The 10 Tools AI Filmmakers Actually Use

Bea hosting, Ada on the tools, same rules as our festivals list. Criteria, in plain words: ranked by verified usage in this library's festival coverage — the tools finished films and festivals actually credit — with each entry's capabilities taken only from the tool's own documentation page, opened in full, as of November 5, 2026. Two honest notes before the countdown. First: Suno, the music tool, ranks second for music in the festival reporting under this list — but its own documentation wouldn't open in this run, so it gets this mention and no capability entry, and a finishing tool takes its slot. The same goes for the voice tool ElevenLabs, whose documentation wouldn't open either. Second: Sora isn't ranked, because it's closed — OpenAI's own help pages say the website and apps were switched off on April 26, 2026 and the service for other builders on September 24, 2026. It's history, and it's good history, but you can't make a film with it this week. Ada, take it away.

10. Descript — You Edit the Video by Editing the Words

The basics: an editor for video and audio, from Descript's own features page. What the page says it does: you edit the transcript and the picture follows — delete a word and that moment of media goes with it; move a sentence and the footage moves too. Its built-in co-editor, Underlord, takes a raw recording and a goal and produces a fine cut; Studio Sound strips background noise and echo and lifts the speaking voice; Regenerate fixes a flubbed line by typing the correction, with a video version that adjusts mouth movements to match; captions, translation dubs and clip-making sit in the same page. The free plan carries 100 AI credits, one time, per the same page.

Ada's notebook: where it sits in the pipeline — the finish, for dialogue-led work. A festival short built from generated shots still needs its talk tightened, its pauses cut and its captions burned in, and that's the tool on the list that treats the script as the timeline. It's a substitute entry, and it says so: it's here because a voice tool's documentation wouldn't open for this run (see the intro), and a tools list that skips the edit isn't a pipeline. It's a shopping basket.

9. Topaz Video — It Upscales and Cleans Up Finished Clips

The basics: a video enhancement toolset, from Topaz Labs' own documentation for its web app. What the docs say it does: three jobs, which you can run in one render. Enhance and upscale — the docs' example takes 720p up to 4K — with named models for named problems: Starlight Precise 2.6 keeps a frame's contents faithful and upscales to 4K (5 credits for 1080p output, 20 for 4K, priced on a 10-second clip at 30 frames per second); Astra 2 is the creative model the docs point at AI-generated and heavily degraded footage, adding detail the source doesn't hold; Proteus is the general-purpose cleanup. Frame interpolation with Apollo 8 builds new frames between yours — up to 30, 60, 90 or 120 frames per second, or slow motion at 2× to 16×. Hyperion 2.5 converts ordinary footage to HDR and delivers ProRes, EXR or H.265, and the docs are blunt that it runs last.

Ada's notebook: where it sits — the very end, after the edit locks. Generated shots often arrive softer and smaller than the screen they'll play on, and that's the pass that makes a 720p take presentable next to a 4K one without regenerating it and losing the performance you liked. It's the second substitute entry, in the slot the music tool couldn't take (see the intro), and it's earned its keep in maker credits elsewhere in this library's coverage — the finish is a real job, and that's its tool.

8. Pika — It Generates a Shot and Splits the Sound Into Separate Parts

The basics: a video generation tool, covered here through its newest model, Pika H3 Flex, from Pika's own launch post. What the post says it does: a 15-second clip with sound that lines up with the picture, generated natively at 480p or 768p, with up to 12 reference files in one generation — images, video and audio together, each told what it should contribute. Through its developer service you can ask for the soundtrack in five separate parts — dialogue, score, effects, foley and room tone — and Pika's upsampler takes a chosen clip up to 1080p, 2K or 4K for the finish. The model's weights are released openly, per the same post.

Ada's notebook: where it sits — shots, with one foot in the sound department. That stem split is the notebook detail that matters: every other video tool on this list hands you sound baked into the picture, and Pika hands you the mix in pieces, so the editor can turn the music down without touching the conversation. Filmmakers storyboard with references, generate with references, and finish with stems — this entry covers two of those three steps in one tool.

7. Luma Dream Machine (Ray2) — You Set a Shot's First and Last Frames

The basics: Luma's Dream Machine, running its Ray2 model, from Luma's own model announcement and developer documentation. What the pages say it does: base clips of 5 to 10 seconds at 540p, 720p or 1080p, with extension holding a shot together out to 30 seconds, upscaling to 4K, and sound support listed in the model's spec table. Its listed features are keyframes — you set the pictures a shot travels between — seamless looping, camera moves named in the prompt, and style references. The developer docs default to the Ray2 model and expose the same controls: aspect ratio, looping, keyframes, duration. Ray2 is for paid subscribers, per the announcement.

Ada's notebook: where it sits — shots, especially the slow ones. The keyframe habit is the transferable skill here: pick the first frame and the last frame, let the model find the road between them, and you've directed the shot instead of requesting it. That's the same instinct as the still-first workflow in our Astana winner's story, linked below — the picture comes first, the movement second, one decision at a time.

6. Midjourney — It Makes the Still Images a Film Is Planned With

The basics: an image generator, from Midjourney's own getting-started documentation. What the docs say it does: you type what you'd like to see into the Imagine bar and it makes a set of four images; from there the docs walk you through upscalers, variations, zoom out, pan and an editor for refining the keepers. You can feed it pictures as well as words — image prompts to steer content and composition, style references to match a look — and its own guide points onward to animating your images into videos. A subscription is required before you create anything, per the same page.

Ada's notebook: where it sits — the very start of the pipeline, the storyboard. That's how a maker's notebook actually fills up: forty versions of a room before a single frame moves, a character's face settled in stills while faces are cheap, a look locked so every later tool has something to aim at. It doesn't make your shots. It makes the decisions your shots depend on, and it makes them at the speed of typing.

5. Higgsfield — It Puts the Whole Workshop in One Platform

The basics: a platform of video and image tools, from Higgsfield's own changelog — the same company whose festival tops our companion list, where every entry was made inside the platform. What the changelog says it does: it ships tools weekly, and the recent entries name the filmmaking kit plainly. Intone makes talking characters: pick a character and a location, add speech — an uploaded recording or generated audio — and get a lip-synced video with a voice that stays the same from scene to scene, at 720p or 1080p, with speech limited to 30 seconds per generation. Content Factory turns a topic into a finished clip of 15 seconds to 5 minutes, with each voiceover line generated separately so it can be redone on its own. Shorts builds vertical pieces from 10 seconds to 2 minutes around a character and a motion reference. And the platform hosts other makers' models too — the changelog records Kling 4.0 arriving with 3-to-30-second generations, per Kling.

Ada's notebook: where it sits — everywhere, which is its point and its risk. One workshop means one login, one credit balance and characters that carry from the talking-head tool into the festival entry. It also means your whole film lives inside somebody else's building. Five thousand finished films came out of that building for its festival, per the company's own count, printed as its figure — so the workshop clearly works. Makers just choose, eyes open, between a rented studio and a toolbox they own piece by piece.

4. Runway — Gen-4.5 Makes the Shots and Act-Two Copies a Real Performance

The basics: Runway's video platform — the pair its own festival's rules name as the reference tools — from Runway's own help pages. What the pages say they do: Gen-4.5 makes video from words or from a starting picture, in lengths of 2 to 10 seconds, at 720p and 24 or 25 frames per second, for 12 credits per second, on the Standard plan and up. Act-Two is the performance tool: you provide a video of someone acting the scene and a character picture or video, and it transfers the movement, speech and expression onto your character — up to 30 seconds, at 5 credits per second with a 3-second minimum. With a character picture you also get gesture control; with a character video it keeps the original camera move and handles the face.

Ada's notebook: where it sits — shots, and the acting inside them. Every other generator on this list invents a performance from a description; Act-Two starts from a person actually performing, which is why the Astana winner's trick — cast a real face when the invented ones fall short — has a software cousin here. Record the scene on your phone, badly lit, in your kitchen. Then give the performance to the character who couldn't be in the room.

3. Google Veo 3.1 — It Makes Clips of 4, 6 or 8 Seconds

The basics: Google's Veo 3.1 video model, from Google Cloud's own model documentation. What the docs say it does: one go makes a clip of 4, 6 or 8 seconds — up to four clips at a time — in tall 9:16 or wide 16:9, at 24 frames per second, with output at 720p, 1080p or 4K on the main model; the Fast and Lite versions top out at 1080p. It works from words alone, from a still picture, or from a first and last picture the clip travels between; you can hand it reference pictures to keep a person or a thing looking the same, though reference work runs only at 8 seconds; sound generation is supported; and a finished clip can be extended further.

Ada's notebook: where it sits — shots, at the precision end. Eight seconds is a leash, and good makers treat it as one: one action, one camera idea, nothing asked of the clip that a clip can't hold. A seven-minute festival short at these lengths is dozens of separate goes before the re-takes, and holding the seams between them is the job — Veo's reference pictures are one honest answer to that job, stated in its own docs and in nobody's launch adjectives.

2. Kling VIDEO 3.0 — It Makes a Multi-Shot Scene in One Generation

The basics: Kling's VIDEO 3.0 model, from Kling's own model guide. What the guide says it does: one go runs 3 to 15 seconds, at a length you choose inside that range, with sound made alongside the picture — including speech in Chinese, English, Japanese, Korean and Spanish. It can plan several shots inside a single go, cutting between views without you stitching pieces, and it can lock onto a person or object you show it, so the same face stays the same face while the camera moves; an element can carry its voice with it. The guide prices the work per second: at 1080p, 12 credits a second with sound and 8 without; at 720p, 9 with sound and 6 without — a five-second 1080p clip with sound costs 60 credits, by the guide's own example.

Ada's notebook: where it sits — shots that want to be scenes. The multi-shot go is the entry's whole argument: a cut is a directing decision, and a tool that lets you describe the cut inside the generation — wide, then close, then the hands — saves you from faking edits in the timeline later. Kling also lives inside other platforms on this list, including Higgsfield's, which is part of why its name keeps surfacing in maker credits across this library's coverage.

1. ByteDance Seedance — Festival Entrants in Japan Used It Most

The basics: ByteDance's Seedance video model family, from ByteDance Seed's own pages — the 2.0 model page and the 2.5 launch announcement. Why it's first: at AI Film Festival Japan, Seedance was the most-used video tool among the entries, per the festival reporting linked below, and Reply's own release credits Seedance as one of the three tools its Venice winner was made with. What the company's pages say it does: the 2.0 model is built on one architecture that makes sound and picture together, taking words, images, audio and video as inputs. The 2.5 announcement extends a single generation to 30 seconds — the longest single go on this list — with further extensions after that; it accepts up to 30 images, 10 video clips and 10 audio clips as references in one pass; and it offers editing aimed at specific moments in a clip, including green-screen and camera-perspective changes, per the same announcement.

Ada's notebook: where it sits — shots, at the length where shots start becoming scenes. Thirty seconds in one pass doesn't just save stitching; it changes what a maker dares to attempt, because a scene with a beginning, a turn and an end can be judged as a scene instead of as three clips and a hope. The festival field in Tokyo reached for it most, the Venice winner carries its name in the credits, and its own documentation backs the ranking with the longest leash in the business. That's a number 1 you can check three ways.

Bea to close: that's the pipeline, end to end — storyboard on Midjourney, shots on Seedance, Kling, Veo, Runway, Luma and Pika, a workshop in Higgsfield if you want one roof, the edit in Descript and the polish in Topaz. The two empty chairs at the table belong to the music and voice tools whose pages wouldn't open for us this run; when their documentation opens, they'll be re-ranked on the same basis as everything else here, and not before. Bring receipts. We do. — Bea