qwen-3-5-plus-video-feedback

Qwen 3.5 Plus Video Feedback

Overview

Analyze videos with OpenRouter using the qwen/qwen3.5-plus-02-15 model. Designed as the feedback loop for Veo 3.1 video generation — pass a generated video and a prompt to get structured analysis and quality feedback.

Analyze a public video URL

uv run {baseDir}/scripts/analyze_video.py \
  --video-url "https://example.com/generated.mp4" \
  --prompt "Does this video match the description: a cat jumping over a fence? Rate quality 1-10."

Analyze a local video file (auto-uploaded to S3)

uv run {baseDir}/scripts/analyze_video.py \
  --video-file /path/to/video.mp4 \
  --prompt "Describe what is happening in this video and assess its overall visual quality."

Save analysis to a file

uv run {baseDir}/scripts/analyze_video.py \
  --video-url "https://example.com/clip.mp4" \
  --prompt "Evaluate prompt adherence and suggest improvements." \
  --output feedback.txt

Arguments

Argument	Required	Description
`--prompt`	Yes	Question or instruction for the model
`--video-url`	One of these	Public or presigned URL to the video
`--video-file`	One of these	Path to a local video file (uploaded to S3, then analyzed via presigned URL)
`--output`	No	File path to save the analysis text

--video-url and --video-file are mutually exclusive.

S3 Configuration (required for `--video-file`)

Local files are uploaded to an S3-compatible bucket and accessed via a presigned URL (valid 1 hour). Set these environment variables:

Variable	Required	Description
`S3_ENDPOINT_URL`	Yes	e.g. `https://s3.us-east-1.amazonaws.com` or `https://nyc3.digitaloceanspaces.com`
`S3_ACCESS_KEY_ID`	Yes	Access key ID
`S3_SECRET_ACCESS_KEY`	Yes	Secret access key
`S3_BUCKET`	Yes	Bucket name
`S3_REGION`	No	Region (default: `us-east-1`)
`S3_KEY_PREFIX`	No	Key prefix for uploads (default: `video-analysis/`)

The bucket does not need to be public — presigned URLs are used.

Supported video formats

video/mp4, video/mpeg, video/mov, video/webm

System prompt customization

The skill reads an optional system prompt from assets/SYSTEM_TEMPLATE. Edit it to tune the model's feedback style (scoring rubric, artifact checklist, output format).

Behavior

When --video-file is used: uploads the file to S3, generates a presigned URL, passes it to the model. Upload progress is printed to stderr.
Prints the full analysis to stdout.
If --output is provided, also writes the text to that file and prints Analysis saved to: <path>.

Chaining with veo3-1

VIDEO_URL=$(uv run veo3-1/scripts/generate_video.py --prompt "..." --url-only | grep '^VIDEO_URL:' | cut -d' ' -f2)
uv run {baseDir}/scripts/analyze_video.py \
  --video-url "$VIDEO_URL" \
  --prompt "Evaluate prompt adherence, list artifacts, rate 1-10."

Troubleshooting

Symptom	Resolution
`OPENROUTER_API_KEY is not set`	`export OPENROUTER_API_KEY="sk-or-..."`
`S3_ENDPOINT_URL is not set`	Set S3 env vars (see S3 Configuration above)
`Video file not found`	Check the path passed to `--video-file`
`AuthenticationError` / HTTP 401	Key invalid or out of credits. Check https://openrouter.ai/settings/keys.
`Missing Content-Length` / upstream error	The provider doesn't support base64 data URLs — use `--video-file` (S3 upload) or `--video-url` with a direct HTTP URL.
`uv: command not found`	macOS/Linux: `curl -LsSf https://astral.sh/uv/install.sh \| sh`. Restart terminal.

For transient errors (HTTP 429, network timeouts), retry once after 30 seconds. Do not retry the same error more than twice — surface it to the user instead.