DeepSeek Harness says the current model does not support images. Should I change the configuration or the model?

This answer refers to the version in the original report. Check the controls and behavior in your own version.

First check the provider and exact model ID selected in DSH. The provider’s web app may use a different model or endpoint than the API route configured in Harness.

• The attachment is blocked before sending. In builds with this control, open Settings → Models, edit the provider, expand the model options, and check Input types. Enable Image after confirming that the endpoint accepts images. For manual configuration, pi-ai custom models use input: [text, image] and the direct DeepSeek adapter uses inputModalities: [text, image].
• The provider returns an error after sending. Test a small image with the same gateway and exact model ID. Check the provider’s required image format and size limits. If that endpoint accepts only text, switch to a vision-capable model. Retest in a fresh session if the old one already contains an image.

For a quick check, use a screenshot with a random number and ask the model to read it. If the request succeeds but the model still cannot read it, check whether the outgoing request retained the image. Settings labels can vary by version.

References

• Original user report
• DeepSeek Harness provider guide
• Site guide · Model catalogs and image input