Comparisons of AI video generators go stale quickly. Models are renamed, features move between tiers, and products shut down. So this guide does two things. First, it records what each vendor says on its own official pages as of October 2026, with links so you can check whether anything has changed. Second, and more useful in the long run, it gives you a way to choose a tool for a specific short film, whatever the current lineup looks like.
No prices, durations or resolutions appear here unless a vendor states them on a page we could open. Where a vendor page does not say something, we do not fill the gap.
Start with the film, not the model
Most filmmakers pick a generator because of a demo reel. That is backwards. A short film has specific demands, and they are rarely the ones that make a good demo. Before you compare tools, write down what your film actually needs:
- Recurring characters. Does the same person appear in many shots? Then character consistency outranks raw image quality.
- Dialogue. Will characters speak on screen? Then you need generated speech with lip movement, or a plan to record voices separately and sync them.
- Existing footage. Are you transforming live-action material rather than generating from scratch? Then video-to-video editing matters more than text-to-video.
- Finishing. Will the film be graded and mixed in a professional pipeline? Then export formats matter.
- Rights and disclosure. Where will the film be shown? Festivals and distributors may ask what tools you used and under what terms.
Keep that list next to you as you read the sections below.
Sora (OpenAI): discontinued
If you were planning to use Sora, check its status first. In March 2026, OpenAI announced that it is discontinuing Sora, both the app and the API, as Engadget reported. OpenAI's help center article What to know about the Sora discontinuation is the place to confirm the exact shutdown dates and how to export anything you made there.
The practical lesson for filmmakers: if a project depends on a single hosted model, archive your generations, prompts and reference images as you go. If the service disappears mid-production, you keep your material, even if you cannot generate more shots in the same style.
Veo (Google DeepMind)
As of October 2026, Google DeepMind's Veo page presents Veo 3.1 as its leading video generation model and describes it as designed for filmmakers and storytellers. According to that page:
- Veo generates audio natively, including sound effects, ambient noise and dialogue.
- You can guide generation with reference images of a scene, character or object, and the page also lists style references, first and last frame control, scene extension, and object insertion and removal.
- Outputs are offered in 1080p and 4K.
- Veo is reachable through several Google products, including the Gemini app, Flow, Google Vids, Google AI Studio and the Gemini API.
- Generated videos are marked with SynthID watermarking.
The same page notes that natural, consistent spoken audio is still an area of active development, especially for short speech segments. If your film depends on dialogue, test a full scene before you commit.
Runway
As of October 2026, Runway's homepage presents Gen-4.5 as its headline video model, described in terms of motion quality, prompt adherence and visual fidelity, inside a workspace for generating and editing video, images and audio. For short films, three parts of Runway's documentation are especially relevant.
Consistent characters and locations
Runway's Gen-4 research page says the model can generate consistent characters, locations and objects across scenes using visual references combined with instructions.
Editing existing footage
Runway's changelog describes Aleph 2, including a September 2026 entry about restyling footage already on your timeline through Runway's Adobe plugins. That makes Runway worth testing if your film mixes live action with generated material.
Finishing
The same changelog lists an August 2026 update that lets you export Gen-4.5 and Aleph 2 generations as ProRes 4444 with PCM audio or as a PNG sequence. For a film headed to a colorist and a sound mix, that is a practical advantage over compressed web exports.
Kling
As of October 2026, the Kling AI homepage headlines Kling 4.0, described in terms of audio-visual performance, alongside an image model, a node-based canvas and an agent for scripts and storyboards. The homepage itself does not detail reference features, but Kling's Element Library user guide does:
- An element is a reusable asset for a character, prop or scene, built from at least two and up to four reference images.
- The guide lists Kling VIDEO 3.0, VIDEO 3.0 Omni and Kling O1 as models that can generate with elements from the library.
- In VIDEO 3.0 Omni, character elements can also have a voice bound to them.
For a short with a small recurring cast, a library of saved elements is a useful way to work. Check whether the newest model on the homepage supports the same element features before you plan around them.
How to choose: a test-shoot method
Feature lists only get you so far. The reliable way to choose is the same way a cinematographer chooses a camera: shoot a test.
- Pick your hardest scene. Not your prettiest. The one with the most shots of your lead, or the dialogue exchange, or the transition between live action and generated footage.
- Write it as a shot list. Use the AI shot planner so every tool gets the same wording for character, setting and look.
- Prepare the same references for every tool. A small set of character images and one location image. See the guide on consistent characters in AI films.
- Generate the scene in two or three tools. Keep notes on how many generations each shot needed before it was usable.
- Cut each version together. Judge the sequence, not individual clips. A tool that makes beautiful single shots but drifts between them will lose on the cut.
- Read the terms. Before committing, read each tool's current terms of use on commercial use and ownership of outputs, and save a copy with the date.
Questions to score each tool on
- Does the lead look like the same person across five consecutive shots?
- Does the location hold across a shot and its reverse?
- Can it produce dialogue you would put in front of an audience, or will you record voices separately?
- Can you edit or restyle footage you already have?
- Does the export fit your finishing workflow?
- Does the tool mark its outputs, and how will that interact with festival disclosure rules?
- If the service changed or closed tomorrow, would you still have your raw material?
Combining tools is normal
Few AI shorts are made in one tool. A common pattern is to generate key character shots where reference control is strongest, generate environments or inserts where image quality is strongest, and handle voices, music and finishing elsewhere. That only works if your prompts and references are organized, which is why planning matters more than tool choice.
Whatever you use, keep a production log: tool, model version, date, prompt and reference files for every shot you keep. Festivals increasingly ask for this, and the guide to AI film festivals explains why.
After the film is finished
An AI short still needs an audience. Once you lock the cut, plan a teaser and a set of vertical clips. You can do it by hand with the teaser planner, or upload the finished short to ViceForge to get a first pass of vertical 9:16 clips with captions added automatically, then refine them in your editor. The guide on promoting a film with clips covers what to do with them.
Checklist before you commit to a generator
- Confirmed on the vendor's official page that the product is still available.
- Tested your hardest scene, not a demo prompt.
- Used identical references and prompt wording across tools.
- Judged the edited sequence, not individual clips.
- Checked export formats against your finishing plan.
- Read and saved the current terms of use.
- Set up a production log and an archive of all generations.
Vendor details above are as of October 2026 and come only from the official pages linked in the text. Features and availability change often, so check those pages before you rely on them.