ScreenScene
ScreenScene is a Hermes Agent plugin that turns a screenshot and a short list of narration beats into a narrated explainer video. For each beat it zooms the camera to a region of your screenshot, highlights it with an outline or a spotlight, and plays an ElevenLabs voiceover. Every camera move is sized to the length of its spoken beat, so the visuals and the voice stay in step.
How it works
- Hermes looks at the screenshot and chooses 3 to 6 beats. Each beat has a short narration and a focus region.
- The plugin sends each beat's text to ElevenLabs and measures the length of the returned audio.
- Manim renders a camera path over your real screenshot, with each move timed to its beat.
- ffmpeg combines the video and the narration into one MP4.
Tool
screenscene_render
| Parameter | Required | Description |
|---|---|---|
image_path | yes | Path to a PNG or JPG screenshot |
beats | yes | 1 to 8 beats. Each has narration (up to 500 characters) and an optional region as [x0, y0, x1, y1], fractions of the image from the top left. Use [0, 0, 1, 1] for a whole-image beat. Optional per beat: highlight (box, spotlight or none) and color (hex such as #2EC4B6). Colours rotate automatically if not given |
voice_id | no | An ElevenLabs voice your account can use. Default is a premade voice |
quality | no | standard (720p30, default), high (1080p30, sharper small text, slower) or draft (480p15, fastest) |
It returns the path of the finished video, its length in seconds, and the number of beats. Videos are saved in ~/screenscene_output/.
Example prompt: "Use screenscene_render on ~/shots/dashboard.png. Make 4 beats: an overview, then three zooms on the most important areas. Keep each narration short."
Requirements
- Manim Community Edition (
manim) on your PATH, or setSCREENSCENE_MANIMto its full path ffmpegandffprobeon your PATH (or setSCREENSCENE_FFMPEGandSCREENSCENE_FFPROBE)- An ElevenLabs API key with Text to Speech access, stored as
ELEVENLABS_API_KEY. Free accounts can use their own premade voices but not library voices through the API
Limitations
- Choosing regions needs a model that can look at images. If your model cannot, give the regions yourself in the prompt.
- A four beat video takes about one minute to render at the default quality.
highis slower. - Small text is only readable if the camera zooms in. Use tight regions around specific content, since full-width bands barely zoom.
- Region accuracy depends on how well the model locates parts of the image. Highlight boxes are only as precise as the coordinates provided.
- Narration uses ElevenLabs credits, about one call per beat.
Disclosures
- Network calls: HTTPS requests to the ElevenLabs API (
api.elevenlabs.io) for text to speech. The narration text of each beat is sent. The API key is read through Hermes secret handling and is never written to files or logs by this plugin. - Subprocesses: runs
manim,ffmpegandffprobewith fixed argument lists and timeouts. No shell is used. Child processes receive a minimal environment (PATH,HOME,LANGand a privateTMPDIR) and do not inherit Hermes provider keys. - Files: reads the screenshot path you give it. Writes temporary files to a private temp folder (deleted after each run) and the final MP4 to
~/screenscene_output/. - Background processes: none. Rendering runs as a bounded step during the tool call.
- Telemetry: none.
Installation
hermes plugins install solocreativeone/screenscene
hermes plugins enable screenscene
Then make sure ELEVENLABS_API_KEY is set for Hermes. To check a local copy:
hermes plugins doctor .
License
MIT License. See LICENSE for details.
README from the reviewed commit 111e781; it updates when the author re-pins.

solocreativeone