Gemini
Long context, the Google stack, and a quick image when I need one.
Where it has an edge
- Very long inputs, whole repositories or hour long transcripts
- Anything already sitting in Docs, Sheets or Drive
- Fast image generation
- Cheap bulk passes over a lot of text
How I prompt it
I use the context window properly: drop in the full document or transcript and ask for a specific extraction instead of a summary. Naming the output shape helps, such as a table with three columns, and saying which sections to ignore. For images I describe the scene, the framing and the style in that order, then ask for two or three variations.
Images
Yes. It is quick at clean, simple scenes. Text inside an image is still unreliable, so I keep words out of the picture and add them in the layout.
Where it slips
It sometimes answers from a summary of the document rather than the document itself. Ask it to quote the line it used and the problem usually shows up.