AI tools for graphic designers work best on production, not on the deliverable itself. Figma’s 2025 survey of 2,500 users found 78% say AI makes their work more efficient, but only 31% of designers use it for core design work like asset generation (Figma, April 2025). For product work, we think the gap is provenance. A prompt-first model invents a plausible product. It does not render yours.
A ranking of AI design tools on output quality answers the wrong question. The output is already good enough to look at. It is not good enough to ship, because the thing in the frame is not the client’s product.
There is a second gap underneath it. The stages that survive automation are the ones that repeat, and they mostly repeat per product. A tool that speeds up one asset saves little on a catalog unless you can reuse the setup.
This is for working designers, in-house or agency, who have already tried the tools and found that the results do not ship. It covers what the adoption data shows, which stages of a design workflow survive automation, how to keep the real product in the frame, and what it costs.
Key Takeaways
-
Adoption is high, reliance is not. 78% of Figma’s 2,500 respondents say AI makes their work more efficient. Only 32% agree they can rely on its output (Figma, April 2025).
-
Designers use AI less in core work. 31% of designers use AI in core design work like asset generation, against 59% of developers using it for code generation (Figma, April 2025).
-
The failure is provenance. A text prompt produces a product that resembles the brief. Starting from the real product photo produces the product itself.
-
Automate production, keep judgment. Angles, cutouts, relighting, scene swaps, and placement variants repeat. Concept, composition, and sign-off do not.
-
Build the step once, then run it again on each product. The stages worth automating repeat per SKU, which is where the setup time pays back.
-
Stay on your canvas. The DesignerBox Figma plugin makes product photos, try-ons, videos, and cut-outs without leaving the file.
-
Image models have different strengths, and each vendor describes them in its own documentation. In a DesignerBox workflow, you pick the model for each step.
Do graphic designers use AI tools? What the adoption data says
Figma surveyed 2,500 of its users, published the results in April 2025, and split them by discipline. The split is the story.
78% of respondents agree AI significantly enhances the efficiency of their work. Only 32% agree they can rely on the output of AI in their work (Figma, April 2025). Efficiency and reliance are not the same claim, and the 46-point spread between them is the gap this article is about.
The discipline split widens it. 59% of developers use AI for core development tasks like code generation, against 31% of designers using it in core design work like asset generation. Developer satisfaction runs 82% against 69% for designers, and 68% of developers say AI improves work quality against 54% of designers (Figma, April 2025).
The widest gap is in core design work, the part that ships.
Why prompt-first output fails a design brief
A text-to-image model is trained to produce something that matches a description. Your brief is more than a description: a specific object with a specific finish, a logo lockup with fixed clearance, and a color that has a hex value in a brand book. A brand guidelines template shows how those rules are written down as numbers and named files.
Ask a prompt-first model for “a matte black ceramic mug on a linen backdrop” and it returns a mug. It is a good mug. It is not the client’s mug. The handle curve is wrong, the glaze breaks differently, and the logo is an approximation of a logo.
That is not a quality problem, so a better model does not fix it. Nothing in the pipeline ever saw the product. This is also why AI output gets flagged as generic AI in review: it is drawn from an average, and averages look like averages. We covered the visual symptoms of that separately in why AI images look generic and how to fix it.
The fix is to change what the pipeline starts from. Upload the actual product photo and every downstream asset derives from it. The angles are that mug’s angles. The cutout is that mug’s silhouette. Product accuracy through edits is a measurable property, covered in how accurate AI product photos really are.
Which stages of design work survive automation
Not every stage repeats, and only the repeating stages are worth automating. Here is the split.
| Design stage | Hand to AI | Why |
|---|---|---|
| Brief, strategy, concept | No | Judgment about what the work should argue. No repeatable input. |
| Moodboard and reference | Partly | Fast to generate, cheap to discard, never ships. |
| Asset production: angles, cutouts, relight, scenes | Yes | Identical operation, different SKU. Pure repetition. |
| Composition and layout | No | Hierarchy and balance are the design. |
| Placement variants and resizes | Yes | Mechanical reformatting of approved art direction. |
| Retouching and cleanup | Yes | Consistent standard, no interpretation. |
| Final art direction and sign-off | No | The reason a designer is on the payroll. |
Everything in the yes column shares one property: the decision was already made once, and the work is applying it again. That is the setup work that fills a design week. Agencies make the same argument when scaling creative production without hiring. The other route is to buy the hours, and four ways to outsource graphic design compares those models. If cutouts are the first stage you give to AI, compare AI background removers by job before you pick one. For retouching and cleanup, the best AI photo editors by job compares ten tools.
One setup for every product
The yes column has a second property that decides whether any of this is worth building. Those stages do not repeat a few times inside one job. They repeat once per product, so a brand with 40 new SKUs needs the same four operations 160 times.
That changes what you should aim for. A prompt you retype for each product is a new decision each time. After 160 new decisions, the catalog no longer matches itself. A saved setup is one decision applied 160 times.
Three things make a step reusable rather than repeatable by hand.
- The brand is a record. Colors, fonts, logos, voice and rules sit in a brand profile. You set the brand once, and the workflow reads it on every run.
- The product photo is the input. Provenance and repeatability are the same fix. The setup stays fixed, and the source image changes per SKU.
- The cost is visible before the run. The model and the steps change what a run costs, and you see that number before you press Run.
When the workflow is ready, you can publish it as an app. An app is a form a colleague completes without opening the workflow. A saved workflow runs the same way on the next product, and batch runs that workflow over a whole sheet of products at once. You keep or discard per row, and re-run one row on its own. How an AI creative workflow is put together explains the build, and producing product images in bulk covers where large catalog work breaks.
Working from the canvas you already use
Moving files between a design file, a generation tool, a retoucher and a download folder takes time. Each move is a chance for the brand to drift.
The DesignerBox Figma plugin makes product photos, model try-ons, videos, and clean cut-outs on the Figma canvas. You make them without leaving the file. The practical walkthrough is in AI product photography in Figma. What Figma’s own AI already does is in the guide to Figma AI tools.
For designers who work in an AI chat instead, DesignerBox has 68 tools over MCP. An AI chat such as Claude, ChatGPT or Cursor can then run the same workflows and apps, one product after another.
Model choice per shot
Image models have different strengths. Designers who want control over every model setting often use a node canvas such as ComfyUI. The ComfyUI alternatives shortlist covers the GPU time and the weights license that come with it.
Typography, editing and keeping a character the same across scenes are different strengths. Each vendor describes its model’s strengths in its own documentation, linked in the table below. You pick the model for each step when you build the workflow, and every run after that uses it. A template comes with its model already picked.
| Need | Model | What the vendor says |
|---|---|---|
| Editing an existing image, and legible text in the image | Nano Banana Pro | Google, November 2025 launch post: lighting and aspect-ratio edits, and “correctly rendered and legible text” |
| Typography and small details | FLUX 2 Flex | Black Forest Labs: “Specialized models for typography and keeping small details” |
| The same character or object across scenes | Kontext Multi | Black Forest Labs: keeps “a reference character or object” the same across scenes. BFL’s API docs now call FLUX.1 Kontext Pro “a legacy editing model” (docs.bfl.ai) |
| Styles and layouts that follow the brief | Seedream 5 | ByteDance Seed, Seedream 5.0 Lite page: “Master various styles and layouts” |
The model also changes the cost of the shot, so compare models on one shot before you choose. The full catalog is on the model list.
If you are weighing DesignerBox against a design tool your team already uses, the comparison pages set it against named tools one at a time, and the ad creative roundup ranks the shortlist. Our Canva alternatives shortlist widens that to seven tools.
If a broad model catalog is what draws you to a tool, read how model lists change over time in our guide to DaVinci AI alternatives.
Plans and run costs for designers
Every DesignerBox plan carries a monthly credit allowance, and the same number is charged in USD, GBP and EUR. What changes between the plans is which jobs you can reach.
| Plan | Notable for designers |
|---|---|
| Free | Runs on sample products. Results carry a watermark. No video. |
| Basic | No watermark. No video. |
| Pro | Uploading your own photos and the commercial license start here. No video. |
| Premium | AI video, virtual try-on, upscaling, the image editor and the video editor start here. |
| Ultra | Team features, shared brand kits, white label. |
On annual billing, you get the same credits each month, and they reset every month.
Some costs are fixed. An avatar run returns nine fixed poses for 25 credits. Video has a range: an 8-second clip costs 40 to 560 credits, depending on the model. What AI video costs explains the video side.
Client work needs the commercial license, so start on the Pro plan. Every plan below Ultra is one seat, so a design team that wants shared brand kits needs the Ultra plan. Full plan detail is on the pricing page.
What a designer still owns
The tools removed production time. They did not remove the reason anyone hires a designer.
Nothing in this stack decides what the campaign should say, whether the crop is doing work, or when the whole route is wrong and needs restarting. Only 32% of Figma’s respondents say they can rely on AI output in their work. The judgment call stays with the designer. The same holds for merch: a designer who sells on print on demand still sets the lettering and checks the print file, which AI tools for print on demand covers tool by tool.
What changes is the ratio. Less time on the fourteenth resize, more on the decision that made the first one good.
Anyone can make an AI picture. Making hundreds that still look like your brand is the hard part. The full workflow from the first product photo to the finished ad, in one subscription. The templates, workflows, apps, batch, the image editor, the video editor, the brand record and the file library are in one place. Four separate tools would each keep their own version of the brand.
The first job to build is the one you repeat most. Start from a template, add your brand and your products, and run it.
FAQ
Do graphic designers use AI?
Yes, but selectively. 78% of Figma’s 2,500 respondents say AI makes their work more efficient, while only 31% of designers use it in core design work like asset generation (Figma, April 2025). For comparison, 59% of developers use AI for core development work like code generation.
Will AI replace graphic designers?
In our view, concept, composition and art direction still need a designer. AI tools remove repeated production. In Figma’s survey, only 32% of respondents say they can rely on AI output in their work. Keep a human review step.
Which parts of a design workflow should you automate first?
Start with the stages that repeat per product: angles, cutouts, relighting, scene swaps, resizes and retouching. Those are the same operation on a different SKU, so one setup covers all of them. Leave concept, composition and sign-off alone. They have no repeatable input, so there is nothing to save.
What is the best AI tool for a graphic designer?
Pick by stage. For repeated production work, pick a tool that starts from your product photo and saves the setup. For concept and layout, keep your own judgment. In DesignerBox, you build a workflow once with your brand and run it on each product photo.
Can AI generate an image of my actual client’s product?
It can, if the pipeline starts from a photo rather than a text prompt. Upload the product shot, and the angles, cutouts, relights, and scenes come from that image. Check every result against the real product before it goes to the client.
Can I use AI-generated images in commercial client work?
In DesignerBox, the commercial license starts on the Pro plan. Terms for any other tool vary by vendor and by model, so check the license on the specific plan you hold before delivering client work.
How much does AI design tooling cost per month?
DesignerBox has a free plan, and the paid plans are on its pricing page. The cost of each run depends on the model, and you see it before you press Run.
Sources
- Figma, “Figma’s 2025 AI report: Perspectives from designers and developers”, survey of 2,500 Figma users, published 24 April 2025 (figma.com/blog/figma-2025-ai-report-perspectives), read 15 September 2026
- Google, Nano Banana Pro launch post, November 2025, read 15 September 2026
- Black Forest Labs, FLUX.2 model page, FLUX.1 Kontext model page and FLUX.1 Kontext Pro API reference, read 15 September 2026
- ByteDance Seed, Seedream 5.0 Lite page, read 15 September 2026
- DesignerBox plans, credits a month, plan gates, the avatar run and the 8-second video range: DesignerBox pricing and product pages (designerbox.ai), October 2026
- DesignerBox MCP tool count: DesignerBox MCP page (designerbox.ai/mcp), September 2026
Figma survey figures re-verified on figma.com as of September 2026. Model strengths verified against each vendor’s own page as of September 2026. DesignerBox plans and plan gates from designerbox.ai as of October 2026. Individual results vary.
