Skip to main content

AI analysis

Serpent’s AI analysis connects to a cloud AI service that you provide. Prepare a provider account, API key, a vision-capable model, and check the provider’s pricing first.

Configure AI

On Windows open Main menu → Settings → AI; on macOS open Settings → AI from the system menu.

AI configuration

Connect a provider

  • API format: choose the provider interface, such as DashScope Multimodal (native), OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, or Gemini Native. With a relay, follow the relay’s interface instructions.
  • Base URL: usually leave it empty. For a relay or self-hosted gateway, enter the address supplied by that service. DashScope native uses an /api/v1 root.
  • API Key: paste the provider key. The eye icon shows it temporarily; after saving, the full key is not displayed. Enter a new key to replace it.
  • Model: enter the provider’s vision model name. With a valid key, the picker can load models; if that fails, enter the name manually.
  • Language: choose one language—Chinese, English, Japanese, or Korean—for descriptions and tags.
  • Click Test connection, then Save after it succeeds.

Never put an API key in a script, plugin, screenshot, issue report, or MCP configuration.

AI write toggles

  • Description: describe visual content, style, and mood.
  • Tags: generate keywords for type, style, subject, and visual characteristics.
  • AI rating: generate a 1–5 reference score.
  • Force existing tags: choose only from tags already in the library.

AI content is stored separately from manual content; manual descriptions, tags, and ratings are not overwritten.

Advanced settings

  • Concurrent analysis limit: default 16, range 1–32; lower it if the provider rate-limits requests.
  • Maximum image edge: default 2048 px, range 512–4096; images are reduced when needed and never enlarged.
  • Maximum tags: default 8, range 1–32.
  • Chinese description limit: default 100 characters, range 20–500.
  • English description limit: default 60 words, range 10–200.
  • Output style: Normal, Concise, or Rigorous.
  • Rating rubric: define what scores from 1 to 5 mean.
  • Custom description prompt: add description guidance such as “focus on composition and lighting”.
  • Custom tag prompt: define the tag vocabulary. Never put a key or secret in a prompt.

Automatic analysis

Accept the disclaimer and enable Automatically run AI analysis on new assets to analyze new images, videos, and 3D models in the background; it is off by default and turning it off does not remove existing AI content.

Supported assets

  • Images and camera RAW: a resized image is sent.
  • Video: a timestamped contact sheet is sent, not the original video.
  • 3D models: a four-view image is sent.
  • Audio, text, and unsupported formats: not analyzed by AI.

Manual analysis and results

Choose AI analysis from an asset context menu or use the batch action for a multi-selection. Analyze unanalyzed assets skips assets that already have results; use normal AI analysis to re-analyze one.

Start AI analysis from the asset context menu

Results appear in the AI section of the asset Inspector. A successful re-analysis replaces the current AI result and does not keep history; clearing AI content does not remove manual content or tag entities.

Jobs and failures

Open Window → Background jobs to see queued, running, paused, failed, and completed AI jobs. You can pause, resume, cancel, or retry failed items. Network, rate-limit, and timeout failures are usually retryable; configuration, permission, quota, and unsupported-format problems must be fixed first.

If a video contact sheet or thumbnail is not ready, retry media generation from Background jobs before retrying AI. Failure messages include a short reason.

Privacy and cost

Serpent does not upload an entire library or provide AI search. Only explicitly submitted, supported assets are sent to the selected provider; check its privacy terms and pricing before enabling AI.