June 2026 Update: Brandmachine Agent gets sharper

Benjamin Nassler•June 15, 2026

A couple of weeks ago we published the Autonomous Photo Studio framing: humans bring creative direction, agents handle everything that can be deduced from it. The fully autonomous side of that is currently in research preview with a small group of brands.

Alongside it, we've been shipping smaller, step-by-step improvements to make Brandmachine Agent itself easier to find and more useful across the product. Here's what landed this week, plus a quality-score feature worth a section of its own.

Brandmachine Agent, easier to find

Brandmachine Agent has always been there, sitting alongside your work. Powerful but easy to miss. The most common feedback we've heard is some version of: "I didn't know the agent could do that."

This release tackles that head-on.

Tools surfaced per view. On every screen, the agent now shows what it can do for you there. The actions are visible at a glance, so you can see whether the agent can help before you ask. The discoverability problem goes away.

Ask why something didn't work. If a generated image didn't come out the way you wanted, you can just ask: "Why did this go wrong?" The agent reads the image directly, compares it to what you asked for, and tells you what happened. The image reading is powered by Gemini 3.5 Flash, which the agent now runs on, and it generalizes: any time the agent needs to look at an image you're working with, it can.

A few smaller things came with it too. Replies are shorter and less repetitive. Long chats nudge you to start fresh so the context stays clean. You can leave feedback without leaving the conversation.

You direct. The agent handles.

A quality score on every image

Every generated image now gets an automated quality verdict: a score and a short written comment. In the studio, each output tile carries a color-coded indicator. Anything below 90% is flagged at rest. Hover for the score and the comment.

At catalog volume the bottleneck moves. Generating images is solved. The hard part is finding the weak ones fast among hundreds of strong ones. The studio now grades its own work and lets you zero in on what needs attention.

Also shipped

A few smaller things that make the rest of the work less surprising.

The agent got better at catching its own mistakes:

When you change the lighting, mood, or setting of a campaign image, it works out whether that attribute belongs to the shooting location or to the image's own styling, and asks where to apply it.
When you report that a design looks wrong, it diagnoses the mismatch against the brief and proposes the right fix instead of guessing which image to redo.
Paint edits that change the silhouette (a new hemline, neckline, sleeve length) now regenerate both the front and back so the views stay consistent. Small local marks like a button still only touch the view you edited.

What you see is what you ship:

Finished images now download as either a lossless PNG or a smaller sRGB JPEG with the right color profile. Pick the format that fits where the image is going.
High-res and paid downloads hand you the final processed image (your edit if you made one, otherwise the brand-color-corrected version), never the raw AI output.
Brandmachine Color Correct (the auto-snap-to-your-exact-hex feature from May) just got a scaling upgrade. Much more reliable on large product shots.

Easier to start, easier to bring teammates on:

Empty screens show a guided checklist of the real prerequisites, with shortcuts into each step.
A new Learn button on each feature opens a short guided video tour that ends by dropping you into the create flow.

Smaller wins

The app is installable to your home screen as a PWA.
Rename a product input by clicking its title in Product Studio. The change updates across all of that product's images.
The campaign-image inspector shows the full product list instead of capping at the first three.
Shot Planner is out of beta.

Try it

Open the agent panel and ask it something. The quality scores show up automatically on anything you generate. Tell us what breaks. The fastest way we get the next round of autonomy right is by hearing where the agent still falls short for you.

Back to Blog