SenseNova 6.8 Code2Video

SenseNova 6.8 Code2Video

Builds short videos by writing and rendering executable code, with a 27B model fine-tuned from Qwen3.8-27B. Suits motion designers, teachers and marketers who want video drafts from a prompt.

Made by sensenova on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: SenseNova 6.8 Code2Video is a 27B image-and-text model, fine-tuned from Qwen3.8-27B, that makes short videos by generating code and rendering it. The card says the same code can hand work to image, video, 3D and editing tools. Its focus today is short videos built entirely in code.

Who it suits: People who want a video draft produced from a written brief, such as motion designers, teachers and marketing teams. The card lists math explainers, data stories, character animation and product promos among its uses.

Industry | Who uses it | How they use it | Example
Education | Teachers and tutors | Turn a topic into an animated math explainer | A short explainer built from a prompt
Marketing | Brand teams | Draft a product promo as motion design | A promo clip with animated type
Media | Data writers | Build a data story with moving charts | A chart that animates into a short story

If you need long films or live footage, you can skip it.

Where to use it:

• Docker on a GPU server, served with LightLLM or vLLM
• An agent that follows the production skill on GitHub
• Windows, macOS or Linux, for downloading the files

Use case | Who | What they get back
Math explainer | Teachers | A short animated video made from code
Product promo | Marketers | A rendered clip with motion and type
Data story | Analysts | An animated video of charts and numbers

How to set it up:

• Python with pip, to download the files
• Docker with GPU access
• Several GPUs, since the card's commands split the model four ways
• About 54.7 GB of disk for the BF16 weights

1. Install the Hugging Face hub package.
2. Download the model to your computer.
3. Start the server in Docker, pointing the volume at the folder you downloaded.

pip install -U huggingface_hub
hf download sensenova/SenseNova-6.8-Code2Video

The card's vLLM command looks like this:

docker run --rm --gpus all --ipc=host -p 8000:8000 \
  -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 \
  -v /path/to/SenseNova-6.8-Code2Video:/model \
  vllm/vllm-openai:latest \
    --model /model \
    --tensor-parallel-size 4 \
    --max-model-len 524288 \
    --reasoning-parser qwen3 \
    --enable-auto-tool-choice \
    --tool-call-parser qwen3_coder \
    --host 0.0.0.0 --port 8000

How to use it: Once the server runs on port 8000, send it a video brief and let it write the code. The repo also holds example trajectories, and a production skill on GitHub supplies the workflow and rendering tools.

Tool | What it does
Model weights | The weights, config and tokenizer
Example trajectories | Logs of messages and tool calls, with figures
Production skill | A workflow, guides for making videos and rendering tools

Example: You ask for a short product promo. The model writes executable code for the scenes and motion, the code renders the frames, and you get a short video back.

For newcomers and experts: If you are new, think of it as a video maker that works through code. Start by reading the example trajectories to see how a brief turns into code, before setting up a GPU server.

Under the hood, it is a language model that writes programs, so any tool a program can call becomes part of the workflow. The trade-off is that quality depends on the code it produces and on the rendering tools around it. The card calls the results preliminary.

You write a video brief
The model generates executable code
The code renders frames and motion
The code may call other creative tools
A short video comes out

Good to know:

• Context: up to 512K tokens
• Its stated focus is short videos built entirely in code
• Editing, compositing, animation and 3D are named as later steps
• License: Apache 2.0
• Cost to run: no paid service named, but the card's commands use four GPUs
• Data it reaches: your prompts and the files downloaded from Hugging Face, plus any tools the code calls
• Not for: long films or work the card lists as future steps

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 27.4B
• Weights: about 54.7 GB in BF16; about 13.7 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,460
• Likes: 107
• License: apache-2.0
• Languages: en, zh
• Quantized versions: 6
• Fine-tunes: 1
• Tags: transformers, safetensors, qwen3_5, image-text-to-text, code-video, video-generation, agent, claude-code, qwen3.8, conversational

Links

• Model: https://huggingface.co/sensenova/SenseNova-6.8-Code2Video
• Files: https://huggingface.co/sensenova/SenseNova-6.8-Code2Video/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download sensenova/SenseNova-6.8-Code2Video

Then follow the usage in its model card. Nothing in this kit runs by itself.