Figma now edits images natively. Remove background, expand, boost resolution, erase objects, and replace a background all run on canvas without a plugin. What Figma’s published tools do not cover is work that depends on your physical product: new camera angles of the same object, a garment on a model, or product video. That gap is where an external studio earns its place.
You have one product photo, a PDP layout open in Figma, and six placements to fill by Friday. The photo is a three-quarter front view. The layout needs a top-down, a detail crop, and the same product on a model.
No amount of editing gets you there, because the shots you need were never taken.
This covers what Figma’s AI does with images today, where its published tool list stops, the DesignerBox Figma plugin, and the MCP route for automating the same work.
Key Takeaways
- Figma removes backgrounds natively. Isolate, Expand, Erase, Boost resolution, Remove background and Edit with prompt all ship on canvas (figma.com, July 2026). A plugin whose pitch was background removal has nothing left to sell.
- Weave tools add Replace background and Change lighting, rolling out to Professional plans and above with Full seats from 24 June 2026, in open beta and consuming no AI credits until general availability (help.figma.com, July 2026).
- Figma’s published tool list does not include virtual try-on or new camera angles of the same physical product. Both need a model that starts from your product photo.
- The DesignerBox Figma plugin puts generation on the canvas. Add it from designerbox.ai/plugins/figma.
- MCP covers the automated route. The DesignerBox MCP server is live with 43 tools across 8 groups.
- Images reach a Figma file through upload_assets, not through the write-to-canvas tool, which documents no image support yet (developers.figma.com, July 2026).
- An image costs 5 credits and the free plan starts with 112, so roughly 22 images before you pay anything.
What Figma’s AI does with images today
Figma got good at images. As of July 2026 the published on-canvas list is Isolate, Expand, Erase, Boost resolution, Remove background, and Edit with prompt (figma.com/solutions/ai-image-editor, July 2026).
| Tool | What it does |
|---|---|
| Isolate | Lasso an object to edit or reposition it without touching the background |
| Erase | Remove an object or text from an image |
| Expand | Extend the background to fit a new aspect ratio without distortion |
| Boost resolution | Raise the clarity of a low-resolution image |
| Remove background | Isolate the main subject |
| Edit with prompt | Make edits by typing an instruction |
Weave tools in Figma Design sit on top of that, adding Replace background, Change lighting, Text to vector illustration, and Apply color palette, with Generate mockup and Transfer style also listed. They rolled out gradually from 24 June 2026 to Professional plans and above with Full seats. They are in open beta, free to use, and consume no AI credits until general availability, at which point credit usage applies (help.figma.com, July 2026). They are not available on Starter, K-12 Education, or Government plans.
Read that list twice if you were about to pay for a background remover. Figma does it, on canvas, at no credit cost during the beta.
Where Figma’s published tools stop
Every tool above edits pixels you already have. That is the boundary, and it is a clean one.
Expand invents more background. Boost resolution sharpens what is there. Edit with prompt changes the image you uploaded. None of them shows you the back of the shoe, because the back of the shoe is not in the file.
Figma’s published tool lists do not describe:
- A new camera angle of the same physical product. Not a similar product. Yours, from a side the camera never saw.
- A garment on a human model. Virtual try-on does not appear in Figma’s documented tool list as of July 2026.
- Product video. The image tools are image tools.
That is scope, not criticism. Figma is a design tool that learned to edit images well. Generating your product from angles that were never shot is a different job, and it needs a model that starts from your actual product photo.
The DesignerBox Figma plugin
The plugin brings DesignerBox generation to the canvas you are already working in. Start at designerbox.ai/plugins/figma and add it to your files from there.
It covers the work Figma’s own image tools do not: new camera angles of your actual product, a garment on a model, and product stills generated from one source photo, landing in the file rather than in a downloads folder.
For work you want to automate or rerun across a catalogue, the MCP route below goes further. The two are complementary. The plugin handles canvas work, MCP handles pipelines.
The automated route: MCP
MCP, the Model Context Protocol, is an open standard for connecting AI applications to external systems. It was created at Anthropic and open-sourced in November 2024 (anthropic.com, November 2024), then donated to the Linux Foundation’s Agentic AI Foundation in December 2025 (anthropic.com, December 2025). Claude, ChatGPT and Cursor all support remote MCP servers over HTTPS (claude.com, developers.openai.com, cursor.com, July 2026).
The DesignerBox MCP server is live, exposing 43 tools across 8 groups, including image and video generation, avatars, brand profiles, and saved workflows. Your AI client drives the same models the web studio does.
That gets you the shot. Landing it on the canvas is the second half.
Figma shipped its own on-canvas design agent in May 2026 (help.figma.com, May 2026). For how in-canvas assistants compare with the autonomous kind, see what to hand an AI creative agent and what to keep.
Getting the image onto the Figma canvas
Figma runs its own remote MCP server (developers.figma.com, July 2026), and this is where most write-ups get it wrong.
Figma exposes two different tools, and they do not have the same powers:
- use_figma, the general write-to-canvas tool, states “No assets (image) support yet” in Figma’s own FAQ. If you have seen a grey box where your image should be, that is why (help.figma.com, July 2026).
- upload_assets does handle images. Figma documents it as uploading PNG, JPG, GIF and WebP assets into a file, up to 10MB each. Give it a node URL and the image lands as a fill on that node. Give it none and Figma creates new frames with the images as fills (developers.figma.com, July 2026).
Images go through upload_assets, not through the tool that writes your designs. Two caveats before you build a pipeline on it:
- Figma restricts MCP access to clients in its own catalogue, naming VS Code, Cursor and Claude Code (developers.figma.com, July 2026). Claude Desktop is absent from that named list.
- Write to canvas is in beta and free during the beta, and Figma states it will become a usage-based paid feature (developers.figma.com, July 2026).
If that is more plumbing than the job deserves, the short path is unglamorous and reliable. Generate in DesignerBox, drag the file into Figma. It takes seconds, and nothing in it is in beta.
The workflow, start to finish
- Shoot one accurate photo of the real product. Everything downstream derives from it, so this frame is the one that has to be right. Even lighting, true colour, no clutter.
- Generate the shots the layout needs. Photo Angles turns that single frame into multiple views. Outfit to Image and virtual try-on put a garment on a model.
- Pick the model per shot if it matters. The catalogue runs 13 image and video models across six providers, on one bill.
- Bring it into Figma. Use the plugin, drag the file in, or drive upload_assets from a catalogue client.
- Finish on canvas. Expand to the aspect ratio each placement wants, boost resolution, erase what the layout does not need. Figma is genuinely good at this part.
Step 5 is the point worth sitting with. The division of labour is doing the work: DesignerBox produces shots that do not exist, Figma finishes the ones you have.
What it costs
Generating or editing an image costs 5 credits. The free plan includes 112 credits, so roughly 22 images, with no credit card.
| Plan | Monthly | Credits |
|---|---|---|
| Free | $0 | 112 |
| Basic | $15 | 500 |
| Pro | $35 | 1,000 |
| Premium | $75 | 2,500 |
| Ultra | $200 | 8,000 |
Three limits worth knowing before you plan around them:
- Video is the expensive operation. It bills per second of output, and it adds up faster than people expect. A single 8 second clip with native audio can run 6,400 credits, which is more than the entire Premium monthly allocation of 2,500. Budget video separately or not at all.
- Virtual try-on and AI video need Premium at $75/month. They are not on Basic or Pro.
- The commercial licence starts at Pro. If the image runs as a paid ad, that is the floor. Worth checking against pricing before a campaign, not after.
Against what a product photoshoot costs, the arithmetic is not close. The trade is that you still need one accurate photo of the real thing.
FAQ
Can Figma remove an image background without a plugin?
Yes. Remove background ships natively in Figma Design, alongside Isolate, Erase, Expand, Boost resolution and Edit with prompt (figma.com, July 2026). Weave tools add Replace background for Professional plans and above with Full seats, in open beta since 24 June 2026 with no AI credit cost until general availability (help.figma.com, July 2026).
Does Figma have virtual try-on?
Virtual try-on does not appear in Figma’s published tool lists as of July 2026. Figma’s image tools edit the image you already have. Putting a garment on a model means generating something the photo does not contain, which is a different capability. DesignerBox includes try-on from the Premium tier at $75/month.
How do I get the DesignerBox Figma plugin?
Add it from the plugin page at designerbox.ai/plugins/figma. It puts DesignerBox generation on the Figma canvas: new angles of your product, on-model shots, and stills built from one source photo. For repeatable or automated work, the DesignerBox MCP server exposes 43 tools to any client that supports remote MCP over HTTPS.
How do I get an AI-generated image into a Figma file?
Two ways. Drag the exported file onto the canvas, which always works. Or use Figma’s MCP server: upload_assets accepts PNG, JPG, GIF and WebP up to 10MB and places them as a fill on a node or as new frames (developers.figma.com, July 2026). The general write-to-canvas tool does not handle images yet, so the upload path is the one to use.
Which AI clients can connect to Figma’s MCP server?
Figma names VS Code, Cursor and Claude Code as catalogue clients (developers.figma.com, July 2026). Claude Desktop is not on that named list. The DesignerBox MCP server has no such restriction and works with any client that supports remote MCP over HTTPS, including Claude, ChatGPT and Cursor.
Do I need a product photo, or can I start from a prompt?
You can start from a text brief, but for product work you want the photo. A prompt produces a plausible object. Your photo produces your object, with the right logo placement, the right stitching, the right colourway. For a PDP that distinction is the whole job.
Can I use AI product images in paid ads?
The commercial licence starts at the Pro tier ($35/month) in DesignerBox. Verify the current terms on the live pricing page before a campaign, and check each ad platform’s own AI disclosure rules separately, since those are set by the platform and change independently.
Figma capabilities verified against figma.com, help.figma.com and developers.figma.com as of July 2026. MCP facts verified against anthropic.com, claude.com, cursor.com and developers.openai.com. DesignerBox pricing and credits from the product brief. This category changes monthly, so re-check any capability claim before relying on it. Individual results vary.
