AI video generator directory: hosted, open and avatar tools

Adobe Firefly Video

Adobe's own Firefly Video Model, which Adobe says is trained on licensed content such as Adobe Stock and public domain material. The same page lists partner models: Google Veo 3.1, Luma Ray3, Kling 3.0 and Runway Gen-4.5. Adobe says most models give clips up to 8 seconds (some up to 15) and downloads go up to 4K. Adobe notes that commercial eligibility depends on the model you pick.

Dreamina

In our picks, #5

ByteDance's creative site and the home of the Seedance video models. Its video list names Seedance 2.5, which the page says generates up to 30 seconds, plus Seedance 2.0 and partner models Sora 2 and Veo 3.1. Image models include Seedream 5.0 Pro and Lite. There are also avatar tools.

Grok Imagine

xAI's image and video generator, built into Grok. The input image is the first frame of the clip.

Hailuo AI

In our picks, #2

MiniMax's video site, which now shows MiniMax H3 as its current model. It offers credits for completing a survey and subscription plans, and it does not state how many free credits you get.

Higgsfield

A multi-model platform rather than a single model. Its site lists its own Cinema Studio 4.0 alongside Seedance 2.5, and it connects to ChatGPT and Claude through an extension.

Kling AI

In our picks, #1

The site lists Kling 4.0 and Video 3.0, and describes Kling 4.0 as an audio-visual model. It also offers Motion Control and ships desktop apps for macOS and Windows.

Luma

Luma's current product is Luma Agents, which runs its own Ray3.2 model and partner models including Veo 3.1, Kling 3.0 and Seedance 2.0, and also handles image, audio and text tasks. Individual plans are Plus at $30 a month (10,000 credits), Pro at $90 (40,000) and Ultra at $300 (150,000), with lower prices when billed yearly. Credits reset each billing cycle.

Moonvalley Marey

Moonvalley's Marey model, which the company describes as trained on licensed data. It adds camera control from a single image, motion transfer from a reference video and trajectory control from drawn paths.

Neural Frames

Audio-reactive music videos: you upload a track and the site says it analyzes eight stems to drive the visuals. You can pick from several video models (it lists Kling, Seedance, Runway and Stable Diffusion) or train a custom model on your own images for character consistency. It also lists lipsync and 4K upscaling. The company says it is bootstrapped and based in Berlin.

Pika

In our picks, #4

Now a multi-model creative platform. The site lists Pika 2.5, Seedance 2.5, MiniMax H3 and Wan 3.0 for video, plus image and audio models and tools such as Extend Video and Character Studio.

PixVerse

Lists its own PixVerse V6, which it says outputs 1080p with native audio, and R2, described as a real-time world model, next to partner models such as Seedance 2.5 and MiniMax H3. Also lists lip sync, character consistency and a command-line tool.

Runway

In our picks, #6

The pricing page lists Gen-4.5 and Aleph 2.0 as Runway's own models, alongside partner models such as Kling 3.0 and Seedance 2.5. Paid plans are $12 (Standard), $28 (Pro) and $76 (Max) a month on annual billing, or $15, $35 and $95 billed monthly. Credits are monthly (625, 2,250 and 9,500). The API is sold separately with its own pricing.

Sora

OpenAI's deprecations page lists the Sora 2 models and the Videos API as deprecated on March 24, 2026, with the API ending on September 24, 2026 and no recommended replacement. The web and mobile apps shut down on April 26, 2026. It stays listed so a search for it lands somewhere useful.

Veo

Google's video model, shown on its page as Veo 3.1. Google says it generates sound effects, ambient noise and dialogue natively with the video. The page cites 1080p and 4K output and lists 8 seconds as the standard clip length.

Vidu

Lists Vidu Q3, which it says generates audio with the video, and Vidu Stream, a real-time interactive model. The site cites up to 1080p, 16-second audio-visual generation, and a My References tool for keeping a character consistent. It also offers an off-peak mode aimed at cheaper experiments.

HunyuanVideo

A video model with over 13 billion parameters. The repository lists a minimum of 45 GB of GPU memory for 544x960 clips and 60 GB for 720x1280, both at 129 frames, and recommends 80 GB. FP8 weights are available to cut memory use. A separate image-to-video model, HunyuanVideo-I2V, is linked from the same page. The license excludes the EU, UK and South Korea.

LTX-Video

Open weights in 2B and 13B sizes, with distilled and FP8 versions of each. Supports keyframe conditioning and video extension. Weights are on Hugging Face. The README now says development has moved to LTX-2, which it describes as generating synchronized audio and video, and points to a separate repository for it.

Mochi 1

An open-source text-to-video model with 10 billion parameters in the diffusion model. The model card lists 480p output and clips of up to 84 frames at 30 fps, about 2.8 seconds, and calls the checkpoint a research preview. Weights are on Hugging Face and code on GitHub. Genmo also runs a hosted playground, and the model works in ComfyUI.

Stable Video Diffusion

An image-to-video model: it animates a single still. The XT model produces 25 frames at 576x1024, roughly four seconds or less. The card lists limits: no text control, weak text rendering and difficulty with realistic faces. The license allows commercial use below USD 1 million in annual revenue; above that, a separate license from Stability is needed.

Wan

The Wan2.2 repository lists T2V-A14B, I2V-A14B, TI2V-5B, S2V-14B and Animate-14B. Stated single-GPU needs: at least 80 GB of VRAM for T2V-A14B and S2V-14B, and at least 24 GB for TI2V-5B, which fits an RTX 4090. Weights are on Hugging Face and ModelScope.

D-ID

Talking avatar videos from a script, a deck or a document. You can pick a ready-made avatar or upload a face image (JPEG or PNG, up to 10 MB). Standard output is up to 1280x1280, with 1080p on Pro and above, and clips run up to 5 minutes.

HeyGen

Avatar and presenter video from text, images, audio, slide decks or a YouTube link for translation. The pricing page cites translation in 175+ languages on paid plans (30+ on Free) and lists voice cloning from the Creator plan.

Synthesia

Avatars that present a script in 160+ languages and voices, per the pricing page. Dubbing covers 70+ languages on most plans and 140+ on Enterprise. Interactive avatars handle real-time conversations. Works from text, documents, links and PowerPoint files. API video creation needs the Pro plan or higher.