The Lock-in agent
Studio's chat agent plans a campaign with you, asks questions, proposes generations with prices and runs them. Modes, playbooks, planner models, audio and costs.
For the complete documentation index, see llms.txt. This page is also available as Markdown.
The Lock-in agent is Studio's chat, at /studio/chat. You describe a goal, and the agent plans a campaign, asks you questions, proposes images, video and audio with prices, runs them, and can look at an image it made. You can open the page without an account, but sending a message needs you to sign in.
#What the agent can do
- Ask you questions to settle the brief.
- Look up the available models and load a playbook.
- Generate images, edit an image, generate video and generate audio.
- Analyse an image it created and check on running generations.
#Ask mode and Auto mode
The mode chip next to the message box reads Ask or Auto and opens a two-row menu. Ask before generating shows each proposal with its price and waits: the card has an Approve button with the credits on it, Always allow (approve and generate without asking from now on) and Stop (Esc). Generate without asking runs each generation as soon as it is priced. A proposal can hold up to 4 outputs, and a message can carry up to 4 proposals.
#Playbooks
The slash list holds 12 built-in playbooks. The time next to each is the estimate the list shows.
| Playbook | Estimate |
|---|---|
| Product photoshoot | 3 min |
| Launch campaign | 6 min |
| UGC video | 6 min |
| Ad variations | 3 min |
| Social carousel | 3 min |
| Thumbnail | 2 min |
| Brand kit | 5 min |
| Lookbook | 4 min |
| Cinematic trailer | 10 min |
| Character sheet | 3 min |
| Faceless explainer | 8 min |
| Logo reveal | 3 min |
#Planner models and what a message costs
The planner model is the chat model that thinks with you. Studio lists Auto plus 30 models. Each message costs credits set by the model you choose, from 4 to 369. For example, GPT-6 Luna is 4 credits a message, GLM-5.3 Flash is 5, Claude Opus 5.5 is 146, and Claude Fable 5 is 369. The full list is in Planner models.
- Auto is free for your first 50 messages each UTC day, then 5 credits a message.
- Auto routes a simple turn to GPT-6 Luna, a normal turn to Gemini 3.8 Flash, and a turn with an image you attached to Claude Opus 5.5.
- Some planner models read text only: GLM-5.3, GLM-5.2, DeepSeek V4 Flash and DeepSeek V4 Pro. If you attach an image, the Studio swaps in the model Auto uses for images and tells you.
#Audio lives here
Studio has no standalone audio page. Speech, music and sound effects are made by asking the agent, and are charged at the approval card.
| Audio model | Kind | Credits |
|---|---|---|
| ElevenLabs v3 | Speech, up to 5,000 characters | 184 per 1,000 characters |
| ElevenLabs Turbo v2.5 | Speech, up to 5,000 characters | 92 per 1,000 characters |
| MiniMax Speech 2.8 HD | Speech, up to 10,000 characters | 184 per 1,000 characters |
| MiniMax Speech 2.8 Turbo | Speech, up to 10,000 characters | 111 per 1,000 characters |
| ElevenLabs Music v2.5 | Music, 3 to 180 s, can be instrumental | 1,103 per minute |
| Google Lyria 3.5 | Music, the model picks the length | 184 per song |
| ElevenLabs Sound Effects v2 | Sound effects, 0.5 to 22 s | 4 per second |
| MMAudio v2 | Video to audio, 1 to 30 s, needs a video from the chat | 2 per second |
#Limits per message
- Up to 6 images attached. Only the images on your latest message count as vision input, and at most 4 of them are sent to the model.
- Up to 8,000 characters of text.
- Up to 4 proposals, each with up to 4 outputs.
#Temporary chat
Open a temporary chat with ?temporary=1 on the chat address. It is kept in memory only, is dropped when you leave, and lasts at most 24 hours from when it began. Its generations still land in your Library.