Adobe MAX 2026: a Premiere agent, Firefly as a studio, and a local MCP

Adobe MAX 2026: a Premiere agent, Firefly as a studio, and a local MCP

Update (Nov. 10): removing objects got easier in Photoshop, Lightroom, and Premiere

Photoshop, Lightroom, and Premiere got better tools for erasing whatever is left over in a scene. The Verge rounded up the announcements. Below is what each one does, what is already out, and what is still in beta, checked against Adobe's own pages.

Premiere: Generative Remove, in beta. You draw a mask around the object, track the mask through the shot, and click Remove. According to the beta announcement in the Adobe community, from November 4, only the masked area is rebuilt and the rest of the frame stays untouched. Shadows and reflections only disappear if you include both in the mask. The limits matter: files from 480p to 1080p, no HDR, up to 30 frames per second, and 10 seconds at most. Processing happens in the cloud and uses generative credits. On the same page, a reply from the Adobe team mentions 40 credits per second, which means 400 credits for a 10-second clip.

Photoshop: Remove takes the shadow and reflection with it. Before, to erase a person from a photo, you selected the person, then the shadow, then the reflection in the puddle. The Related Elements feature, which had been in beta since October according to the beta page in the Adobe community, finds the object's traces on its own (shadow, reflection, splash, smoke) and erases everything at once. According to DPReview, Adobe showed an improved Remove at MAX that also catches water ripples, dust, and fog. The Photoshop release notes for version 28.0, from November, list the feature. In the beta it cost generative credits, and Photoshop Essentials pointed out a flaw worth knowing before you use it: because the whole image is regenerated, the result can come out softer. The workaround is a layer mask that shows the original photo outside the erased area.

Lightroom: shadows and reflections join the selection. The Lightroom release notes for version 10.0, from November, say that when you paint over an object with the detection tool, Lightroom now selects its shadow and reflection too. TechCrunch and DPReview also mention expanded spot removal in Lightroom. Removing reflections from glass, windows, and water with less blur shows up in the Lightroom Classic 16.0 notes.

What this changes for people who post

For photos and covers, it changes a lot. A visible cable in a thumbnail, people walking behind you, and a window reflection on your face are the most common flaws for anyone who films at home, and now they go away in one click instead of five selections.

For video, the 10-second limit defines the use. Generative Remove is for the short shot that got ruined: the lavalier mic that fell into the frame, someone who crossed the background of your vlog, a competing brand's logo on a T-shirt during a sponsored post. It is not for cleaning up an hour-long podcast. And 400 credits for a 10-second clip is expensive enough that you should ask yourself whether it is worth just cutting the shot.

One caveat I would make clear: a removal tool does not change who owns the content. Erasing a watermark or credit from someone else's video is still using someone else's content, with or without the mark on screen.

Adobe did not share availability by country for these features, so there is still no official confirmation for the Brazilian market. If you subscribe to Creative Cloud, update Photoshop and Lightroom and install Premiere (beta) from the Creative Cloud app to see whether they already show up in your account.

Adobe opened MAX 2026 in Miami on Tuesday (Nov. 10) by spreading AI agents across Premiere, Photoshop, After Effects, and Firefly, and announced a local MCP so assistants like Claude, ChatGPT, and Gemini can work directly on project files. For anyone editing short-form video, the agent matters less than three practical things: removing objects from video in Premiere, a mobile Premiere that cuts long video into Shorts, and an open door to automating editing from outside the app.

Almost everything is still beta, part of it is closed beta, and Adobe said nothing about pricing or countries. Below, I separate what you can test from what is a promise.

What changed in Premiere?

Premiere got an AI assistant in beta that runs multi-step tasks from a text request. According to IT Brief, it applies an effect to every clip in a sequence, tracks a masked object frame by frame, and adjusts color and exposure across the whole sequence at once.

Also arriving:

• Generative Remove (beta): erases an object from the video. TechCrunch describes it as object removal in video. The limits and cost are in the update at the top of this post.
• Color Mode: the color grading environment Adobe introduced in April shows up at MAX as Premiere's new color system for professionals.
• Audio: Enhance Audio, Separate Crosstalk (separates voices speaking at the same time), and Separate Sound Source, plus the Generative Media Tool, which generates video, sound effects, and music inside the timeline.

Separate Crosstalk is what I would test first. A podcast with two mics bleeding into each other is the most common problem for anyone cutting conversations, and fixing it inside the editor saves a trip to another program.

Will mobile Premiere make cuts on its own?

It will try, but only in closed beta. MakeUseOf reports on Quick Cuts: you pick footage from your camera roll, describe the video you want, and get an editable first cut. The app's assistant also turns a long video into several Shorts or builds highlights from an event.

On iPhone, the app will offer hundreds of thousands of licensed songs from Universal and Warner, according to IT Brief. The catch is in the fine print: the music is cleared for personal, non-commercial video. Sponsored posts, branded content, and monetized video still depend on Adobe Stock Audio. For anyone who lives off sponsored posts, that changes little.

There is no date for the beta to open and no confirmation for Brazil.

What about Firefly?

Firefly stopped being just a generator and became a studio with an agent. TechCrunch lists an "always-on" agent that gathers context from your work, partial image editing (change one element without touching the rest), video assembly from clips, images, and sound effects, and one-step Quick Actions, like removing a background or generating a video thumbnail.

The always-on agent is in preview. Firefly Builder, which turns text into a website or presentation prototype, is in closed beta. Photoshop gets Relight and sliders for prompt-generated styles.

On the model side, Adobe already offered models from Google, Runway, ElevenLabs, Black Forest Labs, OpenAI, Kling, and ByteDance. With the acquisition of Topaz Labs, it adds Wonder 4, for upscaling and noise removal, and ImageHDR, which converts a regular photo to HDR.

Why is the MCP the news that will last?

Because it takes editing out of the app. MCP (Model Context Protocol) is the standard that connects an AI assistant to a tool. Adobe will release a local MCP for Photoshop, Illustrator, Premiere, and After Effects, and a third-party agent will be able to edit the native file on your computer, with layers, vectors, bins, and sequences staying editable.

Adobe's examples cited by MakeUseOf: removing a logo from several campaign files, or pulling a brief from Google Docs and sending the result through Slack. The closed beta is expected in early 2027, and the company has not confirmed which agents will be supported.

Vice president Deepa Subramaniam told TechCrunch that Adobe is still deciding what to expose and has "no need or desire" to limit what the MCP can reach. That is a strong statement from a company that has always kept users inside Creative Cloud. I read it as an admission that repetitive work is going to be asked of a chat assistant, not clicked through a menu.

This move is not only Adobe's. Edits on Mac also shipped with MCP, and I have already shown here how to automate clipping with an AI agent via MCP today, without waiting for 2027.

What changes for you

It depends on where you edit.

• You already subscribe to Creative Cloud and edit in Premiere: update and look for the assistant and Generative Remove in the betas. Test on a throwaway project, not on client work.
• You cut podcasts and livestreams: Separate Crosstalk may save audio that goes in the trash today. Mobile Premiere cutting into Shorts is the feature closest to your work, but it is closed.
• You do not subscribe to Adobe: nothing here justifies subscribing today. Almost everything is beta, there is no new pricing, and the automatic cutting part is not open.

In April, when Adobe launched the Firefly AI Assistant, I wrote that a general-purpose agent does not replace a dedicated clipping tool. Quick Cuts edges into that territory, but the gap remains: finding the stretch that holds attention in a three-hour livestream, captioning it, and publishing to several social networks is a different job. On Cut.Pro, that whole flow (AI clipping, captions, scheduling on TikTok, Instagram, Facebook, Threads, X, LinkedIn, Bluesky, Pinterest, Telegram, Mastodon, Discord, Dailymotion, Tumblr, Slack, Car.Media, and YouTube) already runs, and through our MCP you can ask Claude or ChatGPT to cut and schedule without opening any editor.

What to do now

If you use Premiere, turn on the betas this week and measure one thing only: how much time the assistant saves on a task you repeat every week, like applying the same color correction to 20 clips. If it does not save time, turn it off.

If you cut long video, do not switch workflows because of a closed beta. Follow Quick Cuts when it opens and compare it with what you use today, using the same criteria as the comparison of AI tools for podcast clips.

And if you already talk to Claude or ChatGPT every day, it is worth connecting an MCP now to some tool that already offers one. When Adobe's opens up, you will already know what you can delegate and what needs your eye.