How to use the Venice agentic chat

What the agentic chat changes

When you open Venice now, you land in the agentic chat window. Instead of picking a model, opening an image editor, and stitching steps together yourself, you describe what you want and the agent runs the workflow for you. It selects tools, generates images, searches the web, and edits media based on your prompt.

This guide covers three workflows that show the range: building a brand style guide from a logo, running deep research with a follow-up update, and editing an image and turning it into video.

Workflow 1: a brand style guide from a logo

Attach an image and describe the deliverable in plain language. In the demo, the user uploads the Venice logo and prompts:

Help me create a brand style guide using my logo as the basis. It's called my business, the Venice Agentic Agency.

The agent returns a full style guide: new logo options, font selections, textures, mockups, and a brand architecture diagram rendered as a mermaid chart, plus a style guide document. You do not manually combine or edit anything; the single prompt drives the whole output.

Note that text inside generated images may show spelling or rendering glitches. That comes from the default open-source image models, covered in the next section.

Model selection, privacy tiers, and credits

Use the selector arrow to see and change the model behind your workflow. The choice affects both cost and privacy.

The default in the demo is Kimi K2.5: conversations stay private and requests are included with your subscription. Switching to Kimi K2.6 can be faster but spends credits on tokens. You can also pick a frontier agentic model like Claude Sonnet or Opus, which costs more and shifts you from private to anonymized requests, since your prompts route through Venice to Anthropic rather than staying fully local to Venice.

One side effect of an older model: Kimi K2.5 may assume the current year is 2025 based on its training, which matters for time-sensitive tasks.

For images, models without a green credits icon are included at no extra cost. The agent defaults to these. The demo highlights a zero-credit Turbo image generator as the default for Pro accounts, described as the best output you can get without spending credits, which is why the agent reaches for it first.

When you want higher-quality images, look for the option to enable premium image models and turn it on. By default the agent stays on the free models; enabling premium grants permission to use credit-based models for better results.

Workflow 2: deep research with current context

Switch back to the default private model and ask for research. The demo prompt:

Research and create a report on the current issues in AI privacy and surveillance.

The agent searches the web across multiple angles (facial recognition, data collection, government monitoring, algorithmic bias, predictive policing), pulls articles, and compiles a thorough report.

Because the model's training can lag the calendar, steer it with a follow-up to pull in the right timeframe:

Update for year 2026.

That triggers a fresh pass and adds current topics to the report. Treat explicit date instructions as a normal part of research prompting, especially with older models.

Workflow 3: deep image editing

Upload a photo and describe the edit in one prompt. The demo starts with a headshot:

Put me in outer space on a flying saucer with green skinned aliens. Photo realistic, not cartoon.

Then it iterates on the result with another plain-language instruction:

Make it widescreen 16 by 9 and put Elon Musk on his own flying saucer, flying towards us in the background.

In the older Venice chat, each of these meant clicking the plus, choosing edit image, dragging the file in, picking a model, describing the change, opening advanced settings, and generating. The agentic chat collapses all of that into the conversation, so you keep editing by just saying what to change.

From image to video, then extending the story

Once you have an image you like, ask the agent to animate it:

Turn it into a video, high speed interstellar chase. Me and the aliens are riding the spacecraft like a skateboard.

The agent defaults to a private video model and produced a 10-second clip with audio in the demo. To build a sequence, continue with a follow-up:

Generate a second shot to continue the story.

These examples are a starting point, not the limit. Venice notes the agentic chat will keep improving, with more tutorials and announcements posted on the Venice Twitter and YouTube channels.

Key takeaways

• The agentic chat runs multi-step workflows from a single prompt, no manual model or screen switching.
• Model choice sets your trade-off: private and subscription-included, faster but credit-spending, or frontier models with anonymized routing.
• Image generation defaults to zero-credit open-source models; enable premium models only when you want higher quality.
• The same chat carries you from research to image editing to video, and you steer date-sensitive results by stating the year explicitly.