# The Best AI for Video, by Job: Editing Your Footage vs Generating New Clips

> To edit footage you shot, Claude Opus 5.5 now does about 95% of Pat Simmons's Premiere Pro edit. To make new clips from a prompt or a photo, here is what Kling, Gemini Omni, Grok Imagine, Seedance and the rest do and cost, from each company's own pages.

By Pat Simmons · November 8, 2026 · 16 min read

Canonical: https://www.aiformortals.co/blog/best-ai-for-video

The best AI for video depends on the job. To edit footage you shot yourself, Pat Simmons now has Claude Opus 5.5 do about 95% of his edit inside Premiere Pro, through an editing skill and Premiere Pro bridge he published free on GitHub. Before that, in his September 11, 2026 video, GPT-6 Astra did a full edit by clicking through Premiere itself: it cut about 48 minutes of raw footage to about 26 in 56 minutes and used about 15% of his usage on the $100-a-month ChatGPT plan.

To make new clips from a prompt or a photo, Pat hasn't run video generators against each other, so this page compares them from each company's own pages. Kling 4.0 Flash makes 3 to 20-second clips at 720p for 0.6 API units a second, against 1.0 for Kling 4.0. Gemini Omni 1.2 Flash makes clips up to 15 seconds with sound for about $0.07 a second at 720p, and Grok Imagine 1.5 turns a photo into a clip with sound for $0.14 a second at 720p. For animated explainers that Claude or Codex builds for you, HeyGen's HyperFrames is free and open source.

Last checked November 8, 2026: every price, plan and limit below comes from the company's own pricing or launch page, opened that day and linked where it's stated.

## Which AI is best for editing video?

For editing footage you've already shot, the AI Pat Simmons uses now is Claude Opus 5.5 in Claude Code, which does about 95% of his Premiere Pro edit (how Claude Opus 5.5 does 95% of Pat's Premiere Pro edit). Each time he finishes recording a take, Opus 5.5 transcribes it in Premiere, cuts the pauses and bad takes, and adds it to a review sequence while he records the next one, all on a separate Mac Mini.

Most of that work runs through a Premiere Pro bridge, an MCP server (a small add-on that lets an AI agent run another app's commands) that controls Premiere with scripts. Opus clicks through Premiere's buttons itself, which is called computer use, only for Transcribe and the Pauses filter, because those two exist only as buttons. Pat still gives the finished cut a once-over, and that check is the 5% he does himself. The skill and the bridge are free on GitHub; Premiere Pro itself costs US$22.99 a month on an annual plan billed monthly, with a 7-day free trial.

Before Opus took over, Pat taught GPT-6 Astra his editing style one step at a time in the video at the top of this page. Astra cut pauses and bad takes, then added jump cuts, eased push-ins, word-by-word text animations and B-roll. It worked through a mix of the Mac accessibility tree (a text list of an app's menus, panel names and timecodes that every Mac app publishes for blind and visually impaired users) and computer use, because the accessibility tree doesn't expose the clips on the timeline or let it trim their edges. The skill he built is on GitHub as a Codex skill.

_GPT-6 Astra running Premiere's Pauses filter on its own. "No results" means every pause in the clip has been cut. From the video above at 8:08._

On the final run, Astra edited a whole video end to end in 56 minutes, turning about 48 minutes of raw footage into about 26. It wasn't a finished cut: Pat had told it to pull in any B-roll as an example, so the B-roll was random, and the text in the practice run was placeholder wording. His caveat in the video was speed: "it is slow. You can see it took 56 minutes to go through this. However, what you can do is you can just have this running in the background."

Pat has never run Astra and Opus 5.5 on the same footage, so neither has a measured win at editing. In the September 11 video he said "Astra is especially good at computer use, much better than even Fable or Opus." In his October 12 post, he describes moving his own editing to Opus 5.5 with the bridge, which skips most of the clicking. On whether this replaces editors, his words were: "people on social media are absolutely going to say that editors are cooked. That is certainly not true," and "this is Astra just doing the grunt work that you probably don't even like doing as an editor."

 The free option: FilmCraft with Claude

If you don't pay for Premiere, FilmCraft, the free Premiere Pro clone Claude can edit in, is the setup Pat tested. FilmCraft is open source under the MIT or Apache-2.0 license and runs on Mac, Windows, Linux and in a web browser. It ships with its own MCP server, and in Pat's fork Claude Opus 5.5 found 21 pauses in 28 seconds of audio and cut every one in about 40 seconds. It can't open Premiere's .prproj project files yet, so an existing Premiere edit has to come across as an FCP7 XML, AAF or OTIO file.

Editing by chat: CapCut in ChatGPT and Descript

Pat hasn't tested either of these, so this is what each company says. CapCut's plugin for ChatGPT turns a prompt or your own clips into an editable draft, and you then ask in the chat to shorten scenes, reorder clips, change the music or add captions, or open the project in CapCut's editor. CapCut says it's designed for the ChatGPT desktop app. Our CapCut in ChatGPT write-up walks through CapCut's own demo.

Descript edits video by editing its transcript, and its AI co-editor, Underlord, makes edits from a prompt. Its free plan includes text-based editing and 60 minutes of media a month, and paid plans start with Hobbyist at $16 a month billed yearly or $24 month to month.

## What is the best AI video generator?

No single video generator wins from Pat's tests, because he hasn't run them against each other. Going by each company's own specs, Kling 4.0 makes the longest, highest-resolution clips here (30 seconds, up to 4K), Kling 4.0 Flash and Gemini Omni 1.2 Flash are the cheap ways to draft, Grok Imagine 1.5 is built around turning a photo into a clip with sound, and Seedance 2.5 makes 30-second stories with the sound made together with the picture.

ModelLongest clipSoundResolutionWhat you can start fromPrice (checked November 8, 2026)

Kling 4.030 secondsTwo-channel stereo720p, 1080p, 4KText, up to 10 images, up to 5 videos1.0 API unit a second at 720p, 1.5 at 1080p
Kling 4.0 Flash20 secondsNot listed in Flash's specs720pText, images, references0.6 API units a second at 720p
 Gemini Omni 1.2 Flash15 secondsYes, made with the picture360p or 720p, upscaled to 1080p or 4KText, images, motion clips up to 3 seconds, audio cuesAbout $0.07 a second at 720p (API)
 Grok Imagine 1.515 secondsEffects, ambience and dialogue in the same pass480p, 720p, 1080pText, an image, audio$0.14 a second at 720p, $0.25 at 1080p (API)
Seedance 2.530 seconds, extendable twiceMade together with the videoNot listed on ByteDance's pagePrompts, reference videos, editsNo price on ByteDance's page; included in Runway Pro and Max and HeyGen Pro

Kling 4.0 Flash is the cheaper, faster version of Kling 4.0, and it gives up the longer clips and the higher resolutions: Flash stops at 20 seconds and 720p, while Kling 4.0 goes to 30 seconds and 4K. At the API rates above, Flash costs 40% less than Kling 4.0 at 720p. Our Kling 4.0 Flash vs Kling 4.0 pricing page covers the credit math and starter prompts.

_Kling's developer spec table: Flash takes the same text, image and video inputs as Kling 4.0, and tops out at 20 seconds and 720p. Image: Kling AI developer docs._

Gemini Omni 1.2 Flash came out on November 5, 2026, and the change Google leads with is editing by chat: you ask for a change to a clip you already made, like a new background or different weather, and it changes that clip instead of starting over. In our 10-clip Gemini Omni 1.2 Flash test, an AI for Mortals staff test, the ten clips cost $8.58 in total and took 33 to 66 seconds each. Scene and background edits worked when we asked plainly, and both requests to change the camera movement failed.

_One Gemini Omni 1.2 Flash clip and three chat edits from our test: the original, a rainy city window at night, daytime instead of night, and anime._

The Grok Imagine 1.5 announcement says sound effects, background noise and dialogue are made in the same pass as the picture, and that the Fast version on grok.com and the Grok apps makes a 6-second 720p clip in about 25 seconds. In the Grok app, Imagine comes with the Plus plan at $8 a month, and the $100-a-month Super plan adds 1080p, 15-second videos.

Seedance 2.5 is ByteDance's video model, built, in ByteDance's words, "for 30-second storytelling", with the sound and the picture generated together. ByteDance's page lists no price, so most people reach it through another app's plan: Runway includes it on Pro and Max, HeyGen on Pro, and InVideo on Plus and Max.

Runway, Adobe Firefly and Canva

Runway sells credits for its own Gen-4.5 model and for other companies' models. Its Standard plan costs $15 a month ($12 a month billed yearly) for 625 credits, which Runway says is 52 seconds of Gen-4.5, and Gen-4.5 uses 12 credits a second. The same plans also run Kling 4.0, and Pro and Max add Seedance 2.5.

Adobe Firefly makes video with the Firefly Video Model, paid for with generative credits. Firefly Standard is US$9.99 a month for 2,000 credits, and the US$199.99 Premium plan is the one with unlimited use of the video model. Canva puts Image to Video among its Premium AI tools and its "Create high-quality video clips" tool among its Ultra AI tools, and Canva Free gets limited access to Premium AI and none to Ultra AI.

Sora is gone. OpenAI's help page says the Sora web and app experiences ended on April 26, 2026 and the Sora API on September 24, 2026, so Sora and Sora 2 are no longer ways to make video. Our Sora page keeps the history.

## Is there a free AI video generator?

Yes, with limits. These are the free options each company lists on its own pricing page:

- Google Vids: 15-second Gemini Omni clips at no extra cost with a Google account

- Runway Free: 125 one-time credits

- HeyGen Free: 3 watermarked videos a month, up to 1 minute each

- Synthesia Basic: 10 minutes of video a month, with a watermark

- Canva Free: limited access to Premium AI tools, Image to Video included

The Gemini app's free plan doesn't list video generation; it starts on Google AI Plus at $4.99 a month. And Grok's pricing page marks image generation as limited on its free plan but doesn't list video generation there.

For animated videos and edits, two tools are free outright. HyperFrames' framework and skills are open source under Apache 2.0, and rendering on your own computer needs no account; you pay only for the Claude or ChatGPT plan your agent runs on. FilmCraft is free and open source, as above.

## Image to video AI: turning a photo into a clip

Grok Imagine 1.5 is built for this: xAI's instructions are to "give it a starting image, describe the motion, and choose your resolution and duration", and the clip comes back with sound. Kling 4.0 and Kling 4.0 Flash both list image-to-video, and Gemini Omni 1.2 Flash takes character, product or style images and holds faces, packaging and brand colors steady through the clip.

For a free start, Canva's Image to Video is one of the Premium AI tools Canva Free gets limited access to. If you use Seedance 2.0 through Higgsfield, Dreamina or CapCut, Pat's free Seedance prompt builder skill for Claude turns a request like "Turn this product photo into a slow close-up hero shot" into a prompt you can paste in.

## AI avatar videos: HeyGen, Synthesia and InVideo

HeyGen and Synthesia make videos with AI avatars, digital presenters you pick from a library or make from yourself, and InVideo sells credits for AI video models with a timeline editor around them. Pat hasn't tested any of the three, so this section is what their pricing pages say.

HeyGen's free plan makes 3 watermarked videos a month, up to a minute each, and Creator costs $29 a month for 1080p exports without HeyGen branding and a custom avatar with your own voice and likeness. Pro, at $49 a month, adds 4K export, videos up to 30 minutes and Seedance 2.5. HeyGen also has a plugin for ChatGPT.

Synthesia's Basic plan is free with 10 minutes of video a month and a watermark. Starter is $19 a month and Pro is $89 a month, with a cheaper first month on each ($9 and $49).

InVideo's pricing page lists three plans by monthly credits: Starter with 400, Plus with 2,000 and Max with 8,000. Starter gets Seedance 2.0 Fast and Mini, and Seedance 2.5 starts on Plus. The dollar prices only appear once the page loads in a browser, so check them there.

## How do you make AI videos with Claude or ChatGPT?

You describe the video to an AI agent, and a free tool turns what it builds into a video file. With HyperFrames, Claude Code or Codex writes the video as a web page and HyperFrames records it into an MP4, as covered in HyperFrames vs Remotion for animated videos. In Pat's May video on it, he put it this way: "every video in Hyperframes is written as HTML. This is the same code that builds the websites you use every day."

In Claude Code, HyperFrames' README installs it with two commands, claude plugin marketplace add heygen-com/hyperframes and claude plugin install hyperframes@hyperframes; for Codex and other agents it's npx skills add heygen-com/hyperframes. If you'd rather skip the terminal, the HyperFrames Studio app for Mac, Windows and Linux is free with a free HeyGen account, and you direct the edits there without writing code. Remotion, the older code-to-video tool, is free for individuals and companies of up to 3 people, and bigger companies need a paid license.

_Part of a 30-second promo Codex built with HyperFrames for AgentMail, a startup that gives AI agents email inboxes. Pat's verdict was "the branding is dead on." From Pat's HyperFrames video at 7:16._

In the September 11 Astra video he left motion graphics out of his editing skill, saying of the code-to-video repos: "I don't love the motion graphic output from these repos quite yet. It just isn't there." By October 7, 2026, he was rendering motion graphics for his own videos with the HyperFrames skill, as the HyperFrames post shows.

For explainers with generated footage, Pat made a Vox-style explainer from one prompt, with Claude Fable 5.1 doing the research, the script, the art direction and every call to Seedance 2.5. He also published a beginner's guide to product ads made with Seedance 2.0 (listed in the sources below).

_Pat's video on making a Vox-style explainer with Fable 5.1 and Seedance 2.5 from a single prompt._

Inside ChatGPT, video comes through plugins now that Sora is gone: CapCut's plugin builds and edits drafts in the desktop app, HeyGen has a ChatGPT plugin for avatar videos, and Runway has one too (how to set up Runway in ChatGPT).

## What AI video tools cost

The editing setups Pat uses run on apps and AI plans you may already pay for, while generators charge by the second or by credits. Every price below was checked on the company's own page on November 8, 2026.

ToolWhat it's forFree optionPaid

 Claude Opus 5.5 with Pat's Premiere skillEditing your footage in Premiere ProSkill and bridge free on GitHubPremiere Pro US$22.99 a month (annual, billed monthly), plus your Claude plan
 GPT-6 Astra with Pat's Premiere skillEditing your footage in Premiere ProSkill free on GitHubOne full edit used about 15% of Pat's usage on the $100-a-month ChatGPT plan
 FilmCraftEditing, with Claude through its MCP serverFree and open sourceNone
HyperFramesAnimated videos built by Claude Code or CodexFree, open source; Studio app freeYour agent's plan
DescriptEditing by transcriptFree: 60 minutes of media a monthHobbyist $16 yearly or $24 monthly; Creator $24 or $35
Kling 4.0 and 4.0 FlashGenerating clipsNone listed on the API price pageAPI: Flash 0.6 units a second; Kling 4.0 1.0 at 720p, 1.5 at 1080p
 Gemini Omni 1.2 FlashGenerating and chat-editing clips15-second clips in Google Vids at no extra costGoogle AI Plus $4.99 a month; API about $0.07 a second at 720p
 Grok Imagine 1.5Photo to clip, with soundVideo not listed on the free planPlus $8 a month; Super $100 for 1080p, 15-second videos; API $0.14 a second at 720p
RunwayGenerating clips with Gen-4.5, Kling and Seedance125 one-time creditsStandard $15, Pro $35, Max $95 a month ($12, $28, $76 billed yearly)
Adobe FireflyGenerating clipsLimited daily generationsStandard US$9.99, Pro US$19.99, Pro Plus US$49.99, Premium US$199.99 a month
CanvaImage to video and short clips in designsLimited Premium AIPro US$144 a year
HeyGenAvatar videos3 watermarked videos a monthCreator $29, Pro $49 a month
SynthesiaAvatar videos10 minutes a month, watermarkedStarter $19, Pro $89 a month

Pat's own measured cost is the Astra edit: about 15% of the usage on his $100-a-month plan for a 56-minute run that cut about 48 minutes of footage to about 26. No usage figure was logged for the Opus 5.5 setup. For the coding side of the same models, see the best AI for coding, from 21 builds.

## Update November 8, 2026

This page was created. Every price, plan and limit was checked on the companies' own pages that day; Sora is listed as discontinued per OpenAI's help page, and Grok Imagine 1.5's API price is now $0.14 a second at 720p on xAI's price page. New entries go above this one when Pat posts a new video test.

## Questions people ask

What is the best AI video editor?
For editing your own footage, Pat Simmons uses Claude Opus 5.5 in Claude Code with his free Premiere Pro skill and bridge, and it does about 95% of his edit. If you'd rather not pay for Premiere, FilmCraft is a free clone that Claude can edit in through its MCP server.

Is Sora still available?
No. OpenAI shut the Sora app on April 26, 2026 and the Sora API on September 24, 2026, and you can export old Sora content from sora.chatgpt.com/sunset.

Can Gemini make videos for free?
Not in the Gemini app, where video generation starts on Google AI Plus at $4.99 a month. Google Vids makes 15-second Gemini Omni clips at no extra cost with a Google account.

Is Runway free?
Runway has a free plan with 125 one-time credits, which don't refill. Paid plans start at $15 a month.

Is HeyGen free?
Yes, with limits: HeyGen Free makes 3 videos a month, up to a minute each, with a watermark.

Can Canva make AI videos?
Yes. Canva lists Image to Video among its Premium AI tools and "Create high-quality video clips with Canva AI" among its Ultra AI tools, and Canva Pro costs US$144 a year.

Does Adobe Firefly make video?
Yes. Firefly's video generation uses generative credits on any plan, and the US$199.99-a-month Premium plan includes unlimited use of the Firefly Video Model.

Can ChatGPT make videos?
Through plugins. ChatGPT has had no video app of its own since Sora shut down, and CapCut's plugin builds and edits video drafts in the ChatGPT desktop app, and HeyGen has a ChatGPT plugin for avatar videos.

What is text to video AI?
Text to video AI makes a short clip from a written description. You type the scene, the camera move and the sound you want, and a model like Kling 4.0, Gemini Omni 1.2 Flash or Seedance 2.5 generates the clip, in these three cases with sound included.

How long can AI video clips be?
Up to 30 seconds per generation for Kling 4.0 and Seedance 2.5, 20 seconds for Kling 4.0 Flash, and 15 seconds for Gemini Omni 1.2 Flash and Grok Imagine. Avatar tools run longer: HeyGen Pro makes videos up to 30 minutes.

Can AI edit my videos for free?
The tools can be free, though the AI model usually isn't. Pat's Premiere skill and FilmCraft are free on GitHub, and HyperFrames is free to render on your own computer, but each runs on an agent like Claude Code or Codex, which bills to your Claude or ChatGPT plan.

## Sources

- Kling AI: API pricing (checked November 8, 2026)

- Kling AI: Kling 4.0 and Kling 4.0 Flash release

- Google: Gemini Omni 1.2 Flash

- Google AI Studio: Gemini API pricing (checked November 8, 2026)

- Google: Gemini subscriptions (checked November 8, 2026)

- SpaceX Super Intelligence: Grok Imagine Video 1.5

- SpaceX Super Intelligence: API pricing (checked November 8, 2026)

- SpaceX Super Intelligence: Grok plans (checked November 8, 2026)

- ByteDance Seed: Seedance 2.5

- OpenAI Help: What to know about the Sora discontinuation

- Runway: Pricing (checked November 8, 2026)

- Adobe: Firefly plans (checked November 8, 2026)

- Adobe: Premiere plans (checked November 8, 2026)

- Canva: Pricing (checked November 8, 2026)

- HeyGen: Pricing (checked November 8, 2026)

- Synthesia: Pricing (checked November 8, 2026)

- InVideo: Pricing (checked November 8, 2026)

- Descript: Pricing (checked November 8, 2026)

- CapCut: CapCut × ChatGPT

- HyperFrames docs: FAQ

- HyperFrames docs: Install the Studio app

- HyperFrames on GitHub

- Remotion: License and pricing

- FilmCraft on GitHub

- per-simmons/opus-premiere-editing on GitHub

- per-simmons/astra-premiere-editing on GitHub

- Pat Simmons on YouTube: Codex + Hyperframes = INCREDIBLE Motion Graphics

- Pat Simmons on YouTube: Fable 5.1 + Seedance 2.5 Just Unlocked Vox-Style Motion Graphics

- Pat Simmons on YouTube: How to Make Viral Product Ads with Seedance 2.0 (Beginner's Guide)