Answer
Text can read supported images, videos, audio and files. Image and Video generation accept still-photo references; an uploaded video cannot be used as a generation reference. Audio accepts no attachments. Check the separate message and attachment limits for current counts and sizes.
Image mode uses replied photos before new attachments. Image and Video use replied photos without their caption, so repeat any needed instructions in your new prompt. Video uses only the first photo in a multi-photo reply and cannot combine an image-bearing reply with a new image. Remove the new photo or choose a compatible type.
GIF, Clip and location-only replies hide AI controls. Documents, audio and polls leave Text available and disable incompatible modes. Incompatible choices stay greyed out and preserve your inputs. Attachment-addition choices disappear at capacity. During uploads, typing remains available but @ai and the model picker cannot change.