video-thumbnail-creator 1.6.1

CLI tool to extract and select thumbnail images from video files

video-thumbnail-creator

CLI tool to extract and select a thumbnail (poster) image from a video or image file. Supports manual selection, fully automatic AI selection (Claude Vision), and a semi-automatic suggest mode where the AI proposes a frame and you confirm.

Every generated JPEG has creation metadata (EXIF) embedded so that the exact parameters used during generation are preserved and can be read back with the info subcommand.

Note: This tool extracts and saves a still image only. Embedding the image into the video file (e.g. with AtomicParsley) is out of scope and must be handled by the caller.

---

Requirements

• Python 3.10+
• ffmpeg and ffprobe available in $PATH (required for video input; not needed for image-only input)
• macOS only: sips — built-in macOS image tool, used for wide-gamut color space conversion (TIFF/HEIC with Rec.2020 or Display P3). On Linux/Windows, Pillow is used as a fallback but may not handle wide-gamut images correctly.
• An Anthropic API key for auto / suggest modes (set via CLAUDE_API_KEY env variable or video-thumbnail-creator config set claude.api_key <key>)

---

Installation

pip install video-thumbnail-creator

Or in editable/development mode:

pip install -e .

---

Usage

video-thumbnail-creator extract <input_path> [OPTIONS]

Options

Option	Description
--mode manual	Interactive: open mosaic, enter frame number 0–19
--mode auto	Fully automatic: AI selects the best frame
--mode suggest	AI suggests a frame; you confirm or override
--format poster	Output format 2:3 (1080×1620) with 1:1 crop + text area (default)
--format landscape	Output format 16:9 (1920×1080) — existing behaviour
--embedded-image prefer	Use embedded cover art or sidecar image if present; otherwise extract frames
--embedded-image ignore	Always extract frames (ignore any embedded cover art or sidecar images)
--embedded-image ask	Prompt user when embedded cover art or sidecar image is found (default)
--crop-position POSITION	Set crop position directly (left, center-left, center, center-right, right); skips interactive prompt and AI crop selection
--overlay-title TEXT	Title text to overlay on the output image
--overlay-title-from-filename	Use the input filename stem as overlay title
--overlay-category TEXT	Category label shown above the title (poster format only)
--overlay-category-logo PATH	PNG logo shown instead of category text (poster format only); wide logos centered above title, square/portrait logos to the left
--overlay-note TEXT	Small text centered at the bottom of the poster text area (poster format only)
--style NAME	Poster style to use (default: internet). Run video-thumbnail-creator styles to list available options
--description TEXT	Optional video description for AI context (max 1000 chars)
--output-dir PATH	Output directory (default: same directory as the video)
--output-name-suffix SUFFIX	Suffix appended to the video filename stem (default: -poster)
--json	Emit machine-readable JSON to stdout
--no-badges	Disable automatic technical badges (4K, HD, HDR) on the poster
--fanart	Generate an additional clean 16:9 fanart image (for Infuse/Emby) with -fanart suffix

---

info Subcommand

Read and display the creation metadata embedded in a generated poster image:

video-thumbnail-creator info /path/to/poster.jpg

Default output:

Poster Metadata:
  Version            1.3.0
  Source             frame
  Frame Index        12
  Crop Position      center-left
  Format             poster
  Mode               auto
  Input File         2025-11-01_Herbst-Spaziergang.mp4
  Overlay Title      Herbst-Spaziergang
  Category           Videoschnittstudio Silvan Kurmann
  Note               1. November 2025
  AI Reasoning       Sharp, well-lit frame with child running towards camera…
  Created            2026-02-26T14:30:00

JSON output (--json):

video-thumbnail-creator info --json /path/to/poster.jpg

{
  "vtc_version": "1.3.0",
  "source": "frame",
  "frame_index": 12,
  "crop_position": "center-left",
  "format": "poster",
  "mode": "auto",
  "input_file": "2025-11-01_Herbst-Spaziergang.mp4",
  "overlay_title": "Herbst-Spaziergang",
  "overlay_category": "Videoschnittstudio Silvan Kurmann",
  "overlay_note": "1. November 2025",
  "ai_reasoning": "Sharp, well-lit frame with child running towards camera...",
  "created_at": "2026-02-26T14:30:00"
}

The embedded metadata enables future re-generation of posters (e.g. with a new template) without needing AI calls or interactive prompts.

---

Poster Format (2:3)

The default poster format produces a 1080×1620 image composed of two sections:

┌──────────────────┐
│                  │
│   1:1 crop of    │  ← 1080×1080 square crop (with subtle vignette)
│   selected frame │
│                  │
├──────────────────┤  ← 10px separator line (#2a2a2a)
│▓  [category]    ▓│  ← optional category text or logo above title
│▓                ▓│
│▓    Title       ▓│  ← bold, auto-sized (40–72px), centered, drop shadow
│▓                ▓│
│▓         note   ▓│  ← optional small note text at bottom-right
│▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓│  ← solid dark background (#1a1a1a)
└──────────────────┘
        2:3

• Top section (1080×1080): A 1:1 square crop of the selected high-res frame with a subtle radial vignette effect (15–20% edge darkening). The horizontal crop position (left / center-left / center / center-right / right) can be set directly with --crop-position, chosen by AI in auto/suggest mode, or prompted from the user in manual mode.
• Bottom section: The text area fills the entire bottom section edge-to-edge. Layout details depend on the active style (see Poster Styles):
  • Title (optional --overlay-title): bold, auto-sized, word-wrapped, centered, white with drop shadow.
  • Note (optional --overlay-note): small text centered at the bottom of the text area.
  • Category / Logo (optional): placement varies by style — in the text area above the title, or as an overlay band at the top of the image.

---

Supported Image Formats

When <input_path> points to an image file, the mosaic/frame-extraction pipeline is skipped and the image is used directly as the high-res source for crop-position selection and poster/landscape composition.

Format	Notes
JPEG (.jpg, .jpeg)	Used directly (copied as-is for sRGB images)
PNG (.png)	Converted to JPEG via Pillow
TIFF (.tiff, .tif)	Wide-gamut (Rec.2020, P3) converted to sRGB via sips (macOS)
HEIC/HEIF (.heic, .heif)	Converted to sRGB JPEG via sips (macOS)

Note: Wide-gamut color space conversion (Rec.2020, Display P3) requires macOS with sips. On other systems, Pillow is used as a fallback but may not handle wide-gamut TIFF files correctly.

---

Examples

Manual mode

video-thumbnail-creator extract /path/to/video.mp4 --mode manual

The mosaic of 20 frames is opened in the system image viewer. Enter the frame number (0–19) at the prompt. In poster format, you are also prompted to choose a crop position. The result is saved next to the video.

Manual mode — landscape with text overlay

video-thumbnail-creator extract /path/to/video.mp4 \
  --mode manual \
  --format landscape \
  --overlay-title-from-filename

Automatic AI mode (poster, default format)

video-thumbnail-creator extract /path/to/video.mp4 \
  --mode auto \
  --description "Documentary about rocket launches" \
  --overlay-title "2025-05-15 – Starship IFT-7"

The AI selects the best frame from the 20-frame mosaic, then chooses the optimal 1:1 crop position for the poster, and renders the text in the blurred bottom area.

Poster with category, logo, and note

# Category text above title, note at bottom-right
video-thumbnail-creator extract /path/to/video.mp4 \
  --mode auto \
  --overlay-title "Starship IFT-7" \
  --overlay-category "Space Exploration" \
  --overlay-note "2025-05-15"

# Category logo (wide, centered above title)
video-thumbnail-creator extract /path/to/video.mp4 \
  --mode auto \
  --overlay-title "Starship IFT-7" \
  --overlay-category-logo /path/to/channel-logo-wide.png \
  --overlay-note "Episode 7"

# Square/portrait logo (left of title)
video-thumbnail-creator extract /path/to/video.mp4 \
  --mode manual \
  --overlay-title "My Documentary" \
  --overlay-category-logo /path/to/icon-square.png

Note on --overlay-category vs --overlay-category-logo: If both are provided, the logo takes precedence and a warning is printed to stderr.

Using embedded cover art (MP4/M4V/MOV)

# Use embedded artwork if present, fall back to frame extraction
video-thumbnail-creator extract /path/to/video.mp4 \
  --embedded-image prefer \
  --mode auto

# Always skip frame extraction and use embedded cover art
video-thumbnail-creator config set defaults.embedded_image prefer

The --embedded-image option controls how embedded cover art and sidecar images are handled:

• prefer: Use the embedded image or sidecar image if found; otherwise extract frames normally.
• ignore: Always extract frames, even if embedded cover art or sidecar images exist.
• ask (default): Prompt the user when an embedded image or sidecar image is found.

Note: Embedded image detection is only supported for MP4, M4V, and MOV containers. For other formats the tool falls through to sidecar detection and then frame extraction. The --embedded-image option is independent from --mode; the mode only affects how the crop position is determined after the source image is resolved.

Using sidecar thumbnail images

When extracting from a video file (e.g. video.mp4), the tool also checks for existing thumbnail images ("sidecar" files) in the same directory with the same filename stem:

video.jpg / video.jpeg / video.png / video.tiff / video.tif

The detection priority for video input is:

1. Embedded image (inside the video container)
2. Sidecar image (next to the video file)
3. Frame extraction (mosaic flow)

The --embedded-image option controls sidecar image handling the same way it controls embedded image handling.

# Use sidecar image if present, fall back to frame extraction
video-thumbnail-creator extract /path/to/video.mp4 \
  --embedded-image prefer \
  --mode auto

# Always use frames, ignore any sidecar images
video-thumbnail-creator extract /path/to/video.mp4 \
  --embedded-image ignore \
  --mode manual

When a sidecar image is used, the JSON output has "source": "sidecar":

{
  "poster_path": "/path/to/video-poster.jpg",
  "frame_index": -1,
  "mode": "auto",
  "format": "poster",
  "source": "sidecar",
  "reasoning": "Sidecar image file was used as source. | Crop: ...",
  "crop_position": "center",
  "input_path": "/path/to/video.mp4"
}

Using --crop-position to skip interactive and AI crop selection

# Set crop position directly — skips AI crop selection (saves API costs)
video-thumbnail-creator extract /path/to/video.mp4 \
  --mode auto \
  --crop-position center

# Useful for batch processing where crop position is already known
for f in /videos/*.mp4; do
  video-thumbnail-creator extract "$f" --mode auto --crop-position center-left
done

The --crop-position option accepts: left, center-left, center, center-right, right. When provided with --format poster, it skips both the interactive crop prompt (manual mode) and the AI crop selection call (auto/suggest modes).

Image file input (JPEG, PNG, TIFF, HEIC)

# Create poster from a TIFF image (auto color space conversion on macOS)
video-thumbnail-creator extract /path/to/photo.tiff \
  --mode auto \
  --overlay-title "Herbst-Spaziergang"

# Create poster from JPEG with manual crop
video-thumbnail-creator extract /path/to/photo.jpg \
  --mode manual \
  --overlay-title "Mein Foto" \
  --overlay-category "Familie Kurmann"

For image input, ffmpeg/ffprobe are not required. The image itself is used as the high-res source; only crop-position selection and poster composition run.

Semi-automatic suggest mode

video-thumbnail-creator extract /path/to/video.mp4 \
  --mode suggest \
  --output-dir /tmp/thumbs \
  --output-name-suffix -thumb

In poster format, after the frame is selected the AI also suggests a crop position, which you can confirm or override at the prompt.

---

Configuration

Settings can be stored in ~/.config/video-thumbnail-creator/config.toml so you don't have to pass them on every invocation. The directory and file are created automatically on the first config set.

Priority order (highest to lowest)

1. Explicit CLI arguments
2. Config file values
3. Built-in defaults

Commands

# Store a value
video-thumbnail-creator config set claude.api_key "sk-ant-..."

# Read a single value
video-thumbnail-creator config get claude.model

# Show all stored values
video-thumbnail-creator config list

Allowed keys

Key	Description	Default
claude.api_key	Anthropic Claude API key	(none)
claude.model	Claude model name	claude-opus-4-5
tools.ffmpeg	Path to ffmpeg binary	ffmpeg
tools.ffprobe	Path to ffprobe binary	ffprobe
defaults.output_name_suffix	Suffix for output filename	-poster
defaults.mode	Default selection mode	manual
defaults.format	Default output format (poster or landscape)	poster
defaults.embedded_image	Default embedded image handling (prefer, ignore, ask)	ask

Example config.toml

[claude]
api_key = "sk-ant-..."
model = "claude-opus-4-5"

[tools]
ffmpeg = "ffmpeg"
ffprobe = "ffprobe"

[defaults]
output_name_suffix = "-poster"
mode = "manual"
format = "poster"
embedded_image = "ask"

Output filename

The output filename is formed by appending the suffix to the video file stem:

2025-05-15_Starship_IFT7.mkv  +  suffix "-poster"  →  2025-05-15_Starship_IFT7-poster.jpg

---

Output

Default (no --json)

stdout contains only the absolute path of the created image:

/path/to/poster.jpg

All status messages, progress info, and AI reasoning are written to stderr.

JSON mode (--json)

{
  "poster_path": "/path/to/poster.jpg",
  "frame_index": 12,
  "mode": "auto",
  "format": "poster",
  "source": "frame",
  "reasoning": "Sharp, well-lit frame that is representative of the content.",
  "crop_position": "center-left",
  "overlay_title": "My Video Title",
  "overlay_category": "Space Exploration",
  "overlay_note": "2025-05-15",
  "input_path": "/path/to/video.mp4"
}

The overlay_category field is only present when --overlay-category is provided (and --overlay-category-logo is not used alongside it). The overlay_note field is only present when --overlay-note is provided.

When embedded cover art is used as the source:

{
  "poster_path": "/path/to/video-poster.jpg",
  "frame_index": -1,
  "mode": "auto",
  "format": "poster",
  "source": "embedded",
  "reasoning": "Embedded cover art was used as source image. | Crop: ...",
  "crop_position": "center",
  "input_path": "/path/to/video.mp4"
}

When an image file is used as the source:

{
  "poster_path": "/path/to/photo-poster.jpg",
  "frame_index": -1,
  "mode": "auto",
  "format": "poster",
  "source": "image",
  "reasoning": "Image file was used as source. | Crop: ...",
  "crop_position": "center",
  "input_path": "/path/to/photo.tiff"
}

The source field is "frame" when a video frame was used, "embedded" when embedded cover art was used, "sidecar" when a sidecar image file was used, and "image" when an image file was used as input. When source is "embedded", "sidecar", or "image", frame_index is -1.

When --fanart is used, the JSON output includes an additional fanart_path field:

{
  "poster_path": "/path/to/video-poster.jpg",
  "fanart_path": "/path/to/video-fanart.jpg",
  ...
}

---

Fanart Image (--fanart)

The --fanart flag generates an additional clean 16:9 JPEG alongside the normal poster or landscape output. This image has no text overlays, no badges, and no gradients — just the pure source frame scaled to 16:9. It is intended for media servers such as Infuse and Emby that look for a file with a -fanart suffix.

# Generates both "My Video-poster.jpg" and "My Video-fanart.jpg"
videos-thumbnail-creator extract "My Video.mp4" --fanart

Output resolution:

• 4K source (width ≥ 3840 or height ≥ 2160): 3840 × 2160
• Otherwise: 1920 × 1080

Non-16:9 sources: A blurred background fill is applied automatically (same visual approach as the existing frame extraction) so the output is always exactly 16:9 without black bars or stretching.

---

Poster Styles

All visual design constants (colors, fonts, layout, badge placement) are defined as built-in named styles. Select a style with --style <name> (default: internet).

List available styles

video-thumbnail-creator styles

Style	Description
internet	Category header overlaid at top of image, matte black text area, large bold fonts, badges on image — optimised for YouTube/web video posters

Examples

# Default style (`internet`)
video-thumbnail-creator extract video.mp4 \
  --overlay-title "My Title" \
  --overlay-category "My Channel" \
  --overlay-title "My Video Title" \
  --overlay-note "15. March 2025"

---

Exit Codes

Code	Meaning
0	Success
1	General error (file not found, ffmpeg missing, etc.)
2	No selection made (user cancelled)
3	AI selection failed (no API key, timeout, invalid response)

---

Integration

Library Integration (Python Import)

video-thumbnail-creator ships a full Python API at three levels — pick the one that fits your use case:

CLI (Terminal user)
  └→ High-Level API: create_thumbnail()     – one call, everything automatic
       └→ Mid-Level API: ThumbnailSession    – multi-step, caller controls each step
            └→ Low-Level API: extract_frames(), compose_poster(), etc.  – individual building blocks

High-Level API – create_thumbnail()

One call for fully automatic thumbnail creation (AI selects frame and crop position):

from video_thumbnail_creator import create_thumbnail

result = create_thumbnail(
    "/path/to/video.mp4",
    overlay_title="My Film",
    output_dir="/output/",
    fanart=True,
)
print(result.poster_path)   # absolute path to the poster JPEG
print(result.fanart_path)   # absolute path to the fanart JPEG (or None)
print(result.reasoning)     # AI explanation

Works with image files too (JPEG, PNG, TIFF, HEIC) — ffmpeg is not required:

result = create_thumbnail("/path/to/cover.jpg", format="poster")

Mid-Level API – ThumbnailSession

ThumbnailSession gives you full control over each step. Use the context manager for automatic cleanup of temporary files.

Automatic — AI decides everything, step by step:

from video_thumbnail_creator import ThumbnailSession

with ThumbnailSession("/path/to/video.mp4") as session:
    suggestion = session.suggest_frame(title="My Film", description="A documentary")
    session.select_frame(suggestion["frame_index"])
    crop = session.suggest_crop()
    result = session.compose(
        crop_position=crop["crop_position"],
        overlay_title="My Film",
        output_dir="/output/",
    )

Suggest — AI suggests, caller confirms or overrides:

with ThumbnailSession("/path/to/video.mp4") as session:
    print(session.mosaic_path)   # show the mosaic to the user
    suggestion = session.suggest_frame()
    # ... show suggestion to user, let them confirm or pick a different index ...
    chosen_index = int(input(f"Frame [{suggestion['frame_index']}]: ") or suggestion["frame_index"])
    session.select_frame(chosen_index)
    result = session.compose(crop_position="center", output_dir="/output/")

Manual — caller decides everything, mosaic is just a visual aid:

with ThumbnailSession("/path/to/video.mp4") as session:
    # Display session.mosaic_path to the user, then:
    session.select_frame(7)
    result = session.compose(
        crop_position="center-left",
        overlay_title="My Film",
        format="poster",
        output_path="/output/my-film-poster.jpg",
        fanart=True,
    )

Image input — no frame extraction needed:

with ThumbnailSession("/path/to/cover.jpg") as session:
    # session._frame_selected is already True; call compose() directly
    result = session.compose(format="poster", output_dir="/output/")

Low-Level API – Individual Functions

Use the building blocks directly when you need maximum control:

from video_thumbnail_creator import (
    get_video_properties,
    extract_frames,
    create_mosaic,
    extract_single_frame_highres,
    compose_poster,
    compose_fanart,
    detect_badges,
)

props = get_video_properties("/path/to/video.mp4")
frame_paths = extract_frames("/path/to/video.mp4", "/tmp/frames/")
mosaic = create_mosaic(frame_paths, "/tmp/mosaic.jpg")
highres = extract_single_frame_highres("/path/to/video.mp4", 5, "/tmp/highres.jpg")
badges = detect_badges(props)
compose_poster(highres, "center", "My Title", "/output/poster.jpg", badges=badges)
compose_fanart(highres, "/output/fanart.jpg", is_4k=props["is_4k"])

---

CLI Integration (Subprocess)

Because stdout contains only the file path (or clean JSON), this tool is also easy to integrate via subprocess when you cannot import it directly:

import subprocess, json

result = subprocess.run(
    ["video-thumbnail-creator", "extract", video_path, "--mode", "auto", "--json"],
    capture_output=True, text=True, check=True,
    env={**os.environ, "CLAUDE_API_KEY": "sk-ant-..."},
)
data = json.loads(result.stdout)
poster_path = data["poster_path"]

---

Recent Changes

This section lists the release notes for the three most recent versions. For older versions, see the Releases page or the respective version on PyPI.

v1.6.1

• Internet poster layout refined: Long titles are now better contained so they no longer crowd the date or note area at the bottom
• Protected note area: The overlay note/date now has its own reserved space in the lower section, improving readability and visual balance
• Stronger category header: The category label at the top is larger and better proportioned relative to the title, while still scaling down automatically for longer labels
• Cleaner visual spacing: The top header band and lower text area were rebalanced for a calmer overall composition in the internet style

v1.6.0

• Default style: internet style is now the default poster style; the legacy default style has been removed
• Library API: ThumbnailSession class for step-by-step thumbnail creation with full control over each stage (frame selection, crop position, composition)
• Convenience function: create_thumbnail() one-liner for fully automatic thumbnail generation from Python code
• Three integration levels: High-level (create_thumbnail), mid-level (ThumbnailSession), and low-level (individual functions like extract_frames, compose_poster, etc.)

v1.5.0

• Built-in named styles: Replaces external poster_templates.toml with a built-in style registry; new --style CLI flag and styles subcommand to list available styles
• New internet style: Category label rendered as overlay band at the top of the square image, matte black text area (no blur), large fonts — optimised for YouTube/internet video posters
• Historical note: Earlier releases still included a legacy default style with blurred text area; that style has since been removed

v1.4.6

• Layout: Increase overlay category font size in poster composer by @Copilot in #34
• Layout: Remove rounded frame border from poster layout; promote separator line as sole divider
• Code cleanup: Modularize cli.py, centralize fonts/AI-retry, remove dead code, optimize hot paths

v1.4.2

• Batches: Rename FHD→HD, add HFR badge support (4K60/HD48), remove PNG logo config for technical badges
• Thumbnail selection: Prioritise recurring presenter over topic objects in thumbnail selection

v1.4.1

• Library API: New ThumbnailSession class for step-by-step thumbnail creation with full control over each stage (frame selection, crop position, composition)
• Convenience function: New create_thumbnail() one-liner for fully automatic thumbnail generation from Python code
• Three integration levels: High-level (create_thumbnail), mid-level (ThumbnailSession), and low-level (individual functions like extract_frames, compose_poster, etc.)