ai-agent

Monet

Name: Monet
Author: NOVAglow646

[CVPR 2026] Official codes of "Monet: Reasoning in Latent Visual Space Beyond Image and Language"

219 stars10 forksPythonUpdated 3/19/2026100% free · open source

What it does

Monet lets you run reasoning tasks (e.g., visual question answering, image‑based commonsense inference) directly in the latent space of a pretrained vision model, without needing separate language prompts.

When to use it

•You need to answer complex questions about images that go beyond simple captioning (e.g., “Is the person likely to be late for work?”).
•You want to prototype a visual‑reasoning feature for a product demo without training a full multimodal model from scratch.
•You have a dataset of images and structured queries and need a fast, zero‑shot baseline to benchmark against.

Ready-to-paste prompt

python run_query.py --image data/raw/office.jpg --question "Is the person about to miss their meeting?"

Heads up: Monet requires a CUDA‑enabled GPU with PyTorch ≥1.13; on CPU it will raise a runtime error. Make sure your environment’s `torch` version matches the CUDA toolkit installed, and that you have at least 8 GB of GPU memory for the default CLIP‑ViT‑L/14 backbone.

Saves to your device

Use with Claude

New

Skip the builder — one click puts this in Claude, Cursor, Antigravity and more.

✅ Light setup

Installs with a command or two; your AI agent can do it for you.

Try it instantly — no install

Claude Code

mkdir -p ~/.claude/skills/novaglow646-monet && curl -fsSL https://workflowstacks.com/api/skills/novaglow646-monet/claude-skill -o ~/.claude/skills/novaglow646-monet/SKILL.md

Open in another AI app

Opens the app with this repo with the prompt ready to go — no copy-paste needed.

Connect the whole catalog (MCP)

claude mcp add --transport http workflowstacks https://workflowstacks.com/api/mcp

Adds a WorkflowStacks connector to Claude Code: search and load any skill here by chatting.

What's inside — free to inspect

No purchase needed

Read the entire source before you build — unlike paid marketplaces that hide it behind a buy button.

top-level files

folders

59.7M

source size

—

license

Key files

README.md

requirements.txt

File tree

deepspeed/

images/

inference/

monet_qwen_model/

RL/

script_examples/

src/

transformers/

.gitignore

README.md

requirements.txt

Use with Claude

New

Skip the builder — one click puts this in Claude, Cursor, Antigravity and more.

✅ Light setup

Installs with a command or two; your AI agent can do it for you.

Try it instantly — no install

Claude Code

mkdir -p ~/.claude/skills/novaglow646-monet && curl -fsSL https://workflowstacks.com/api/skills/novaglow646-monet/claude-skill -o ~/.claude/skills/novaglow646-monet/SKILL.md

Open in another AI app

Opens the app with this repo with the prompt ready to go — no copy-paste needed.

Connect the whole catalog (MCP)

claude mcp add --transport http workflowstacks https://workflowstacks.com/api/mcp

Adds a WorkflowStacks connector to Claude Code: search and load any skill here by chatting.

Quick Actions

Details

Creator

NOVAglow646

Language

Python