Lens, train Moondreamto master logo detection.anything.

Production-ready vision AI in minutes. We handle the infrastructure so you can focus on what matters.

Start fine-tuning

Perfection in 3 steps

1

Bring examples.

Collect images for the task you want Moondream to learn.

2

Fine-tune.

Teach Moondream with SFT or RL. Pass your data to the API and we handle the rest.

3

Deploy.

Use your model through the API or run it locally with Photon.

What is fine-tuning?

Moondream starts as a general model trained on broad, public information. Fine-tuning makes it great at one specific task by teaching it the products, documents, categories, or internal information that matter to your business.

Who is this for?

This is for teams putting vision AI into production. If you already know the task and need the model to master that job, fine-tuning is how you get there. It is built for teams that need frontier performance at real-time speed.

Example fine-tunes

Basketball game frame for ball-handler detection

Player with Ball Detection

Detect the player holding the basketball in NBA broadcast footage. After fine-tuning, F1 jumps from 0.28 to 0.79 and false positives drop from 61 to 2.

F128.3%→78.8%

State Farm Logo Detection

Detect State Farm logos in NBA broadcast frames. The fine-tuned model reaches perfect F1 of 1.0 with zero false positives and zero missed logos, up from 0.38 on the base model.

Street view frame used for GeoGuessr country prediction

GeoGuessr Countries

Predict the country from a single street-view image. SFT fine-tuning takes accuracy from 28.6% to 71.1%, beating GPT-5.4 at 69.8%. Trained on 25 images per country.

Accuracy28.6%→71.1%

Hand gesture showing scissors

Rock Paper Scissors

Classify hand gestures as rock, paper, or scissors. RL fine-tuning with only 5 training examples per class reaches 98.8% accuracy from a 54.8% baseline. GPT-5.4 gets 100%.

Accuracy54.8%→98.8%

Retinal fundus image for glaucoma classification

Glaucoma Detection

Classify retinal images by glaucoma stage (normal, early, advanced). RL fine-tuning takes accuracy from 17.6% to 69.2%, more than double GPT-5.4 at 33.2%.

Accuracy17.6%→69.2%

3x3 grid of video frames for action description

Video Action Description

Given a 3x3 grid of frames taken from a video, describe the action. SFT fine-tuning raises description accuracy from 54% to 74%.

Accuracy0.5413→0.7391

Frequently asked questions

Ready to take Moondream to production?

Get started

Start with the docs and run your first experiment in a few API calls.

Start fine-tuning

Need help? We'll build it for you.

We can help define the task, prepare the data, run training, validate results, and hand off a model your team can use.