Using AI chat
Chat is a conversation with an AI that can also generate content. It runs on whichever chat-capable provider you have active, and the same underlying assistant is available both here in AI Studio and inside the editor, where it can additionally edit your canvas directly.
Quick overview
- Switch the composer to Sohbet (chat) mode, alongside the Image and Video modes.
- Type your message, and attach an image if it helps.
- Turn on a skill with a
/command for a focused job. - Continue the thread, or start a new one from history.
Step by step
Open AI Studio from the app sidebar (the item tooltipped Create with AI) to reach the composer. Everything below happens in that composer.
1. Send your first message
- In the composer, switch the mode control to Sohbet (Chat).
- Pick the provider you want to chat on from the provider selector. If none is connected yet, add one first from Settings, AI Settings, Provider & Model tab (see Connecting a provider).
- Type a question, for example
Suggest 3 name ideas for a minimalist stationery brand, one word each. - Press Enter, or click the send button.
- Read the reply. To dig in, type a follow-up in the same box, for example
Make the second one sound more premium.The assistant sees the recent messages in this thread, so it understands what "the second one" refers to.
2. Attach an image and ask about it
- Click the paperclip button, drag an image onto the composer, or paste one from your clipboard. You can attach more than one.
- Ask about it, for example
Here is a screenshot of my poster draft. What would make it read more clearly from across a room? - Send. If your active model cannot see images but the same provider offers a vision-capable model, the app quietly switches to that model for this one message, so the attachment still works.
- If DeepSeek is your active provider, note that it has no vision models at all and no fallback exists, so an attachment cannot be read there. Switch to a provider that supports vision (for example OpenAI, Google Gemini, Anthropic Claude, a vision-capable Qwen, Kimi, or GLM model, or a custom endpoint) before asking about an image. See Troubleshooting.
3. Turn on a skill
- Type
/copywriterand press Enter, or type/to open the skill menu and choose Copywriter from the list. - Ask for the job, for example
5 headline variations for a summer sale banner, each under 6 words. - The reply stays in the copywriter voice. Keep asking follow-ups and the skill stays on for this provider.
- When you are done, type
/off(or/none) to return to the default assistant.
4. Manage conversation threads
- Open the history sidebar from the panel header.
- Click the plus button to start a new thread, then send your first message. The first exchange sets the thread's title automatically.
- Continue by sending more messages in the same thread. Recent messages are sent back as context, which is what makes follow-ups work.
- For an unrelated topic, click plus again to start a fresh thread instead of continuing the old one, so replies stay focused.
- To remove a thread you no longer need, delete it from the history list.
Attachments and vision
Give the assistant something to look at by attaching a file with the paperclip button, dragging one onto the composer, or pasting from your clipboard, all three accept more than one image at once. Vision support depends on the active model: it works on every model for OpenAI, Google Gemini, Anthropic Claude, and a custom endpoint, is limited to specific models on Qwen, Kimi, and GLM, and is not available at all on DeepSeek. If your active model cannot see images but the same provider offers one that can, the app switches to it for that message automatically.
Screenshot capture is an editor-only control
Inside the editor, the assistant can also capture the current canvas as an attachment. That control is hidden here in standalone AI Studio, since there is no design canvas to capture, attach, drag, or paste still work the same way.
Skills
Type / in the chat box to open the skill menu, or use one of these commands directly.
| Command | Result |
|---|---|
/<skill-name> | Activates that skill for the current provider, for example /copywriter |
/skills | Lists every skill, built-in and custom, with which provider each is assigned to |
/off or /none | Turns the active skill off and returns to the default assistant |
A skill layers a reusable directive onto the provider you already have connected, so the assistant focuses on one job, like copywriting or translation, rather than needing you to re-explain it every time. Each provider remembers its own active skill. See Skills and kits for the full list and how to build your own.
Composer modes
The same composer switches between three modes with one control, and your provider selection is remembered separately for each.
| Mode | Purpose |
|---|---|
| Sohbet (Chat) | A conversation, with attachments and skills |
| Resim (Image) | Generates images, see Generating images |
| Video | Generates video clips, see Generating video |
Conversations
Every chat lives in a history sidebar, opened from the panel header. Start a new thread with the plus button, switch between threads by clicking one, or delete a thread you no longer need. The first exchange in a thread sets its title automatically. Threads are saved to your account, so they carry over between sessions and devices, and recent messages in the active thread are sent back to the provider as context on every new message.
Reasoning effort
Providers that support deeper reasoning expose a reasoning effort level, for example Low, Medium, or High. This is not a per-message toggle in the composer, it is set per provider in AI settings and then applied automatically to every request on that provider, so raise it there before a hard problem and lower it again for quick answers. Unlike your API key, which is stored server side, this level is saved only in the browser you set it in, so it does not travel with your account (see Troubleshooting).
Start a new thread for a new topic
Recent messages in the active thread are sent back to the provider as context on every message, which is what makes follow-ups work. For an unrelated question, start a fresh thread instead of continuing an old one, it keeps replies focused and keeps each request smaller.
Common tasks
A few realistic first prompts to copy and adapt. Chat is text-first, so these read like instructions to a teammate.
| Goal | Example prompt |
|---|---|
| Names for a brand | Give me 8 name ideas for a reusable-bottle brand, short and easy to say, no hyphens. |
| Punch up a description | Rewrite this product description to sound more premium and cut it to two sentences: [paste text]. |
| Caption variations | Write 5 Instagram captions for the launch of a matte black water bottle, each under 120 characters, with one emoji max. |
| Review a design | Here is a screenshot of my pricing page. Name the three biggest clarity problems and suggest a fix for each. (attach the screenshot) |
For a focused job, turn on the matching skill first. For example type /copywriter before the caption prompt, or /translator before Translate this headline into German and keep it under 8 words. Turn the skill back off with /off when you move on.
In the editor
The same assistant, with the same providers, skills, and threads, is available inside the editor, where it can additionally add and edit objects on the canvas directly and take a screenshot of what you are working on. See AI chat assistant.
| AI Studio chat | In-editor assistant | |
|---|---|---|
| Providers, skills, reasoning effort | Same | Same |
| Attach, drag, paste an image | Yes | Yes |
| Capture the canvas as an attachment | No canvas to capture | Yes |
| Edit objects on the canvas from a reply | No canvas to edit | Yes, as a batch of undoable actions |
Use AI Studio chat for general conversation, research, and generation outside a specific design, and the in-editor assistant when you want the reply to change what's actually on your canvas.
One key, many jobs
Chat, skills, and generation all run on the providers you connect. Switch which one is active for chat from AI settings at any time.
Troubleshooting
The assistant cannot see my attached image. Vision depends on the active model. If DeepSeek is your active provider, it has zero vision models and no automatic fallback exists, so it can never read an attachment. Switch to a provider or model that supports vision (OpenAI, Google Gemini, Anthropic Claude, a custom endpoint, or a vision-capable Qwen, Kimi, or GLM model). On providers other than DeepSeek, if your active model is text-only but the provider also offers a vision model, the app switches to it for that one message automatically, so the attachment is still read.
My reasoning effort reset when I opened a different browser or computer. The reasoning effort level is a local preference stored in the browser you set it in, per provider, in that browser's local storage. Your API key is encrypted and stored server side, so it follows you to any device, but the effort level does not sync. On a new browser or computer it starts at Default there, so set it again in Settings, AI Settings, Provider & Model tab.
Follow-up answers are drifting off topic. Recent messages in the active thread are sent back as context, so a long thread that has wandered onto several topics can confuse later replies. Start a fresh thread from the history sidebar for the new topic to keep replies focused and each request smaller.
A skill keeps shaping answers I did not mean it to. Each provider remembers its own active skill, so a skill you turned on earlier stays on until you clear it. Type /off or /none to return to the default assistant, or /skills to see which skill is assigned to which provider.