AI Edit Mode

The mode for work you do together with AI. Let your own Codex or Claude Code operate the editor directly, or turn the 3D scene in the viewport into styled images, an ad look, an ad video and an ad page. The right sidebar has five tabs.

• MCP: connect your Codex · Claude Code to the editor → MCP tab
• Image: generate styled product-page imagery from the view → Image tab
• Ad look: AI looks at the product and suggests background, lighting and framing → Ad look tab
• Ad video: plans camera shots and renders a 1080p mp4 → Ad video tab
• Ad page: drafts the share page as an ad landing page → Ad page tab

Getting started

Choose AI from the editor mode switcher. The viewport is identical to normal edit mode — camera controls, gizmo, floating toolbar, placed items and the bottom timeline all stay. The image and ad tabs send exactly what you see on screen to the AI, so frame the shot first.

• The Image and Ad page tabs are available for saved account items only; a local item that has never been saved gains access after its first save
• Image generation costs 2 EvoCredit per image. Ad look, ad video and ad page cost no EvoCredit and have a request limit
• The Tab cycle order is normal → splat → mesh → style → ai → image viewer

AI box

The box at the very bottom of the editor. Write what you want to make or change, and EVOVA runs the AI for you, working in the editor right away. No Codex or Claude subscription needed.

• Quality: choose 초실감 3D (photoreal), 고품질 3D (high quality) or 빠른 스케치 (quick sketch) from the box's menu. Photoreal takes the most care and the most EvoCredit
• Reference images and video: attach them with + on the left of the box. Large videos are made smaller on your device first
• Improve the result: the button above the box. The AI looks over the current model and fixes what looks off
• Cost: EvoCredit is taken for the work done (1 EvoCredit = 100 KRW). The run stops when your credit runs out, and a failed run is refunded
• When a run ends, tell us how it went with the thumbs up or down. It helps us make better results
• The box can't be used while MCP is on

MCP tab

Lets the Codex or Claude Code you already subscribe to operate this editor directly. Ask the agent in plain words and it looks at the scene, draws splats, checks the result and keeps refining. AI usage comes out of your own Codex · Claude subscription.

Connecting (once)

1. Run the command for your agent in a terminal
   • Codex: codex mcp add evova --url https://app.evova.ai/api/mcpCopy
   • Claude Code: claude mcp add --transport http evova https://app.evova.ai/api/mcpCopy
2. Press Allow on the approval page that opens in the browser
3. Restart the agent if it was already running. A running agent does not pick up a connection that was just added

You only connect once. Reloading the editor does not require connecting again, and approved agents are listed in the MCP tab, where you can disconnect them.

Asking (every time)

1. Open the editor and turn on Enable MCP in the MCP tab. An agent cannot touch a tab whose switch is off. "Agents can operate this tab." means it is ready
2. Start the agent and say what you want. E.g. "Use the evova tools to make a teddy bear in 3D in the EVOVA editor I have open"
3. The result appears in the editor. A border around the window shows while the agent is working

On the first screen of an empty editor, Create with Codex · Create with Claude show the connect command and the sentence to paste on one screen.

Keep this tab in view while the agent works. Work stops if you minimise the tab, switch to another tab or close it.

What an agent can do

• Look: capture the view, capture several sides in one picture, move the camera, and view the form alone (clay · depth · normal · outline)
• Draw: place splats like brush strokes, or draw surfaces and solids from its own script. It can also read and change existing points one part at a time
• Reference images: compare a photo with the scene, find where they differ most and fix it, and project a photo onto points
• Edit: run the editor's existing tools (selection, transform, color grading, style and more)
• Server jobs: train on the server from pictures
• Save: save the scene as an item

Settings

• New scene format: the format of scenes the agent starts. The default, 4.0, has opacity that depends on the viewing direction
• Agent may spend EvoCredit: off by default. While off, tools that cost credit do not run. No tool costs credit at the moment
• Reference images: upload pictures for the agent to look at in this tab
• Recent activity: the tools the agent ran and the requests that were refused are recorded

If the same item is open in two tabs, the tab opened later takes over the connection. With different items open in several tabs, the agent picks which tab to work in.

Image tab

Basic flow

1. Set the item's angle and size in the viewport. What you see is exactly what the AI receives
2. (Optional) Mark your intent on the view with Sketch → see the Sketch section below
3. Write the scene you want in the Prompt. E.g. "on a wooden table in warm afternoon light, a blurred café window in the background"
4. Choose the aspect ratio (16:9 · 9:16 · 1:1) and sample count (1–4), and attach a reference image if needed
5. Press Generate Image. The job is queued on the server and its status in the gallery moves from waiting → processing → done. You can keep working in the editor meanwhile

EvoCredit balance and generation cost

• Above the Generate button, check your balance and the total for this request. Generating 1, 2, 3, or 4 images costs 2, 4, 6, or 8 EvoCredit, respectively
• Your balance also appears in the editor header. Click it to open top-up or transaction history in a new tab
• Organization accounts use the shared organization balance; personal accounts use their personal balance
• Credits are deducted when the job is submitted. Failed or cancelled jobs receive a full automatic refund. If only part of a batch succeeds, the entire job fails and its full charge is refunded
• Editing a result to generate one new image also costs 2 EvoCredit. Retrying a refunded job charges again for the requested number of images
• Generation is disabled when the balance is insufficient. If a request loses its response, use Check generation request to check the original submission

Sketch — marking intent on the view

A way to communicate in pictures what is hard to put into words. You can mark positions and regions such as an area to remove or where a prop should go, or draw the thing itself — the shape or placement you want. Draw the outline of a plant pot next to the item and write "small plant pot" as the color meaning, and the result shows a pot right there. Turn on the Sketch toggle to draw over the viewport.

• Pen (Shift+P) · Eraser · Clear (Shift+U, removes everything) with small · medium · large brushes
• Seven colors (red · green · blue · yellow · magenta · cyan · white). Each color can carry a color meaning: write "area to remove" for red and "where the plant goes" for green, and the AI is told "the red line means area to remove", and so on
• At generation time the AI receives the clean view without the sketch as the source and, if you drew anything, the sketched view as an extra reference. The lines themselves don't end up in the result — only the intent does
• Sketches are AI input only and are not saved with the item. For drawings that should stay in the viewer, use the Draw tab

Writing a good prompt

• Be concrete in this order: scene (where the item sits) · lighting (time of day, light source) · mood (tone, style) · camera (front, 45°, close-up). Results are most consistent that way
• The item's shape and color come from the view, so you don't need to describe them. Focus on the background and surroundings
• Reference images are style guides. Upload a photo with the look you want, or pick one from the gallery, and the colors and composition follow it
• Content that violates policy shows a notice instead of a result; change the prompt or image and try again

Gallery

Results generated for this item accumulate, newest first. Each result offers:

• Download — save as an image file. With several samples, page through with previous · next
• Edit — write an edit prompt on top of the result ("brighter background", "make the table marble") and regenerate
• Retry — run a failed job again with the same settings
• Delete — remove the result
• Double-click a result card's prompt to edit the text; it saves automatically when you click away

Video mode (generating motion video from a first and last frame, or expanding a result image into video) shows buttons on screen but is still in preparation and does not run yet.

Ad look tab

AI analyses the product and suggests an ad look (background, lighting, shadow, camera motion) and framing (product placement, camera position).

1. (Optional) Pick a use: Product page (the product leads, restrained styling) · Brand landing (mood and cinematic styling) · Social ad (eye-catching motion and bold styling)
2. (Optional) Describe the mood you want. E.g. "refined and calm"
3. Press Suggest with AI and the suggestion cards appear
4. Under Apply (background · lighting, cinematic, camera motion, interaction, viewer UI, camera framing, product placement) apply only what you want

• Every setting stays adjustable after applying, and undo reverts it
• The look (background · lighting) is not shown in the AI mode view. Check it with View in style mode
• The look is saved with Save style; camera and object positions are saved with the top Save (Ctrl+S). Save everything saves both
• The same product can be analysed again after one minute

→ Each part of a look is explained in Style Edit Mode.

Ad video tab

AI plans camera shots that suit the product and renders a 1080p mp4 ad video in the current style. You do not need an ad look first.

1. Pick the aspect ratio (9:16 vertical · 1:1 square · 16:9 landscape) and the copy language, then press Plan the video
2. Review and edit the shots (dolly in, orbit, side sweep and more) and the copy (label, headline, sub line, end card)
3. If you want, Insert into the timeline puts the camera keyframes on the timeline so you can refine them (it replaces the camera keyframes already there). For the camera path choose AI shot list or Edited timeline
4. Press Render ad video, then Download it or Add to share page

• Text: choose a typography template (editorial · minimal centre · bold stack · luxury · kinetic editorial), a reveal motion (line reveal · fade) and an end card (end card + button · dimmed · none)
• Music: add an mp3 · wav · m4a file (up to 30 MB) and turn on Put cuts and text on the beat. The file is not uploaded to the server. Only use music you have the rights to
• Quality: Standard (fastest) · High (2× supersampling, about 3–4× the time) · Cinematic (supersampling + motion blur, 10× the time or more)
• The video is rendered in the browser. Use Chrome or Edge
• The share page accepts videos up to 10 MB. If yours is larger, shorten it or render at standard quality

Ad page tab

AI drafts this product's share page as an ad landing page (title, body, details, design). Available for saved items only.

1. Pick the page language and, if you like, describe the mood. E.g. "minimal and refined"
2. Press Draft an ad page with AI
3. Open on the share page to edit shows the draft in a new tab. Review it, edit it and save. It only reaches the share page once saved

Links and the ad video you already added are kept.

→ Publishing and sharing › Share page