VoxAI Agent
Chat to build MagicaVoxel .vox models from a description, a reference PNG, or both. Preview in real time, iterate in conversation, and download via View Voxel.
What is Agent?
Agent is a chat workspace at /agent for all signed-in users. Describe what you want—or attach a PNG—and Agent builds a MagicaVoxel-native model you can edit, polish, and save to your library.
Best for: quick blocky assets, iterative edits, and MagicaVoxel workflows. For complex or high-detail subjects (characters, high-resolution objects), use Image-to-3D with VoxAI V3 on the 3D Voxel page.
Getting There
Header link
Click Agent in the app header when signed in.
From My Images
Open in Agent from any image group. When available, your group prompt is pre-filled in the chat.
From View Voxel
Agent-created models show an Edit in Agent option so you can continue editing.
Prompting Basics
Text only
Name a subject and optional grid size:
minecraft pig, 16 cubedPNG only
Attach a reference image with the Attach image button. Agent builds from the picture.
PNG + text
Combine a reference image with text for size or style notes (e.g. grid size, color tweaks).
Edit vs replace
Ask Agent to change the current model ("add a hat"), or name a new subject to replace what's on screen ("make this a chicken instead"). You don't need to click Start over first.
Settings
Open the gear icon in the Agent toolbar to adjust modeling style, thinking effort, and action model.
| Setting | Options | Impact |
|---|---|---|
| Modeling style | Minecraft (default) / Voxel | Minecraft = chunky boxes with cuts and details. Voxel = finer cells traced from the reference photo with per-cell color. |
| Thinking effort | Low, Medium, High, Extra high, Max. Default is Low. | Higher effort improves reasoning but uses more credits and takes longer. |
| Action model | Grok 4.6, Grok 4.6 (fast), GLM 5.3 Flash, Kimi K3, Step 3.7 Flash, DeepSeek V4 Flash Vision (exp) | Affects speed, vision quality, and credit cost per action. |
While Agent Works
Agent shows status labels so you know what it's doing:
- Thinking… — classifying your request
- Building the model… — creating a new model
- Editing the model… — applying changes (viewer shows a blur overlay)
- Polishing the model… — refining details (viewer shows a blur overlay)
- Answering… — responding to questions about the model or VoxAI
A spinner appears during thinking and building. Edit and polish steps blur the 3D preview while keeping the model visible underneath.
Credits & Saving
| Action | Credits | Notes |
|---|---|---|
| Chat (build, edit, polish, Q&A) | Variable | Typically at least 1 credit per action. Cost scales with model and thinking effort. You need a balance above 0 to start. Exact debit is shown in-app per action. |
| Save as new | 2 | Creates a new library asset with a name (max 80 characters). |
| Save with current asset id | Free | Only when you opened Agent from an existing library image. Overwrites that asset's .vox file. |
| Downloads in View Voxel | Free | Same as other VoxAI assets. |
Save to download
Use the Save to download menu in the Agent toolbar:
Save as new
Enter a name (2 credits). A toast and chat message link you to View Voxel to download.
Save with current asset id
Overwrites the existing library asset immediately when eligible. No extra save credit.
Draft Resume
Agent saves your session locally in the browser (IndexedDB). If you leave and return, you can resume where you left off. Click Start over to clear the workspace and begin fresh.
Agent vs Image-to-3D
| Agent | Image-to-3D (VoxAI V3) | |
|---|---|---|
| Best for | Quick iteration, named parts, chat edits | Complex subjects, high detail, mesh exports |
| Output | MagicaVoxel-native .vox | .vox, .glb, .json + rigging pipeline |
| Workflow | Conversational build and edit | Image → 3D generation pipeline |
| Credits | Variable per chat action + 2 to save as new | 25 per generation (fixed) |