Updated monthly

Changelog

New models, features, and API changes shipped to Venice each month.

Changelog

February 2025

Models

18 changes

Deepseek Coder V2 Lite
Released Deepseek Coder V2 Lite in beta in both the API and the UI for feedback from beta testers.
Beta
Mistral Codestral 22B
Released Mistral Codestral 22B in beta in both the API and the UI for feedback from beta testers.
Beta
Qwen 2.5 VL 72B
Multi-modal model support for Qwen 2.5 VL 72B (HuggingFace) has been released for beta users in both the API and the UI.
Beta
Qwen 2.5 VL 72B
Qwen 2.5 VL 72B, a multi-modal vision model, is now live for Venice Pro users. It can describe images and generate prompts for use with Venice's image models.
Pro users
70b Parameter Models
All 70b parameter models have had their context lengths extended on both the Venice.ai app and API.
All users
Deep Seek
Improved context handling for Deepseek to enhance chat results within the Venice UI.
All users

Lustify SDXL
Lustify SDXL, the number one requested image model on Featurebase, is now in production and available via API for Pro users.
Pro users
Stable Diffusion 3.5
Set maximum steps on Stable Diffusion 3.5 to 30 to reduce image generation timeouts.
All users

Upscaler
The upscaler has been migrated to real esrgan with further enhancements planned over the coming weeks.
All users

87 changes

Control-Enter to Send Chat
When Enter Submits Chat is disabled, users can now send messages using Control-Enter.
Disable Venice System Prompt
Added a toggle in Text Settings > Advanced Settings to disable Venice's system prompts. When enabled, only user-enabled system prompts are sent to the model.
Include Filename in Context
Uploaded file filenames are now included in the context sent to the LLM, allowing the model to reference them in responses.
RTF Document Support
Rich Text Format (RTF) documents are now supported for upload in the chat interface.
Character Approval Process
A material set of improvements has been made to the character approval process to reduce delays and improve moderation accuracy.
Characters Search
A new search and filtering mechanism has been added to make finding characters easier.

GIF Upload Guard
Added a guard to prevent GIFs from being uploaded to in-painting.
Image Inpainting Beta
Image inpainting allows editing of generated or uploaded images using text prompts. An LLM interprets the change and a segmentation model creates a mask for inpainting.
Beta
Image Variant Mode
Beta users can now generate 1 to 4 images using the same prompt to get varied responses.
Beta
Inpaint Upload and UI Button
Users can upload an image and describe changes, or use the 'Inpaint' button under existing images to edit them.
Inpainting Blend Steps
Additional inpainting blend steps have been added based on user requests.
Paste Image for Inpainting
Users can now paste an image directly into the chat prompt box to perform inpainting.

Public Character Sync
Public characters are now synced back to the browser after cache clear. A manual sync button has been added for user control.

API Page Tier Indicator
The API page now shows which tier your account is operating under, such as explorer or paid.
Encrypted Backups Beta
Encrypted backups have been released in Beta to the Venice Beta Tester group. Interested testers can reach out to moderators in the #product channel on Discord.
Beta
Persistent Storage Option
Added an option in User Settings to request the browser make storage persistent, preventing browsers from clearing old history to free up space.
Safe Venice Mode
Pro users can now enable Safe Venice mode at the account level, which hides adult classified characters and screens adult content in image generation. Disabling requires a 4-digit PIN.
Pro users
Token Dashboard
Provides a guide for users who have staked their first VVV on how to begin using their issued Venice Compute Units (VCU).
Beta
Profile Endpoint Performance
Improved performance of the user profile endpoint, resulting in faster initial load times.

$VVV on dydx
$VVV has been listed on dydx for trading.
Active Stakers Metric
The Token Dashboard now includes 'Active Stakers' in the Staking chart, defined as VVV holders with staked balances who made an API call in the last 7 days.
Circulating Supply Info
Added circulating supply and the percentage of circulating supply that is staked (34%) to the supply chart on the token dashboard.
Claim and Restake Button
Added a 'Claim and Restake' button to allow users to claim and immediately restake rewards in a single transaction.
Coinbase On-Ramp Support
Added Coinbase On-Ramp support to purchase up to $500 in VVV with a debit card. Also allows purchase of $5 in ETH on Base if gas is insufficient.
Configurable Gas Fees
Users can now configure gas fees when interacting with blockchain contracts to handle network congestion.

Akash Docker POC
Released a proof-of-concept Docker container for users wishing to run Eliza with Venice on Akash. Full documentation will follow in future iterations.
Akash Eliza Agent Template
A template for Akash allows users to deploy a fully configured Eliza Agent that connects to Venice’s API without requiring direct Eliza knowledge.
API Docs Overhaul
The API Reference documentation has been completely updated to match the current API spec, including complete examples and properly referenced IDs. Newly added endpoints previously only in Postman are now documented.
Clock Sync Error Warning
Added a user-friendly error if the user's clock is out of sync and session tokens cannot be validated.
Disable Telemetry Setting
Added a setting in App Settings to disable all browser-based product usage telemetry collected by Venice.
Dune Analytics Dashboard
Created a Dune Analytics dashboard to show network utilization over time as recorded on-chain.

51 changes

/image/generate request id
Added a request id field to the image generation JSON response. API docs are updated.
Inpainting Support
Added inpainting support to the API in beta form. The API allows using Venice.ai's LLM to calculate the inpainting mask via the prompt parameter or specifying a custom mask via the mask object.
Image Generation Seed Return
If no seed was provided, the API will now return the seed used for the request in the response.
Image Upscale Failure Details
Image upscale failures now return details of the failure in the INVALID_REQUEST body.
/image/generate hide_watermark docs
Updated API documentation to note that Venice may ignore the hide_watermark parameter for certain generated content.
/image/generate seed randomness
Fixed an issue with the seed parameter on image generation not being fully random on every request.

23 changes

Character Context Order
Characters with context documents will now properly inject that context in the proper order when starting new chats through the web app. Character chats through the API were unaffected by this bug.
Chat
Character UI Bugs
Fixed a number of UI bugs that were preventing users from accessing the latest version of their public characters.
Chat
Code Block Copy Button
Fixed a bug that prevented the Copy option on code blocks from being clickable until the full message rendered.
Chat
Conversation Scroll
Loading conversations with many images should now properly scroll to the bottom of the thread.
Chat
Copy Button Excludes <tool_call><>
When copying Deepseek-generated messages via the UI copy button, the <tool_call><> block is excluded from the clipboard content.
Chat
Multi-modal Vision Model Inference Failures
Fixed an issue where users using multi-modal vision models may have seen failed inference after multiple requests.
Chat

Share feature requests and vote on ideas from the community on Learn Venice.