2.5 is positioned as a price challenger from China rather than an editing specialist; in-context editing in the Wan family is mostly served by Wan VACE rather than by Wan 2.5 itself.
On pricing, Aleph is more expensive per generated second than most alternatives. Runway's 15 credits per second works out to roughly $0.15 at the consumer rate, where Luma and Pika bundle editing into all-you-can-iterate subscription plans. Wan 2.5 is cheaper at roughly $0.10 per second on Alibaba Cloud DashScope. For studios willing to pay for camera angle synthesis and a production-grade workflow, the higher unit cost is the trade-off.
Runway has aggressively pursued film and television integration since 2024, and Aleph is the company's most direct entry into post-production. The Lionsgate partnership signed in September 2024 is structured around a custom AI model trained on Lionsgate's proprietary catalog, separate from Runway's consumer models. Lionsgate executives told the Hollywood Reporter that they expected the AI tools to deliver tens of millions of dollars in savings, with initial applications in storyboarding, background creation, and special effects for action titles. PetaPixel reported in September 2025 that Lionsgate was still working to make the AI-only film concept production-ready, with the studio described as struggling to convert demo capability into reliable output.
Runway also partnered with IMAX on the 2025 AI Film Festival. IMAX screened ten finalist films at ten IMAX theaters across the United States from August 17 through August 20, 2025, in New York, Los Angeles, San Francisco, Chicago, Seattle, Dallas, Boston, Atlanta, Denver, and Washington. Jurors included Gaspar Noe, Harmony Korine, and Jane Rosenthal. More than 6,000 films were submitted to the 2025 edition. Some of the finalist work used Aleph during post-production for relighting, color grading, and shot continuation, although Runway has not published a full breakdown of which films used which models. Valenzuela framed the festival as offering a "premium viewing experience" for AI films, while IMAX chief content officer Jonathan Fischer described the screenings as part of "exploring how tools shape filmmaking."
Netflix has separately spoken about the production economics that tools like Aleph enable. Co-chief executive Ted Sarandos told investors during a July 2025 earnings call that the studio's first use of generative AI in a finished show, on the Argentine series The Eternaut, came out roughly 10 times faster than traditional VFX would have. Sarandos did not name Runway specifically and the cited 10x figure refers to a generative AI VFX pass rather than to Aleph in particular, but the broader investor narrative around generative video in Hollywood gathered momentum in the same period that Aleph shipped.
Industry coverage of Aleph has generally been positive on capability and cautious on production reliability. CineD's launch piece praised the breadth of editing tasks but flagged unanswered questions about resolution, color space, metadata handling, and integration with professional workflows. Winbuzzer credited Aleph with redefining video editing through text prompts while noting that the broader competitive landscape (Adobe Firefly inside Premiere Pro, Google Veo 3, OpenAI Sora) was moving toward similar capabilities. The Decoder framed Aleph as part of a wave of in-context generative tools alongside Black Forest Labs' Flux.1 Kontext image editor and described Runway's advantage as "polished interface and deep set of editing tools," with the expectation that competitors would close that gap quickly.
Reviewers running hands-on tests reported recurring limitations. Techpoint Africa noted that maintaining character and object consistency across shots is still a challenge, that high-quality generations and upscaling burn through credits, and that effective prompting takes practice. Atlabs AI flagged the five-second generation cap as forcing creators to assemble longer pieces from multiple passes, which can introduce continuity drift. Curious Refuge wrote that the model's handling of complex multi-subject interactions occasionally breaks down, especially in scenes with crowds or fast action.
The response from working filmmakers has been a mix of enthusiasm and caution. Filmmaker Darren Aronofsky has spoken supportively about exploring generative tools, while creative unions and some industry groups have continued to raise concerns about IP licensing, job displacement, and the integrity of training data. Disney general counsel Horacio Gutierrez and other studio executives have argued that piracy is piracy regardless of whether an AI model is involved in producing the output, a position aimed at the broader AI video category rather than at Aleph specifically. Runway has maintained that its Lionsgate-style custom model approach, in which a studio's content is used only to train a private model the studio owns, is the right structure for studio collaboration.
Independent creator reaction on X and YouTube has tended toward the practical. Tutorials on Curious Refuge, Blue Lightning TV, and DataCamp have framed Aleph as a tool that compresses what used to take a colorist, a VFX artist, a rotoscope specialist, and a second camera operator into a single text prompt. The same reviewers caution that the tool is best treated as one component in a larger production stack, not a replacement for editorial judgment or non-linear editing software.
Valenzuela himself has used Aleph as a centerpiece of Runway's broader pitch that AI does not have to be framed as artificial general intelligence to be useful. In an October 2025 interview with The Information titled "Not Everyone Is Trying to Build God," he argued that Runway's purpose is to help creators "think about AI as a new kind of camera," and that the next generation of filmmaking will let animators finish in seconds what used to take weeks. In an April 2026 TechCrunch profile he told the publication that AI tools like Aleph could help Hollywood make 50 films for the cost of one $100 million blockbuster, while acknowledging that the industry's economic structures still need to catch up to that capability.