Generative AI: The 2026 Field Guide

Tencent Hunyuan Video

Tencent's open-source video model. Released in late 2024, Hunyuan Video is a 13-billion-parameterParameterOne individual weight. Model size is quoted in parameters, in the millions or billions. More parameters can mean more capability and more cost. diffusion transformerTransformerThe neural network architecture built on attention that underpins modern language and, increasingly, image and video models. It scales well and handles long-range relationships better than convolutional networks. with strong general-purpose video generation. It is one of the major open-source alternatives to Wan and is widely used in the ComfyUIComfyUIThe dominant node-based interface for running diffusion models locally. Started by 'comfyanonymous' in January 2023, now stewarded by Comfy Org. community. Tencent's strategy mirrors Alibaba's: open-weight releases that drive cloud and ecosystem adoption.

LTX Video (Lightricks)

LTX is an open-source video model from Lightricks (the company behind Facetune, Photoleap, and other consumer creative apps). LTX is notable for being unusually fast, it is optimized for real-time or near-real-time video generation, trading some quality for substantially lower compute costs. LTX-2, released in early 2026, is a significant upgrade: it adds built-in audio generation, multi-keyframeKeyframeA specific frame you fix as an anchor, for example the first and last frame of a shot, that the model generates toward or between. support for precise control over shot composition, and outputs up to 20 seconds of 4K video that can run locally on RTX consumer hardware. LTX-2 represents a meaningful step in making high-quality video generation accessible without cloud compute, which is relevant for operators who need to keep generation on-premise for data privacy or cost reasons.

Mochi 1 (Genmo)

Mochi was an open-source video model from a startup called Genmo, released in late 2024. It was briefly competitive with the best open-source alternatives of its era. Genmo as a company has since been less prominent in the field, but Mochi's weightsWeightsThe learned numbers inside a model that encode everything it knows. 'Open weights' means these are downloadable. remain available and it is still used in some workflows.

Novi AI Long Video Agent

Novi AI released a Long Video Agent in early 2026 that pushed the limit to 5 minutes for narrative videos, a significant long-form coherence milestone that the document's the visual-roadmap chapter forecast had pegged for late 2026. The agent uses a multi-step generation pipeline that plans shots, generates them individually, and then stitches them with consistency checks, rather than generating 5 minutes of video in a single pass. This is an important architectural distinction, it is an agentic workflow built on top of existing video models rather than a single model that generates long video natively. The result is usable for certain narrative use cases (storyboarding, pre-visualization, social content) but not yet at the quality level needed for finished production work.

CogVideoX

An open-source video model from a Chinese research group at Tsinghua University. CogVideoX has been incrementally improving across multiple versions and is one of the more academically-grounded open-source video projects. Less polished as a product than Wan or Hunyuan, but useful as a research baseline.

Pyramid Flow

An open-source flow-matching video model that demonstrated the rectified flow techniques (which we covered in the foundations) could be applied effectively to video. Influential as a research contribution; less widely used as a production tool.

With image and video models covered, the next part addresses the specialized models, editing, lipsyncLipsyncSynchronizing the mouth movements of a person in video with an audio track of speech., 3D generation, upscalingUpscalingIncreasing an image or video's resolution after generation, often with a model that invents plausible fine detail rather than simply stretching pixels., that do not fit neatly into either category but are essential parts of the production pipeline.

Other notable video models