Design and edit a template

Design a branded document or image that you can use again with new data. In the studio, the AI chat and the editor work on the same draft, so you can switch between them as you make changes.

These steps use the web app. To create a PDF from code, follow Create your first PDF.

Choose a start

Read the price beside the chat button before you send a message. Opening the page does not spend credits. See credits and billing for AI costs, export costs, and failed work.

Set up your brand kit

Create a brand kit before you use the AI chat. Open Brand kits, create a kit, and add your colours, tone guidance, logos, and fonts.

Use PNG, JPEG, or WebP for a logo. Choose an included font or upload a WOFF2 file. Each file can be at most 10 MiB. Animated images are not supported. Use files you have the right to use.

Choose a kit before you send the first message. The chat keeps that kit, and each new message uses its current details. Changing the kit later does not change a saved template version.

The AI receives the kit's name, colours, tone, and primary logo. Check the draft to see which fonts and logos it uses. Uploading an asset does not guarantee that the AI includes it.

Archiving a kit stops changes and new uploads. It does not delete the kit's assets. Replacing an asset keeps the old file for work that already uses it.

Describe a template

1. Open Build a template and choose your brand kit.
2. Say what the result must show. Name the fields that change between uses.
3. Choose PDF or Image under Output, or let the AI choose.
4. Check the credit cost, then send the message.
5. When the message says Done, choose Show preview and check the result.

For example:

Make a one-page project brief PDF with my brand colours.
Include a project name, client name, due date, summary, and a list of tasks.
Keep those values as data fields so I can reuse the template.

Send another message to change the draft, such as "Make the task text larger." The chat receives your current canvas and source edits before it starts, so it works from your latest changes.

Start with HTML

Open Paste HTML, paste your source, and open it in the editor. This step uses no AI credits. Use the Source tab to change the HTML and the Data panel to fill its fields.

To apply your brand with AI, attach the HTML file in the chat. This sends the HTML to the model provider and uses AI credits. The chat sends the attachment on its own. Any words in the message box stay there for your next message.

The visual editor supports one page of placed text, images, boxes, tables, and lists. It keeps other HTML in source blocks. If your HTML has no editor canvas, it opens in flow mode and the Add controls are off. You can still change the source, preview the result, and publish it.

The editor saves a consistent HTML layout and keeps unsupported source blocks. Check the Source tab before you save if exact source formatting matters. Read templates and data for the supported field syntax.

Rebuild a source image

1. Open Import image and choose your brand kit.
2. Attach one PNG, JPEG, or WebP image you have the right to use.
3. Choose your output, check the cost, and send the file.
4. Open the completed preview and check text, layout, branding, and any numbers.
5. Use the chat or editor to fix the result before you publish it.

The source image can be at most 10 MiB, 7,680 pixels wide, and 4,320 pixels high. It must have one frame and no more than 32 million pixels. Corrupt, animated, and unsupported files fail before they reach the model provider.

The import rebuilds the layout with your brand so you can edit and reuse it. Expect to review the result, since it may not match every pixel. The AI receives instructions to replace the source brand and issuer name. Check both before you publish. The Terms of Service include your warranty that you have the rights to use the source.

For image output, the import uses the source dimensions unless your API request sets both width and height. See PDF and image options for file formats and size limits.

Adapt a gallery template

1. Open a design in the gallery and read its purpose, output formats, and expected data.
2. Choose Copy to my account. Sign in first if needed.
3. Open Properties and change the sample values already loaded in Data.
4. Change the design in the canvas or source. To ask the AI for a brand change, choose your kit and send a message.
5. Check the live layout and press Save template. Use Export when you need a file.

Your copy starts at version 1. Your edits do not change the gallery original. You can move, resize, remove, and add objects. Sample values stay in this browser for your next export.

Edit, preview, export, and save

Select an object on the canvas or in the object list. Drag it to move it, or use the keyboard and inspector controls. Use Undo and Redo to step through your edits.

Fill the template's fields and tables in the Data panel. Choose Fill with example values to start with sample data. Use the JSON view to enter data directly, or Advanced to add a schema.

Canvas shows the layout as you edit and keeps field names visible. Open Preview and press Refresh preview to fill the template with your sample data. Both views are free. Refresh again after you change the design or data.

Export creates a PDF or image and costs one credit when it succeeds. It shows progress, a download, and your recent files. Show preview on a completed chat message opens its existing result without another charge. See downloads and limits for file rules.

Press Save template. The first save asks for a name. Each save keeps the current source and schema as a new version in your account. You choose when to save, including after an AI message. Your saved template stays in your account and does not enter the public gallery.

Save work you need to keep. The browser can offer to restore unsaved edits when you return. Version history shows earlier designs and can restore one as a new version. Read templates and data for version rules and API use.

AI review, privacy, and failures

Check every AI draft before you use it, especially names, amounts, dates, text, and image quality. thirds.ai renders each draft and asks the model to review it. A prompt or HTML message gets one review. An image import gets up to five repair reviews. You still need to check the result yourself.

The AI uses GLM-5.3-Flash through OpenRouter. We send your prompt, submitted HTML, working draft, brand details, source image, and review preview as needed. OpenRouter and its selected model host receive this content.

The model host must not train on or keep your content. The request fails if no host meets those rules. The selected host can change, and we do not promise that processing stays in the UK or EU. OpenRouter can keep request metadata, such as token counts and latency.

We keep chat messages and accepted drafts until account closure. We remove hidden attempts and temporary brand copies when the message ends. AI chat content stays out of backups, logs, and analytics. Read the privacy policy for the full data notice.

The build keeps a record of the input type, selected brand kit, and message history. The draft also records whether AI made it. These records describe how the draft starts. They do not prove rights to a source image or certify the result.

If an AI message fails, you keep the prior draft and get the reserved credits back. Wait for the current message to finish before you send another. Repeated failed or cancelled work can reach a daily limit. You can still use the canvas, source, and preview while the chat is paused. Read retries and limits to find the next step for your error code.