| name | lovart-design-agent |
| description | Operate Lovart AI Design Agent and ChatCanvas in an already-open browser session. Use when Codex needs to create Lovart projects, generate images or videos, edit canvas assets, split or generate named layers, use Brand Kits, references, Skills, Touch Edit, Edit Elements, Remove BG, export assets, or verify Lovart outputs on the live canvas. |
Lovart Design Agent
Core Rule
Treat Lovart as a design agent on an infinite ChatCanvas, not as a one-shot image generator. Work in the live browser page when Lovart is already open, keep the project tab as the deliverable, and verify results from the visible canvas plus the right-side chat or Generated Files list.
Operating Flow
- Attach to the open Lovart tab with the Chrome/browser automation workflow when the user says Lovart is already open.
- Start from a New Project, a built-in Lovart Skill, or the current canvas depending on the user's target.
- Write prompts that specify deliverable type, subject, style, target format, aspect ratio or dimensions if relevant, references, brand constraints, and editability requirements.
- Wait for generation to finish. Do not claim completion while Lovart still shows "生成中", "Generating", or an active timer.
- Select the generated canvas object before asking for follow-up edits so Lovart attaches that object as context.
- Use Lovart's native editing controls for precise edits instead of vague regeneration.
- Keep the final project/canvas tab open as a deliverable unless the user asks to close it.
Task Patterns
Generate New Assets
- Use Agent/Auto mode for broad design briefs, multi-step campaigns, brand systems, and tasks where Lovart should choose the right model.
- Use Image Generator (
A) or Video Generator (S) when the model, aspect ratio, resolution, length, or reference mode must be controlled directly.
- Use quickstart projects or Lovart Skills when the task matches a known format such as logo, Instagram post, UGC image, product listing, mockup, or video ad.
Edit Existing Assets
Use the selected object's toolbar or prompts:
Quick Edit or Tab: fast selected-image edits and A/B variants.Mark / Touch Edit, M, or Ctrl+Click on Windows: localized object/region edits.Edit Elements: semantic layer splitting into movable/editable elements.Edit Text: modify text inside an image while preserving style.Remove BG: isolate a subject with transparent background.Eraser: remove an object and fill the gap.Expand: outpaint or change aspect ratio.Move Object: reposition a selected object while reconstructing the original area.Multi-Angles: generate alternate views of the same subject.Vectorize: convert suitable raster logos/icons/illustrations to SVG.Mockup: place a design onto packaging, apparel, posters, cups, or other real-world surfaces.
Layer Work
Do not stop at a written layering plan. Ask Lovart to actually run Edit Elements or generate named independent layer assets on the canvas.
Use this Chinese follow-up when Lovart only gives instructions:
请继续，不只是说明方案。请基于当前选中的图像实际生成/分离可编辑图层，并把每个图层作为独立对象放在画布中，方便后续单独选择、移动、编辑和导出。
For portrait/person layer work, request named layers such as:
背景层、人物主体层、脸部皮肤层、头发层、服装层、配饰层、光影氛围层
For exportable layer packages, select all intended layer objects and export as PSD when available.
References
Load references/lovart-workflows.md for detailed operating notes, prompt templates, model/reference guidance, and export rules.
