Released: November 2026 · All CNAPS Studio release notes
Release Overview
CNAPS Studio v1.2.4 brings a new Home and Agent, a chat assistant that finds, builds, and runs cards and flows for you. It also refreshes cards and the Community, adds new video upscalers, image + text models, and Audio/Video Loop tools, and makes it clearer why a run is waiting or failed.
✨ What's New
Major Improvements
1️⃣ New Home
Home has a new look with a prompt box in the center, a Ready to try tab, a new card grid, and a hero wall of Community posts.
2️⃣ Agent - Find, Build & Run by Chatting
Just tell Agent what you want. It finds or builds the right card or flow and runs it for you. Agent asks before it runs anything, lets you undo its edits, takes file attachments, and keeps your chat history.
New AI Models & Tools
1️⃣ AI Model - Video Upscaling (Pro plan)
- RelayVSR (Precision): Faithful 4× to 4K that keeps faces, text, and logos true.
- FlashVSR+ Turbo (Creative): The sharpest-looking 4K, with one-step diffusion.
- Kandinsky 6 SR ×2 (Precision): 1080p to 4K, faithful and low-flicker.
- Kandinsky 6 SR ×4 (Precision): 540p to 4K, faithful and low-flicker.
2️⃣ AI Model - Multimodal Language Models
- MiMo-V2.6-Distill-Qwen-9B: 9B agent model for code, tool use, and software screenshots.
- SenseNova-Loop-9B-Preview: Looped 9B vision-language model for deeper reasoning per token.
- MiniCPM-V 4.7 1B: Tiny vision-language model for fast, high-volume questions.
- MiniCPM-V 4.7 35B A3B: MoE vision-language model with 35B total and about 3B active.
3️⃣ AI Model - Image Generation (Iris-3B)
Pixel-space text-to-image with no VAE, driven by sentence prompts.
4️⃣ AI Model - Image Classification (PixAI Tagger)
Tags anime illustrations with 30,877 Danbooru tags.
5️⃣ Tool - Audio Loop & Video Loop
- Audio Loop: Turns music or ambience into a seamless loop with no click at the repeat, cut to the exact length you need (max, a fixed time, or a connected video's length). Pair it with ACE-Step for looping background music.
- Video Loop: Loops a clip with no visible jump, blending picture and sound together. Made for looping social posts, website hero videos, and animated profiles.
6️⃣ Model Updates
Omni now supports First & Last Frame. SAM 3.1 and QWEN get new inpainting options, and Whisper and SparkVSR handle inputs better.
7️⃣ VEO Node Retired
The VEO node has been retired. Use Omni instead.
Platform Improvements
1️⃣ Cards - Redesigned Detail Page & Chaining
Card detail pages have been reorganized and now offer multiple sample sets. Generate several candidates, pick the one you like, and continue from there, or send a result straight into the next card as its input.
2️⃣ Open Past Runs & Refill Inputs
Open any past run and fill its inputs back in with one click to run it again.
3️⃣ Community - New Design
The Community has a new design. Hit Try on any post to run it right away, and find your own posts and saved posts in your Library.
4️⃣ Clearer Run Status
See a countdown until your limit refills and the estimated wait time for your run. When a run fails or is rejected, you'll see why and what to do next. If you think a rejection was a mistake, you can report it.
5️⃣ Studio Updates
Video Compare is more powerful, and nodes now connect automatically. Notifications, Profile, and Settings screens have been redesigned.
6️⃣ Faster Loading & Better Emails
Home and Recipes load faster, and reminder emails for flows you haven't run yet have been redesigned.
7️⃣ Security & Reliability
Deleted accounts can no longer use their API keys or sign in to MCP, and we applied security updates to dependencies and images.
🔧 Bug Fixes
Fixed wrong colors in GLM-Image outputs, GPU workers getting stuck, and servers restarting during boot, along with other small fixes.
