MakeFigures

Computing & research methods Question, approach, finding

Vision–language feature fusion

Bring two encoding branches into a shared fusion module.

AI-generated conceptual examples for visual inspiration, not experimental results.

Reference prompt

Replace the example subjects, steps and findings with your research before generating.

Create a conceptual vision–language architecture. The upper branch runs from Image patches through Image encoder to Visual features; the lower branch runs from Text tokens through Text encoder to Text features. Connect both branches to Cross-attention and then Prediction head. Use a white background, blue and teal modules, clear directional arrows and restrained coral emphasis at fusion. Do not add layer counts, dataset sizes or performance values.
Use this promptUse as visual reference

Use the prompt to start with the example description, or use the image as a composition reference while keeping your own text.

Browsing and copying are free. Generation uses the credits shown in the maker.

Adapt it to your research

  1. Replace the branches with your actual input modalities.
  2. Use actual module names and connection directions.
  3. Keep architecture illustration separate from measured performance.
Browse all templates