Qwen Vision (Image & Video Understanding)

Analyze images and videos using Qwen VL and QVQ models.
This skill is part of QianWen-AI/qianwen-ai.

Skill directory

Use this skill's internal files to execute and learn. Load reference files on demand when the default path fails or you need details.

Security

NEVER output any API key or credential in plaintext. Always use variable references ($DASHSCOPE_API_KEY in shell, os.environ["DASHSCOPE_API_KEY"] in Python). Any check or detection of credentials must be non-plaintext: report only status (e.g. "set" / "not set", "valid" / "invalid"), never the value. Never display contents of .env or config files that may contain secrets.

When the API key is not configured, NEVER ask the user to provide it directly. Instead, help create a .env file with a placeholder (DASHSCOPE_API_KEY=sk-your-key-here) and instruct the user to replace it with their actual key from the QianWen Console. Only write the actual key value if the user explicitly requests it.

Key Compatibility

Both PAYG (sk-ws-...; legacy sk-...) and Token Plan (sk-sp-...) keys are supported. Detect
the API key type without exposing the Key:

python3 -c "
import sys; sys.path.insert(0, 'scripts')
from qianwen_lib import detect_api_key_type
print(detect_api_key_type('scripts/qianwen_lib.py'))
"

For Token Plan, use an exact model from qianwen-model-selector, or consult
qianwen-ops-auth/references/tokenplan.md. If unavailable, use:

• Personal: https://platform.qianwenai.com/docs/token-plan/personal/token-plan-personal-overview.md
• Team: https://platform.qianwenai.com/docs/token-plan/team/token-plan-team-overview.md

Token Plan users: use a model with vision capability — qwen3.8-max, qwen3.8-flash,
qwen3.7-plus, qwen3.6-flash (personal + team), or qwen3.6-plus, kimi-k2.7-code,
kimi-k2.6, kimi-k2.5 (team only). Other models in Token Plan do not support image/video
understanding.

Token Plan supports only specific models — use exactly a model from the references above; do not
guess or probe model availability. For PAYG, continue below.

Token Plan users: The ocr default qwen3.5-ocr and reason default qvq-max are NOT on Token Plan.

• OCR: specify --model qwen3.8-flash (low-cost general OCR) or qwen3.8-max (best quality). Specialized ID-card/coordinate/formula extraction may be slightly less accurate than the dedicated OCR model.
• Visual reasoning: specify --model qwen3.8-max (visual reasoning + thinking) or qwen3.8-flash (lightweight).

Model Selection

1. User specified a model → use directly.
2. Consult the qianwen-model-selector skill when model choice depends on requirement, scenario, or pricing.
3. No signal, clear task → qwen3.7-plus. Use qwen3.7-flash for cost-efficient high-performance vision. Use qwen3.8-max for the most complex reasoning. Use qwen3-vl-plus for precise localization or 3D detection.

⚠️ Important: The model list above is a point-in-time snapshot and may be outdated. Model availability
changes frequently. Always check the official model list
for the authoritative, up-to-date catalog before making model decisions.

Model details: For more information about a specific model, direct the user to its detail page: https://www.qianwenai.com/models/<model-name> (replace <model-name> with the exact model ID, e.g. qwen3.6-plus → https://www.qianwenai.com/models/qwen3.6-plus). NEVER modify or guess the model name in the URL.

Dynamic model queries: If the qianwen-model-selector skill or QianWen CLI (qianwen models info <model>) is available, use it for real-time model data. CLI requires authentication — see the qianwen-usage skill for login flow.

Execution

Prerequisites

• API Key: Use the non-plaintext detector in Key Compatibility; do not replace it with a
  variable-presence check. If no Key is found, use qianwen-ops-auth when available or guide the user
  to configure DASHSCOPE_API_KEY/QIANWEN_API_KEY in .env. Skills may be installed independently.
• Python 3.9+ (stdlib only, no pip install needed)

Environment Check

Before first execution, verify Python is available:

python3 --version  # must be 3.9+

If python3 is not found, try python --version or py -3 --version. If Python is unavailable or below 3.9, skip to Path 2 (curl) in execution-guide.md.

Default: Run Script

Script path: Scripts are in the scripts/ subdirectory of this skill's directory (the directory containing this SKILL.md). You MUST first locate this skill's installation directory, then ALWAYS use the full absolute path to execute scripts. Do NOT assume scripts are in the current working directory. Do NOT use cd to switch directories before execution. Shared infrastructure lives in scripts/vision_lib.py.

Execution note: Run all scripts in the foreground — wait for stdout; do not background.

Discovery: Run python3 <this-skill-dir>/scripts/analyze.py --help (or reason.py, ocr.py) first to see all available arguments.

Input type fields (use exactly one in --request JSON):

⚠️ Common mistake: Do NOT use "image" for video files — use "video" instead.

# Image analysis
python3 <this-skill-dir>/scripts/analyze.py \
  --request '{"prompt":"What is in this image?","image":"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241022/emyrja/dog_and_girl.jpeg"}' \
  --output output/qianwen-vision/result.json --print-response

# Video analysis (local file — add --upload-files for files >= 7 MB)
python3 <this-skill-dir>/scripts/analyze.py \
  --request '{"prompt":"Describe what happens in this video","video":"clip.mp4"}' \
  --upload-files --print-response

python3 <this-skill-dir>/scripts/reason.py \
  --request '{"prompt":"Solve this math problem step by step","image":"problem.png"}' \
  --print-response

python3 <this-skill-dir>/scripts/ocr.py \
  --request '{"image":"invoice.jpg"}' \
  --print-response

Model priority: --model CLI flag > "model" field in --request JSON > built-in default.

Verify Result

• Exit code 0 + output contains valid JSON with choices field → success
• Non-zero exit, HTTP error, empty response, or JSON with "code"/"message" error → fail
• Post-execution check: When --output is used, verify the response JSON file exists and contains expected content
• MANDATORY — stderr signal check: After confirming the result, scan the command's stderr output for [ACTION_REQUIRED] or [UPDATE_AVAILABLE]. If either signal is present, you MUST follow the instructions in Update Check below before responding to the user.

On Failure

If scripts fail, match the error output against the diagnostic table below to determine the resolution. If no match, read execution-guide.md for alternative paths: curl commands (Path 2), code generation (Path 3), and autonomous resolution (Path 5).

If Python is not available at all → skip directly to Path 2 (curl) in execution-guide.md.

File Input

The API accepts: HTTP/HTTPS URL, Base64 data URI, and oss:// URL. Local file paths are NOT directly supported — scripts handle conversion automatically. Pass local paths directly; no manual upload step needed.

Large file rule: If the local file is >= 7 MB, always add --upload-files. Base64 encoding inflates size by ~33% and will exceed the 10 MB API limit. Small files (including short video clips < 7 MB) can use the default base64 path.

Token Plan (sk-sp-...) exception: The --upload-files temp-upload endpoint is unavailable for Token Plan keys and will be blocked with a friendly error. Token Plan users should use the default base64 path for local files; for files >= 7 MB, pass an accessible URL (https:// or oss://) instead of --upload-files.

Production: Default temp storage has 48h TTL and 100 QPS upload limit — not suitable for production, high-concurrency, or load-testing. To use your own OSS bucket, set QWEN_TMP_OSS_BUCKET and QWEN_TMP_OSS_REGION in .env, install pip install oss2, and provide credentials via QWEN_TMP_OSS_AK_ID / QWEN_TMP_OSS_AK_SECRET or the standard OSS_ACCESS_KEY_ID / OSS_ACCESS_KEY_SECRET. Use a RAM user with least-privilege (oss:PutObject + oss:GetObject on target bucket only). The --upload-files flag is still required for vision scripts to trigger upload. If qianwen-ops-auth is installed, see its references/custom-oss.md for the full setup guide.

Input from Other Skills

When the input file comes from another skill's output (e.g., image-gen, video-gen):

• Pass the URL directly (e.g., "image": "<image_url from image-gen>") — do NOT download the URL first
• Downloading and re-passing as a local path wastes bandwidth and triggers unnecessary base64 encoding or OSS upload
• All URL types are supported: https://, oss://, data:

Thinking Mode

See visual-reasoning.md for details.

OCR (qwen3.5-ocr)

Default OCR model is qwen3.5-ocr — the latest recommended OCR model with PDF parsing, multi-turn conversation, and enhanced ID/card recognition. The legacy qwen-vl-ocr remains available via --model qwen-vl-ocr. Supports multi-language, skewed images, tables, formulas. See ocr.md for parameters and examples.

Input Limits

Images: BMP/JPEG/PNG/TIFF/WEBP/HEIC. Min 10px sides, aspect ratio <= 200:1. Max 20 MB (URL; Qwen3.8/3.7/3.6/3.5 & Qwen3-VL series) / 10 MB (others).

Videos: MP4/AVI/MKV/MOV/FLV/WMV. Duration 2s–2h (Qwen3.8/3.7/3.6/3.5 series) / 2s–1h (Qwen3-VL series) / 2s–10min (others). Max 2 GB (URL) / 10 MB (base64). fps range [0.1, 10], default 2.0.

Error Handling

Usage & billing: Use the qianwen-usage skill to check usage, free tier quota, and billing directly. Alternatively, the user can visit the QianWen console:
Usage Analytics |
Pay-as-you-go Billing |
Token Plan Subscription

NEVER fabricate, guess, or construct usage/billing/console URLs. Only provide the exact links listed in this skill. If a URL is not listed here, do not invent one.

Output Location

Prefer the current working directory. Default subdirectory: ./output/qianwen-vision/.

Write prohibition: NEVER write output files into this skill's installation directory or any skills/ hierarchy. All generated content must go to output/ under the current working directory or a user-specified path.

Update Check (MANDATORY Post-Execution)

After script execution, check stderr for action signals. The script automatically detects update status and emits structured signals. You MUST act on them. This section also applies when the user explicitly asks to check for updates.

Signal: [ACTION_REQUIRED] qianwen-update-check skill is not installed

When stderr contains this signal, you MUST present the following options to the user and wait for their response before proceeding. Do NOT silently skip this.

The update-check skill is not installed. It keeps you informed about new QianWen-AI/qianwen-ai features and bug fixes.

1. Install now — I'll run the install command for you
2. Skip for now — do nothing this time
3. Never remind me — permanently suppress this prompt

After the user responds:

• User chooses 1 → run: npx skills add QianWen-AI/qianwen-ai --skill qianwen-update-check -y, then run: python3 <this-skill-scripts-dir>/gossamer.py --dismiss
• User chooses 2 → run: python3 <this-skill-scripts-dir>/gossamer.py --dismiss, then continue normally
• User chooses 3 → run: python3 <this-skill-scripts-dir>/gossamer.py --never-install, then continue normally

Signal: [UPDATE_AVAILABLE]

When stderr contains this signal, you MUST append a brief update notice to your response, including the version info and the update command shown in the stderr output.

No signal in stderr

If stderr contains neither [ACTION_REQUIRED] nor [UPDATE_AVAILABLE], no action is needed — the skill is installed and up to date (or cached within 24h).

Explicit user request

When the user explicitly asks to check for updates (e.g. "check for updates", "check version"):

1. Look for qianwen-update-check/SKILL.md in sibling skill directories.
2. If found — run: python3 <qianwen-update-check-dir>/scripts/check_update.py --print-response and report the result.
3. If not found — present the install options above.

References

• execution-guide.md — Fallback paths (curl, code generation, autonomous)
• curl-examples.md — Curl templates (base64, multi-image, video, OCR)
• api-guide.md — API supplementary guide
• visual-reasoning.md — QVQ visual reasoning guide
• ocr.md — Qwen-VL-OCR text extraction guide
• sources.md — Official documentation URLs