Assistant

The Assistant helps you direct the 3D scene in plain language. Render photos and videos from that scene, then optionally use AI finishing to extend their look. AI finishing is a separate result: your 3D scene and original render stay available.

Open the Assistant

The Assistant lives at the bottom of the left sidebar, under the scene sections.

• Click the Assistant row to expand it.
• Or click Edit with AI on the viewport, in Photo or Video mode. That opens the Assistant on the current angle or clip.

The Assistant and the scene sections share the sidebar. Drag the divider above the Assistant's header to change the split, or collapse Scene to give the Assistant the whole column — one of the two is always in reach. Click the Assistant header again (or press Escape) to collapse it back to a single row; work you started keeps running and reports on that row until you come back.

What you can ask

Change the scene. Anything the 3D scene can express: the look ("add a little film grain", "warmer, softer light"), the environment ("switch to a dark backdrop"), the product ("select the blue colourway", "turn the shoe to face the camera"), props ("put a few rocks around it"), saved angles and clips. The change lands in the live scene, so it holds across every angle and every render.

Generate an image or video edit. Anything the scene cannot contain: a beach at sunset, a hand holding the product, lifestyle context, photographic re-styling of the frame, or "make a video of it spinning". The Assistant hands these to AI editing and shows the result in the conversation.

Ask. "What lights are in this scene?", "which backdrop is this?" The Assistant reads the scene and answers.

You do not need to say which of these you mean. If the request could reasonably be either — "make it darker" could be the backdrop or the whole frame — the Assistant asks, with a couple of options to tap.

How it answers

Each request becomes a turn in the conversation. Depending on what you asked, the reply is one of:

• A scene change — one sentence saying what changed. Every scene change is a saved state; hover an earlier turn and choose Use this to go back to it. The current state is highlighted.
• An image or video hand-off — a short reason ("the scene can't add a beach…"), the exact prompt the Assistant will use, and a Generate image or Generate video button. When you plainly asked for an image or video, it starts right away without the button. Options on the card let you keep the product's exact position before generating.
• A question — with option chips. Tap one, or type your answer.
• Suggestions — a few follow-up ideas ("Warmer", "More contrast") that fill the composer when tapped.

When the next step is something you do by hand in Studio — building a product animation, deleting an angle, rendering — the reply names the place as an underlined link: "…continue in the Animations tab of the clip editor." Click it and Studio opens that panel for you, switching between Photo and Video mode if it has to.

While a request runs, the row above the composer says what the Assistant is doing right now — reading the lights, placing props, checking the frame — alongside the elapsed time, and Stop ends the run.

Image and video results

A generated result appears in the conversation as a card:

• Open shows it in the Studio viewport. Compare with the source render or choose Save to Library when you like the result.
• Back to 3D returns to the live scene and clears the AI-result target. The result stays in this session so you can reopen it.
• Save keeps it in your render library. Until then it stays in this session as a preview.
• Refine makes it the target of your next request. The composer shows Editing · Image result; describe the change ("softer shadows", "more sky") and the Assistant edits that result rather than the live view. Clear the chip to go back to editing the scene.

Image edits use the current angle in Photo mode and the selected clip in Video mode. When you create an AI edit from a changed angle, Glossi saves the angle before rendering it. Refining an existing AI image continues from that image; it does not save camera changes. If a hand-off wants the other kind — a video from a photo angle, say — the card tells you to switch modes first.

Reuse an edit on future photo renders

Explore and refine in the angle’s AI editor. On a result you want to repeat, choose Use on future renders…. Review What should change in each new render?, then save the instructions for that angle.

Write instructions that make sense from a fresh 3D render. A refinement such as “make it warmer” may depend on earlier edits, so describe the complete result you want. Saving instructions does not promise an identical image.

Open On future renders to change the instructions, switch an edit off, or remove it. Test on a new render saves pending camera changes to the selected angle, captures it again, and applies that edit. Each saved edit starts from the original render independently; multiple saved edits create separate results.

Enabled edits run after photo rendering from Export. The export panel identifies which selected angles include saved edits. Choose Background for these renders. Here becomes available again when the selected angles have no enabled saved edits. Your original renders stay available. Downloading a selected AI result downloads that result; it does not run the saved instructions again.

Templates saved in your workspace carry these instructions and their on/off state with the saved angles. Applying a template does not generate an image. Saved AI edits currently apply to photos; video edits remain manual.

Writing a good request

• Say what should change and what should stay. "Place the shoe on wet sand at golden hour. Keep the shoe's position, angle and logo exactly as they are."
• One change per request is easier to judge and easier to undo.
• For scene changes, name what you see: "the key light", "the backdrop", "the blue variant".
• Or point instead of naming: Choose what to edit above the composer lists your materials, parts, lights, props, and angles. Hovering a material or part highlights it in the viewport; picking one scopes your next request to exactly that, shown as an Editing chip you can dismiss.
• Attach a photo with Match a photo's look when a reference says it better than words.
• Or say it: the microphone above the composer dictates into the field. Your words appear as you speak and stay editable; press Enter or the send arrow when you are happy with them. Dictation uses your browser's own speech recognition, so it is available in Chrome, Edge and Safari, and the browser asks for microphone access the first time.

What the Assistant does not do

• Render or export. Use Export in the toolbar for final images and video.
• Delete things. Removing lights, props, angles or clips stays a manual action.
• Guarantee product fidelity in AI edits. An image edit can drift from the source product. Compare before you use it; when exactness matters, make the change in the scene and render again.

Related

• Direct a scene in Studio
• Edit a render with AI — the same AI editing, started from a finished render
• Lighting · Backdrops · Photo mode · Video mode