Image Generation in Semantics Chat

Image generation lets you create a visual directly from a conversation. You describe the image you want in natural language, then the assistant generates the image and displays it in the conversation thread.

You can use this feature to create an illustration, a thumbnail, a presentation visual, a banner, or a working image to refine further.

Generating an image

To generate an image, simply write your request in the conversation.

Examples:

• "Generate an illustration of a virtual assistant helping a customer advisor, in a flat design style."
• "Create a square image to present a document search feature."
• "Make a landscape banner on the theme of generative AI in business."

The assistant automatically prepares the generation prompt and chooses the parameters best suited to your request.

Modifying an existing image

You can also request a new version based on an image already present in the conversation.

The starting image can be:

• an image you uploaded;
• an image previously generated by the assistant.

Examples:

• "Make this image brighter."
• "Keep the same composition, but add a character in the foreground."
• "Turn this image into an editorial illustration."
• "Keep the main subject and replace the background with a transparent background."

Modification works through guided generation: the source image serves as a reference, then a new image is created. It is not a precise zone-by-zone retouching.

Available actions

Once the image is generated, it is displayed directly in Semantics Chat.

You can:

• copy the image to the clipboard;
• download the image;
• request a new version by continuing the conversation.

Each new request produces a new image. You can therefore progressively refine the result by indicating what you want to keep or change.

Tips for getting a good result

The more precise your request, the more likely the result is to match your needs.

You can specify:

• the main subject;
• the expected visual style: photo, illustration, icon, mockup, flat design;
• the desired format: square, portrait, or landscape;
• the intended use: documentation, presentation, thumbnail, banner;
• the mood: understated, professional, modern, warm;
• colors to use or avoid;
• elements that should or should not appear.

Example of a detailed request:

"Create a simple illustration, on a transparent background, showing a conversational assistant generating an image from text. SaaS product style, blue and green colors, no text in the image."

Available formats

The following formats can be used depending on your needs:

Format	Size
Square	1024 x 1024
Landscape	1536 x 1024
Portrait	1024 x 1536

You can specify the expected format in your request, for example: "in landscape format" or "as a square image".

Known limitations

Image generation can produce variations between two similar requests. If you want to keep a specific element, state it explicitly in your modification request.

In its first version, the feature does not allow you to:

• generate several variants in a single response;
• precisely modify a selected area of the image;
• retouch the image pixel by pixel;
• manually choose all advanced parameters;
• automatically save the image as a document attached to the conversation.

Possible errors

Some requests may fail or be refused.

This can happen if:

• the request does not comply with the generation service's security rules;
• the service is temporarily unavailable;
• the generation takes too long;
• a usage limit has been reached.

In this case, Semantics Chat displays an error message in the conversation. You can rephrase your request or try again later.

Availability

Image generation is only available if it has been enabled on the application being used. For cost reasons, the feature is primarily intended for private applications.

If the assistant does not generate an image despite your request, check with the application administrator that the image generation tool is indeed available.