Anyone who edits video runs into this all the time: the boom mic dipping into the frame, a logo on a T-shirt the client doesn't want, a sign that steals attention in the best take. Until now there were two ways out: reshoot, or spend hours in After Effects tracking and painting frame by frame. That's why, among the Adobe MAX announcements on November 10, we stopped on Generative Remove in Premiere: you draw a mask around the object, track it through the clip and the AI fills the hole.
Our take? It's the kind of feature that can save a delivery, but for now it's a beta with tight limits and a cost you need to work out properly, because you pay per second of video. We haven't tried it on a real job yet, so here you'll find what Adobe states, the credit math done with the official numbers and, as always, the alternatives that run locally in ComfyUI.
| What it is | A Premiere tool that erases an object from a video clip and rebuilds the background with AI |
| Who made it | Adobe, with the Firefly Video model; in beta since November 4, 2026 |
| VRAM required | none locally: processing happens in Adobe's cloud and you need a connection |
| Does it run on an RTX 3060 12 GB? | yes, because the computation doesn't use your graphics card; your PC just needs to handle Premiere |
| License | beta feature included in any subscription that includes Premiere |
| Commercial use from Italy | yes, with an active Creative Cloud subscription |
| Where to try it | Premiere (beta), from the Creative Cloud app: Remove button in the mask controls under the Program Monitor |
| Free or paid | generative credits: 40 credits for every second of processed video |
What is Generative Remove in Premiere and how does it work?
It's a video inpainting feature (rebuilding an erased area of the image, frame after frame) built into the Premiere timeline. In the official beta announcement Adobe explains that the mask defines the area to fill and that everything outside the mask stays identical to the original.
We liked two details. First: the work is non-destructive, meaning the original footage isn't touched, and from the AI badge on the clip you can regenerate, revert or rate the result. Second: removal runs in the background, so you can keep editing in the meantime. You can also select several masks in the same generation, for example the mic and its shadow.
Generative Remove is part of the package shown at Adobe MAX along with the AI Assistant (beta), Color Mode now out of beta and the Generative Media Tool, which generates video, sound effects, soundscapes and music right in the timeline with Firefly or with models from Google, Kling, Runway and Luma, as reported on Adobe's blog. If you're interested in the "assistant" side, we covered it in our article on what the After Effects AI Assistant really does.
What are the limits of the beta?
There are three limits, and you should read them before opening your project: clips up to 10 seconds, resolution from 480p to 1080p SDR and frame rates up to 30 frames per second. In practice: no 4K, no HDR, no 50 or 60 fps slow motion, and for a long shot you have to cut the clip into pieces.
On those pieces Adobe gives an important tip: it's better to generate the removal across the whole clip in a single pass, because segments processed separately may not match at the cut. You also need an internet connection, because the footage is processed in the cloud: keep that in mind if you work with confidential material and your client contract forbids uploading it to external services.
How many credits does Generative Remove in Premiere cost?
It costs 40 generative credits for every second of processed video: an Adobe Community Manager wrote this on November 10 in the announcement thread, and the figure is reported by AI Weekly. A 10-second clip, the maximum allowed, therefore comes to 400 credits. Premiere shows the cost before you confirm anyway, and it's always worth checking.
How many credits you have depends on your plan. With Creative Cloud Pro you get up to 4,000 credits a month for premium features, which is about 100 seconds of removal. For single-app subscriptions the numbers going around don't agree (they change depending on when you subscribed), so our advice is to check the credit counter in your account before relying on the feature for a delivery. Remember that credits don't roll over from one month to the next.
Our verdict: to remove a mic from three seconds of interview, the cost is more than reasonable. To clean up a whole video with lots of objects, though, the credits run out quickly, and that's where a local alternative starts to make sense.
Can you remove objects from video for free in ComfyUI?
Yes, with some compromises on video memory. There are two native options in ComfyUI:
- Netflix VOID: the open model (Apache 2.0 license, commercial use allowed) that removes not just the object but also its consequences, like shadows and reflections. ComfyUI has the official "VOID: Video Inpainting" template, and the ComfyUI guide explains the steps: you type the name of the object to remove and SAM3 (a model that creates the mask from a word) cuts it out by itself. The limit is VRAM: the official repository asks for a 40 GB card or more, so on a 12 GB card it's not a practical route.
- Wan 2.1 VACE: the video inpainting template designed precisely to remove or replace objects. The 1.3B version works at 480p and is the one suited to a card like ours; the 14B reaches 720p but needs much more memory.
Honesty first: today, on 1080p quality and the convenience of working inside the timeline, Premiere is ahead. Local wins on privacy (your footage never leaves your PC), on cost per second and on the freedom to try ten times without watching the counter. For photos the local route is already very mature, as we explained in our article on how to remove objects from photos in ComfyUI in 6 steps. To understand how much memory video models need, you'll find everything in our guide to AI video models in ComfyUI, and if you're thinking about changing cards our guide to graphics cards for AI starts right from the RTX 3060 12 GB.
Try it now
If you have Premiere, here's how to run your first test while spending as little as possible:
- Install Premiere (beta) from the Creative Cloud app, in the beta apps section.
- Pick a short clip, 2–3 seconds, in 1080p at 25 fps: it costs 120 credits at most.
- Draw the mask around the object and track it through the whole clip.
- Scrub through the clip and check that the mask also covers shadows and reflections: anything left outside stays in the image.
- Press Remove under the Program Monitor, read the credit cost and confirm.
- Watch the result at normal speed and frame by frame: if it doesn't work, the AI badge takes you back to the original.
The most useful trick is the same as in compositing software: a clean mask that hugs the object gives better results than a loose, rough one. If you're new to nodes and want to try the free route, start from our beginner's guide to ComfyUI.
Other news
- Plenio 2.0 (November 14): the system for making songs in ComfyUI with YuE2 or MiniMax Music 3 now has a guided setup that reads your graphics card and downloads only the right models, plus an MCP server to request a track from Claude. It's on the project's GitHub.
- Geiru Context Suite (November 12): GPL-3.0 ComfyUI nodes that keep colors and textures consistent in long MiniMax H3 chains, up to about two minutes in one pass. GitHub page.
- ComfyUI 0.45.0 (November 10): support for MiniMax-H3-Image, WeVisOne-VL and RelayVSR, plus new nodes for Kling 4.0 Flash and Seedream 5.5. Full notes in the GitHub releases.
Frequently asked questions
Is Generative Remove already in the regular version of Premiere?
No. As of November 15, 2026 it's only available in Premiere (beta), which is installed separately from the Creative Cloud app and can live alongside the regular version. Adobe hasn't given a date for the stable release yet, so for now it's best used on projects where a problem won't block the delivery.
Does it work with 4K clips?
No, the beta accepts footage from 480p to 1080p SDR, up to 30 frames per second and up to 10 seconds per clip. If you shoot in 4K you can work on a proxy or a downscaled version, but the result will stay in 1080p: for 4K deliveries it's better to wait for the final version.
Do you need a powerful graphics card?
No, because the removal is computed on Adobe's servers and not on your GPU. You do need a stable internet connection and a subscription with enough generative credits. The graphics card only matters for how smoothly Premiere runs and if you want to use the local alternatives in ComfyUI.
What's the free alternative to Generative Remove?
ComfyUI has two official templates: Wan 2.1 VACE for video inpainting, which in its 1.3B version works at 480p even on 12 GB cards, and Netflix VOID, more complete because it also removes shadows and reflections, but designed for 40 GB cards. They're free and your footage stays on your computer.
If you want to see how we work with AI locally, on our YouTube channel @tutorial3d you'll find videos with free workflows to download.










