← LeaderboardThe Skill Leaderboard

video-enhancement

Upscale, restore, denoise, interpolate and edit existing videos via inference.sh CLI. Models: Topaz Proteus, Starlight, Astra, Nyx denoise, Apollo/Chronos/Aion frame interpolation, Themis deblur, colorization, Hyperion SDR to HDR, RIFE, Bria video upscale, background removal, green screen, background replacement, video eraser, Pruna P-Video-Upscale. Use for: upscale video to 1080p/4K/8K, restore old or compressed footage, deinterlace, remove noise and grain, slow motion, 60fps conversion, colorize black and white video, remove or replace video background, erase objects from video. Triggers: vi

Published by inference-sh ↗inference-sh/skills

Try your first task

A source-grounded starter task has not been verified for this skill yet.

SKILL.md

name: video-enhancementdescription: "Upscale, restore, denoise, interpolate and edit existing videos via inference.sh CLI. Models: Topaz Proteus, Starlight, Astra, Nyx denoise, Apollo/Chronos/Aion frame interpolation, Themis deblur, colorization, Hyperion SDR to HDR, RIFE, Bria video upscale, background removal, green screen, background replacement, video eraser, Pruna P-Video-Upscale. Use for: upscale video to 1080p/4K/8K, restore old or compressed footage, deinterlace, remove noise and grain, slow motion, 60fps conversion, colorize black and white video, remove or replace video background, erase objects from video. Triggers: video upscale, upscale video, video upscaler, 4k video, enhance video, video enhancement, restore video, topaz video, topaz video ai, proteus, starlight, astra, denoise video, video noise reduction, deinterlace, frame interpolation, slow motion, increase fps, 60fps, rife, colorize video, sdr to hdr, deblur video, video background removal, green screen video, replace video background, video eraser"allowed-tools: Bash(belt *)

Install the belt CLI skill: npx skills add belt-sh/cli

Video Enhancement

Improve existing videos via inference.sh CLI: upscale, restore, denoise, raise frame rate, and remove or replace backgrounds and objects.

Quick Start

Requires inference.sh CLI (belt). Install instructions

belt login

belt app run pruna/p-video-upscale --input '{
  "video": "https://your-video.mp4",
  "target": "4k"
}'

Apps by Task

Upscale

AppApp IDApproachPricing
P-Video-Upscalepruna/p-video-upscaleUpscale to 720p/1080p/2k/4k target, keeps aspect ratio and audio, optional interpolation to 60fps$0.01/s 720p, $0.02/s 1080p, $0.04/s 2k, $0.08/s 4k; 2x above 30fps
Topaz Proteustopaz/proteusPrecision upscaling with manual controls, deinterlacing, face recovery, CGI enhancement$0.0714/Topaz credit
Topaz Starlighttopaz/starlightGenerative upscaling$0.0714/Topaz credit
Topaz Astratopaz/astraCreative AI-guided upscaling with prompt and creativity controls$0.0714/Topaz credit
Topaz Video Upscaletopaz/video-upscaleProteus-family upscaling selected by Topaz model codesnot published
Topaz Video Upscaler (fal)falai/topaz-video-upscalerSimple upscale factor, optional fps interpolation$0.01/s up to 720p, $0.02/s 720p-1080p, $0.08/s above 1080p; 60fps 2x
Bria Video Increase Resolutionbria/video-increase-resolution2x or 4x super-resolution, output up to 8K, max 60s input$0.02/s

Denoise and Restore

AppApp IDApproachPricing
Topaz Denoisetopaz/denoiseNyx models for noise, compression and artifact removal$0.0714/Topaz credit
Topaz Video Utilitiestopaz/video-utilitiesMotion deblur (themis-2), colorization, SDR to HDR (hyperion-2)$0.0714/Topaz credit

Frame Rate and Slow Motion

AppApp IDApproachPricing
Topaz Frame Interpolationtopaz/frame-interpolationApollo, Chronos, Aion models; slowmo up to 16x, 15-240fps$0.0714/Topaz credit
RIFE Video Interpolationinfsh/rife-video-interpolationTarget 24/30/60/120fps, fast and slow-motion modesGPU time, per ms

Background

AppApp IDApproachPricing
Bria Video Background Removalbria/video-rmbgTransparent or solid color background$0.0033/s
Bria Video Green Screenbria/video-green-screenGreen or blue screen behind the foreground$0.0033/s
Bria Video Replace Backgroundbria/video-replace-backgroundNew background from an image or video$0.0033/s

Object Removal

AppApp IDApproachPricing
Bria Video Eraserbria/video-eraserErase regions defined by a mask video; max 5s, 750p$0.02/s

Topaz apps bill in Topaz credits. Check any app with belt app pricing <app-id> and estimate a run with belt app run <app-id> --estimate --input '{...}'.

Proteus vs Starlight vs Astra

ProteusStarlightAstra
TypePrecision upscaling and enhancementGenerative upscalingCreative, AI-guided upscaling
ControlModel choice plus sliders: compression, details, noise, halo, blur, grain, recover detailModel choice only (hq has extra options)prompt, creativity, realism, sharpness
Also doesDeinterlacing, face recovery, CGI/animation-Text-guided detail (prompt limits input to 450 frames)
  • Proteus when you need faithful, tunable cleanup of real footage, interlaced sources, faces, or CGI.
  • Starlight for generative upscaling where the model reconstructs detail.
  • Astra when you want to steer the result with a prompt and creativity level.

Examples

Topaz Proteus

belt app run topaz/proteus --input '{
  "video": "https://old-footage.mp4",
  "model": "proteus",
  "scale": 2,
  "compression": 0.3,
  "noise": 0.2,
  "output_container": "mp4"
}'

Models: proteus (general), proteus-natural (natural look), rhea (detail-rich), theia-detail/theia-fidelity (temporal), artemis-hq/artemis-mq/artemis-lq and other artemis-* (legacy footage), dione-dv/dione-tv and other dione-* (deinterlace), gaia-cg/gaia-hq (CGI/animation), iris-mq/iris-lq (face recovery). Input max 500MB.

Deinterlace an old broadcast:

belt app run topaz/proteus --input '{
  "video": "https://tv-recording.mp4",
  "model": "dione-tv",
  "video_type": "interlaced",
  "scale": 2
}'

video_type: progressive, interlaced, progressive-interlaced. focus_fix: none, normal, strong. output_container: mp4, mov, mkv.

Topaz Starlight

belt app run topaz/starlight --input '{
  "video": "https://low-res.mp4",
  "model": "precise-2.5",
  "scale": 2
}'

Models: precise-2.5 (best quality, default), hq, mini (lightweight), sharp (detail emphasis), fast-2 (speed). upscaling_factor, low_sharpness, and keep_original apply to the hq model only.

Topaz Astra

belt app run topaz/astra --input '{
  "video": "https://clip.mp4",
  "scale": 2,
  "prompt": "crisp details, natural skin texture, sharp foliage",
  "creativity": 0.4,
  "sharpness": 0.6
}'

creativity, realism, sharpness are 0-1. Setting prompt limits input to 450 frames.

Other Upscalers

# Topaz Video Upscale: Topaz model codes (prob-4 default, pnat-1, rhea-1, ahq-12, gcg-5, iris-3, ...)
belt app run topaz/video-upscale --input '{"video": "https://clip.mp4", "model": "prob-4", "scale": 2}'

# Fal Topaz Upscaler: target_fps 16-60; H265 output unless H264_output is true
belt app run falai/topaz-video-upscaler --input '{"video": "https://clip.mp4", "upscale_factor": 2, "target_fps": 60, "H264_output": true}'

# Bria: desired_increase 2 or 4 (required), max 60s input, output up to 8K
belt app run bria/video-increase-resolution --input '{"video_url": "https://clip.mp4", "desired_increase": 4}'

topaz/video-upscale uses capitalized enums (video_type: Progressive; focus_fix_level: None). Prefer topaz/proteus for readable model names.

Topaz Denoise

belt app run topaz/denoise --input '{
  "video": "https://noisy-night.mp4",
  "model": "nyx-xl",
  "noise": 0.5,
  "compression": 0.3
}'

Models: nyx (balanced), nyx-high-fidelity (detail preservation), nyx-xl (heavy noise), nyx-fast (speed). Sliders compression, details, noise, halo, blur take -1 to 1; grain 0-0.1.

Deblur, Colorize, SDR to HDR

belt app run topaz/video-utilities --input '{"video": "https://shaky.mp4", "model": "themis-2"}'      # motion deblur
belt app run topaz/video-utilities --input '{"video": "https://bw-film.mp4", "model": "colorization"}' # b&w to color
belt app run topaz/video-utilities --input '{"video": "https://sdr.mp4", "model": "hyperion-2"}'    # SDR to HDR

The enhancement sliders and video_type/focus_fix apply to themis-2 only.

Frame Interpolation

# 60fps with Topaz Apollo
belt app run topaz/frame-interpolation --input '{
  "video": "https://24fps.mp4",
  "model": "apollo",
  "target_fps": 60
}'

# 4x slow motion, large motion model
belt app run topaz/frame-interpolation --input '{
  "video": "https://action.mp4",
  "model": "aion",
  "slowmo": 4
}'

# RIFE to 60fps, quality mode
belt app run infsh/rife-video-interpolation --input '{
  "video": "https://clip.mp4",
  "target_fps": 60,
  "fast_mode": false
}'

Topaz models: apollo (best quality), apollo-fast, chronos (optical flow), chronos-fast, aion (large motion). slowmo 1-16, target_fps 15-240. RIFE target_fps: 24, 30, 60, 120; slow_motion: true plays at half the interpolated rate.

Background Removal, Green Screen, Replacement

# Transparent background
belt app run bria/video-rmbg --input '{
  "video_url": "https://talking.mp4",
  "background_color": "Transparent"
}'

# Green screen for editing
belt app run bria/video-green-screen --input '{
  "video_url": "https://talking.mp4",
  "green_shade": "chroma_green"
}'

# Replace with an image or video
belt app run bria/video-replace-background --input '{
  "video_url": "https://talking.mp4",
  "background_url": "https://office-background.jpg"
}'
  • background_color: Transparent (default), Black, White, Gray, Red, Green, Blue, Yellow, Cyan, Magenta, Orange.
  • green_shade: broadcast_green (default), chroma_green, blue_screen.
  • Inputs up to 60s and 16K. Audio is kept by default (preserve_audio).

Erase Objects

belt app run bria/video-eraser --input '{
  "video_url": "https://clip-5s.mp4",
  "mask_url": "https://mask-video.mp4"
}'

The mask is a video: white erases, black keeps. Input max 5s and 750p.

Bria output presets (output_container_and_codec): mp4_h264 (default), mp4_h265, webm_vp9, mov_h265, mov_proresks, mkv_h264, mkv_h265, mkv_vp9, gif.

Workflows

Restore Old Footage

# 1. Remove noise and compression artifacts
belt app run topaz/denoise --input '{"video": "https://vhs-transfer.mp4", "model": "nyx-high-fidelity"}' -o step1.json

# 2. Deinterlace and upscale
belt app run topaz/proteus --input '{"video": "<video-url-from-step-1>", "model": "dione-tv", "video_type": "interlaced", "scale": 2}' -o step2.json

# 3. Smooth to 60fps
belt app run topaz/frame-interpolation --input '{"video": "<video-url-from-step-2>", "target_fps": 60}'

Upscale AI-Generated Video

# 1. Generate a 720p clip (see ai-video-generation skill), then:
belt app run pruna/p-video-upscale --input '{
  "video": "<generated-video-url>",
  "target": "4k",
  "fps": 60
}'

fps (16-60) interpolates above the source rate and bills at the higher rate. turbo: true is faster at the same price.

Tips

  • All three Topaz upscalers (Proteus, Starlight, Astra) take scale 1-4.
  • For transparent output from bria/video-rmbg, the app says to use an alpha-supporting output_container_and_codec preset.

Related Skills

# Full platform skill (all apps)
npx skills add inference-sh/skills@infsh-cli

# Image upscaling
npx skills add inference-sh/skills@image-upscaling

# Image background removal
npx skills add inference-sh/skills@background-removal

# Video generation (generate, then enhance)
npx skills add inference-sh/skills@ai-video-generation

# Avatar videos
npx skills add inference-sh/skills@ai-avatar-video

Browse all video apps: belt app list --category video

Documentation