Local Model Vision

Some local GGUF models support image input when paired with a matching multimodal projector file. 3sparks Chat can download the model and projector together for supported local vision models.

How it works

1. Open Local Models.
2. Look for models marked as vision-capable.
3. Download the model.
4. The app downloads the matching projector file when required.
5. Load the model.
6. Add an image from camera, photos, or paste in chat.

Local and remote vision

Vision is not limited to Local Models. Remote providers such as OpenAI, Anthropic, Gemini, and OpenRouter can also support image input when the selected model has vision capability.

Use Images and Vision for the full image workflow.

What to expect

Local image chats stay on device, but vision models use more storage, memory, and processing time than text-only local models. Smaller text models may be faster, while local vision models trade speed for privacy.

Troubleshooting

• If an image is not accepted, confirm the loaded local model supports vision.
• If a projector download fails, delete the model and download it again.
• If the device runs out of memory, unload the model or choose a smaller vision model.
• If you need faster image analysis, use a remote vision-capable provider.