Images and voice
“Make a picture for a post”, “draw a banner for the sale”, “read this text aloud”, “reply to me by voice” — the platform can do this out of the box. You don't need to connect third-party services or hunt for keys: image and speech services are already connected to the platform, and you pay in credits, just like for answers.
Images
Tell the builder in your own words:
Let the agent draw images for posts
It will give the agent the ability to draw and tell you what it costs. From then on the agent draws by itself — in conversations with you, in messengers and in the chat window on your site:
- from scratch — from a description: “a banner for the autumn sale, wide, with the text −20%”;
- from your photo — send a photo and say what to change: “remove the background”, “make it warmer”;
- what it already drew — “make the background blue” about a picture earlier in the same conversation.
The picture arrives wherever the conversation is happening: on your site — inside the message itself; in Telegram, WhatsApp and Viber — as an image; in email, Discord and Slack — as a link. In the dashboard's conversations you see it as a picture.
The agent doesn't know your site's colours by itself. Tell the builder — “our colours are dark green and sand” or colour codes — and it will write them into the agent's personality. If the chat window colour is already set, the builder can take it from there.
Who can ask
A site visitor or a customer in a messenger can also ask for a picture — for example, “show me what the text would look like on the mug”. But you pay for it, so one conversation with a customer is limited to 3 images. After that, the agent says plainly there won't be any more here. Your own conversations and scheduled work have no such limit — the spending cap applies there.
What the service won't draw
The image service itself refuses to draw faces of real people (famous or not) and other companies' trademarks and logos, as well as prohibited content. The agent then says why it didn't work and suggests a different description. It's better to send your own logo as a photo and ask to place it into the picture.
A drawing isn't a photo. The agent won't draw a product you don't have and won't pass a picture off as a photo of a real product. If the agent publishes pictures outside by itself (a social media post through an action), the builder will put that action up for confirmation: you'll see a picture made in your business's name before anyone else does.
Voice
Read text aloud
Let the agent read text aloud
After that, on “read this text aloud” the agent makes an audio recording: in Telegram and WhatsApp it arrives as a voice message, on your site as an audio file. Up to 4096 characters at a time (that's a few minutes of speech); the agent will read longer text in parts.
Reply by voice
Reply to me by voice
Then in your personal chat with the bot in Telegram or WhatsApp, replies arrive as voice messages. You can also have it reply by voice to everyone who writes in messengers — say so explicitly: “reply by voice to all customers”. The builder takes “to me” to mean exactly “only to me”: you pay for every voiced reply.
How it works:
- voice only in Telegram and WhatsApp: there people listen to voice messages right in the chat. On your site, in Viber, Discord, Slack and in email, the reply stays as text;
- a reply that contains a link or a file arrives both as voice and as text: you can't pass on an address by voice;
- when voicing fails (the service didn't respond, credits ran out, the reply is longer than 4096 characters), the reply arrives as text — no silence; the reason is visible in the turn log;
- in the dashboard's conversations, the reply is stored as text — the same text the agent voiced.
By default the agent replies in text. To go back — “reply in text”.
How much it costs
| What | Credits |
|---|---|
| one image — drawn from scratch or edited | 50 |
| voicing | 15 per 1000 characters, proportionally: a short 100-character reply costs 2 credits |
The spending limit is checked before calling the service: when credits run out, the picture isn't drawn or charged, and the agent says so plainly. You only pay for what reached the conversation: a refusal by the service or a failure costs nothing. In the turn log, images and voicing are separate lines with their price. See Credits.
When the service isn't responding
The agent tells the person the truth: no picture or voice right now. And you get a problem — “images aren't being drawn” or “voicing isn't working”. There's nothing to fix: it's a platform service, and the problem closes by itself as soon as the next image or voicing succeeds.
Where the description and text go
The picture description, your photo (if you ask to edit it) and the text to voice are passed by the platform to the image or speech service — a third-party service connected to the platform. Finished pictures and recordings are stored as long as other conversation files and are available by link to whoever the agent sent them to.
If no services are connected (this can happen on a self-hosted platform), the builder will say plainly, when asked for pictures or voice, that this isn't available here.
Next
- Where the agent answers — messengers, website, email.
- How much it costs — credits and limits.