Podcasters · interview shows · coaches
01 — Short-Form Clipping & Repurposing
A long recording turned into a set of vertical clips for Reels, TikTok and Shorts.
- Moments picked for the hook, not just the loudest 30 seconds
- Every clip starts and ends on a complete sentence
- Speaker held in frame through the reframe
- Burned-in captions, plus matching SRT files
- Loudness matched across the whole set so no clip sounds quieter than the next
Proof: clipping demo built from a 60-minute recording, eight finished verticals — see
the case study.
Anyone publishing to social
02 — Subtitles & Captions
Captions burned in for sound-off viewing, or delivered as SRT/VTT for upload.
- Word-level timing, so captions land on the syllable rather than the sentence
- Second pass over the whole transcript before delivery
- Styled to stay legible over busy footage
- Single language per job — I don't sell translation or dubbing, because I haven't wired a pipeline I trust
Straight answer: a single auto-transcription pass is never sold as reviewed. Machine captions mishear proper nouns and technical terms, and the second pass is what catches it.
Brands repurposing one asset everywhere
03 — Vertical Reframing
One source video, delivered in every shape the platforms want.
- 9:16 vertical, 1:1 square and 4:5 feed, from a single pass
- Safe-area checked per platform so nothing important sits under the UI
- Hook text burned in at the top where a crop would otherwise eat it
- A cover frame pulled from the same footage for the thumbnail
Honest note: reframing here is measurement plus human review, not automated face-tracking. On a difficult set it needs a per-clip offset — and I'll tell you before quoting.
Anyone with raw footage and no time
04 — Video Assembly & Finishing
Separate takes joined into one finished piece.
- Cut on the action, not on the timeline grid
- Colour, contrast and brightness matched shot to shot so footage shot at different times reads as one film
- Music bed ducked under the voice
- Titles and an end card
- Master at −16 LUFS with a true-peak ceiling
Proof: shot-to-shot grade match shown with a before/after frame on
the work page.
Brands wanting a motion identity
05 — Motion Titles & Kinetic Type
Animated titles, intros, outros and lower thirds.
- Per-word reveals timed to the audio
- Full kinetic typography for ad and social openings
- Rendered from real vector and web typography, not a stock preset
- Delivered as a transparent overlay or baked in
Method: motion built in code — SVG, CSS and frame-accurate compositing — so a change of wording costs a re-render, not a rebuild.
Performance marketers · small brands
06 — AI Short Video Ads
A finished ad, produced end to end.
- Script written around one message instead of a feature list
- Voiceover, or your own recording cut to
- Word-level burned-in captions so the ad still works with the sound off
- Music bed mixed to −16 LUFS
- Delivered in 9:16, 16:9 and 1:1, with full commercial rights
Why it's cheaper here: rendering runs on a local GPU rather than a rented per-clip tool, so a scene that doesn't work gets re-rolled instead of billed as a retry.
Photographers · brands · archive work
07 — Remove or Add a Person or Object
A person, object, sign, logo or cable taken out of a clip or a still — or a new element added in.
- Background rebuilt behind whatever is removed
- Grain, lighting and motion matched to the surrounding shot
- Added elements placed, scaled, lit and shadowed into the scene
- Audio untouched, and a frame-by-frame check on video
- Both video and stills, priced separately because they are different jobs
Proof: a person removed from moving footage, a person removed from a photo, and an object added to a photo — all three on
the work page.
Anyone with soft or stuttery footage
08 — Upscaling & Frame Interpolation
Higher resolution, smoother motion.
- Restoration upscaling for older or compressed sources
- 24fps to 48fps interpolation without the ghosting frame-doubling leaves behind
- Generate at 720p, deliver crisp at 1080p
Priced around GPU time, and I'll say so upfront: a full upscale pass is roughly 13 minutes of compute per 5 seconds of footage. That's a real cost and it's in the quote, not discovered later.
Podcasters · course creators · anyone recording at home
09 — Audio Cleanup & Mastering
A noisy recording turned into something that sounds deliberate.
- Hum and hiss removal, de-essing, silence trim
- Level drift reduced between sentences
- Loudness delivered at your platform's target — −16 LUFS streaming, −14 podcast, −23 broadcast
- WAV and MP3, plus a short written report of what the noise was, what changed, and what couldn't be fixed
Proof, measured: hiss −11 dB, level variance 8.1→5.9 dB, loudness range 9.6→8.5 LU, and a master at −27.0→−16.5 LUFS with a −1.5 dBTP ceiling on the delivered file.
Explainer videos · multi-market content
10 — AI Voiceover & Narration
A narration track in a chosen voice and pace, cued per line.
- 13 languages, mixed and level-verified rather than dumped out of a tool
- Consistent voice across a whole series
- Paced per line to the edit, not one continuous take
Where I'm careful: Serbian, Croatian and Bosnian get a by-ear check on every job, and Cyrillic script is verified by file size before delivery — it renders empty if a voice doesn't cover it.
YouTubers · product brands · sellers
11 — Thumbnails, Covers & AI Image Sets
Design and image work that matches a brand rather than fighting it.
- Thumbnails with A/B variants and the source files
- Delivered at platform sizes — YouTube 1280×720, gig 1280×769, blog 1000×750
- Product photography: cut-out, relight, studio background, real shadow
- Packaging text repaired or added where a source label is unreadable
- Seamless texture packs and matching cover art
- Carousels, Story covers and feed cards built from a brand's own type and colour system
Proof: a full social set reverse-engineered from a target company's live design system, board shown on
the work page.
Agencies · publishers · anyone with volume
12 — Batch Media Pipeline & Automation
A repeatable pipeline that processes hundreds of files unattended.
- Caption, resize, watermark, convert and publish on a trigger
- Runs on a schedule, with the results waiting for you rather than you waiting on it
- Self-hosted, so it costs nothing per run — no per-file API bill
- Built on workflow automation plus a local GPU for the generation steps
This is the one that's hardest to replace: it combines workflow automation with local media generation and publishing, which is not something you can buy off the shelf as a single service.
Businesses that need a site · or need the one they have fixed
13 — Website Creation
A site that loads fast, gets found, and can be edited by whoever actually runs it — WordPress when you need a CMS, hand-built HTML when you don't.
- WordPress or a hand-built static build, chosen by what you will maintain afterwards rather than what I like building
- The unglamorous half done: canonical tags, a sitemap submitted to Search Console, security headers, an image weight budget
- Product and checkout pages where you're selling, with the licence handling that goes with them
- Deploys checked file by file — a broken upload fails loudly instead of quietly publishing a half-updated site
Proof: two are live and you're reading one —
yakwp.com (WordPress, running its own plugin) and this site (hand-built static, no CMS, no page builder).
Anyone whose week is eaten by the same manual steps
14 — AI Agents & Automation (Hermes)
A persistent AI agent wired into your own tools. It runs the scheduled work, checks itself, and reports back — instead of waiting for someone to type a prompt.
- Built on Hermes, an open-source AI agent framework, with n8n and PostgreSQL alongside it
- It works inside your systems — your CMS, your database, your workflows — not in a chat window beside them
- Scheduled jobs with monitoring: content pipelines, analytics pulls, source watching, backups, and an alert when something stops
- Models run on your own hardware where the data can't leave the building, so there is no per-call API bill
Proof: it runs my own operation — the fifteen scheduled jobs on
the automation page, each one logging to a database you can query. No client deployment exists to show yet, and I'd rather say that than invent a case study.
What I don't offer, and why. Background removal as a standalone service, subtitle translation and dubbing, custom music composition, standalone colour grading and green-screen keying are all listed nowhere on this site — because the capability isn't proven here yet. I'd rather have a shorter list where every line is real.