Skip to content
Computer vision

Shot Reader 8B

Video shot analysis into a ready 3D scene for Blender

2 min readupdated

The Shot Reader 8B model was created by Voyager based on Qwen3-VL-8B-Instruct (an 8.77 billion parameter multimodal language model) and is designed for reverse engineering video shots. It analyzes a continuous video segment and outputs a detailed JSON report: camera height, tilt angle, focal length, movement, as well as the positions, sizes, and trajectories of objects and environment elements. The resulting data makes it possible to directly recreate the scene in Blender or pass it to a coding agent. The new model outperforms major closed models in camera parameter estimation accuracy on test benchmarks and runs locally on consumer graphics cards.

Key features

  • Purpose: reading camera parameters, object positions, and movement from video footage.
  • Camera accuracy: outperforms several large models in determining camera height and movement trajectory.
  • Efficiency: optimized to run locally or via vLLM on a single accelerator.
  • Output format: a structured JSON document with shot specifications, timings, and coordinates in meters.

What you can do

  • Pre-visualization: quickly obtain editable 3D blockouts from reference video clips for setting up lighting and angles.
  • Controlled AI video: use camera parameters and trajectories as precise control inputs for generative pipelines.
  • Library analysis: massively tag video archives by camera movement types, angles, and mise-en-scène.
  • Scene generation: pass the ready JSON descriptor to scenographic agents for automated environment assembly.

Example

Running a video shot processing script with specified target objects to track:

python shot_reader.py clip.mp4 \
  --subject "P01: person (woman in a black polo)" \
  --subject "O01: dog"
Read moreDetails
Metrics (on 300 test shots): camera height and movement accuracy is 69% (compared to 44–46% for competitors). Processing time per shot is 5.9 sec on an H100 (or 0.76 sec in vLLM batch mode).
Hardware: tested on an RTX 5080 (16 GB VRAM) using 4-bit quantization (about 25 seconds per shot including model loading).
Limitations: real-world video events and movement trajectories are more challenging for the model; strictly processes one continuous shot at a time.
Training: fine-tuned (LoRA r64 on attention and MLP layers) on synthetic data from procedural 3D scenes and annotated archival frames.
Video example
Discuss Telegram 61 417

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 16 GB · ready files18 GB
  • FP8 / INT8recommendedweights 8.2 GB9.5 GB
  • 4-bitweights 4.6 GB5.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 3080 · BF16 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # voyager-so/shot-reader-8b
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download voyager-so/shot-reader-8b --local-dir ~/models/shot-reader-8b
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to shot-reader-8b
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Qwen3-VL-8B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 4All on Hugging Face
Merges 1All on Hugging Face

About the model

voyager-so/shot-reader-8b
Author
voyager-so
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

41761

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments