FlexPic / Tutorials / Workflow and API
Plan your work with Chat Mode
Describe a goal in plain language and let the assistant suggest the right tools with prompts filled in, then run them yourself.
Goal
Get from a rough idea to the right tool and a good first prompt without browsing the whole tool list.
Before you start
- You must be signed in. Chat Mode asks you to sign in if you are not.
- A few credits. Each message costs a small number of credits (4 at the default setting), and a message that fails is refunded.
Steps
- In the sidebar, under Agents, click Chat Mode. The page opens with the heading "Say it, and it's made".
- Type your goal in the message box, or click an example such as Create product ads, Design a film poster or Music for my film. More ideas shows others.
- Write what you want in plain language. For example:
Prompt I need a 5 second vertical clip of a coffee cup with steam for an Instagram story - Send the message. The assistant replies with advice and, when it can, suggests tools. Suggestions appear as cards that show the tool name and sometimes a model.
- Click a card. The tool opens with the suggested prompt already filled in.
- Review the prompt, change what you like, check the price on Generate and click it. Chat Mode never generates for you: you still press Generate in the tool.
- Use the Ask first menu beside the message box to change how suggestions behave. With Auto run, the first suggested tool opens automatically with its prompt filled in.
What you should get
A short conversation that ends with one or more tools open and ready, with prompts you can edit before you spend credits.
Cost example
Chat messages use a few credits each, far less than a generation. For comparison, a 5 second Seedance 2.0 clip at 720p costs about 957 credits, so a few messages to plan it are a small share of the total.
Tips
- State the goal, the platform and the length: "9:16 for TikTok, 10 seconds".
- Ask for options: "give me three prompt variations".
- Use the assistant to tighten a long prompt, then test it at low resolution.
Common mistakes
- Expecting the assistant to render the result. It suggests tools; the tools generate.
- Sending a very vague request such as "make something cool". Name a subject and a format.
- Trusting a suggested model blindly. Check what it costs and whether it is available.