The DeepSeek-v4-flash-vision-exp model accepts images alongside text, enabling description, OCR, and chart analysis. Supported formats include JPEG, PNG, GIF, and WebP, detected from file content.

Three methods provide images via the OpenAI-compatible Chat Completions API (base URL: https://api.deepseek.com). First, base64-encoded inline images embed directly in requests, counting toward the 48 MiB request body limit. Second, external HTTPS URLs (max 8192 characters) download images up to 32 MiB within 60 seconds. Third, the Files API uploads images once (up to 64 MiB per file) and references them by `file_id`, ideal for reuse or large payloads.

A `detail` parameter (`low`, `high`, `original`, `auto`) controls preprocessing for URL inputs. Images are resized to roughly 800×800 equivalent, capping at 384 tokens per image. Limits include 600 images per request, 64 MiB total without file IDs (200 MiB with), and 8192 px max dimension.

The model also works with Anthropic-compatible `/messages` endpoint and OpenAI Responses API, using analogous content block shapes. Images are only allowed in user messages; other models return errors.