Skip to main content

How to use dsh-deepseek-vision

Adds sticker support to text-only DeepSeek: registers a visual routing mechanism where images are first described as text by a VL model, then passed to DeepSeek—no model changes or official modifications required.

This article is auto-derived from indexed fields (wiki / faq / compatibility_json), not freshly AI-generated.

This article is derived from the plugin's already-indexed fields (wiki / faq / compatibility_json / readme), not freshly generated by AI. Source field is noted at the end of each section.

Quick start

dsh-deepseek-vision

— source: plugin_wiki.wiki_content

Install & verify

dsh plugin --profile web add dsh-deepseek-vision

Run the command above in your DSH Web Profile. Then enable the plugin in the plugin list.

— source: plugins.install

Key points

— source: plugin_wiki.readme_en (fallback readme_raw)

FAQ

What needs to be done after installation to paste images?

After installation, go to dsh Settings → Plugins → Plugin Config, open the "DeepSeek + Vision (Vision-Language Bridge)" card, fill in the VL model key (or export QWEN_VL_API_KEY in terminal), then select DeepSeek + Vision provider on the Models page, and paste images in the chat window.

What happens to image-containing historical sessions after uninstallation?

Image-containing sessions cannot switch back to text models—official rejection based on inputModalities is expected behavior; new sessions are unaffected, reinstall this plugin to restore.

What happens when VL model description fails?

Default is fail-closed: description failure causes the entire request to terminate with a stable error code (IMAGE_TOO_LARGE / AUTH / TIMEOUT / TRANSPORT, etc.), without silently dropping images; can be set to placeholder to degrade to text placeholder and continue.

Will the same image be billed repeatedly?

No. LRU caching is done in-process by attachmentId (default 64 items), retries, session compression, and subsequent rounds all reuse the same description text.

Must use qwen3-vl-flash?

No. Change vl.baseURL and vl.model in the settings card to use any OpenAI-compatible /chat/completions gateway, such as vLLM self-deployed, OpenRouter, LM Studio, etc.; for complex visual reasoning, switch to qwen-vl-max.

Can it be used in headless mode?

Yes. Gateway routing behaves consistently in both web and headless profiles; the visual settings card is only provided in web, headless can directly edit settings.yaml.

Is it locked to the official DSH version?

Not locked. Official CLI installation path directly installs the release; the non-CLI replicate path will rebuild on the target machine using the target machine's dsh, successful build proves compatibility.

— source: plugin_wiki.faq_json

Compatibility

  • DSH: >=0.1.0-rc.6(兼容 rc.6 与 rc.7;构建锚点 0.1.0-rc.8 出处声明)
  • Node: >=22.19.0

— source: plugin_wiki.compatibility_json

Pitfalls

Review the upstream repo before installing. This guide is auto-derived from indexed fields and may lag the latest release. If anything contradicts the official docs, treat the upstream source as authoritative.

— source: general rule