Anirudh Thakur · Himachal Pradesh, India
AI Product Visuals & Short-Form Brand Videos
Product campaign visuals and short-form videos for launches, social ads and websites.
For brands and creative agencies.
02 · Work
Selected work
Product visuals, brand video concepts and character experiments.
-
Independent concept
ORVALE — Product campaign for social and web
A fictional skincare campaign combining an unchanged pack photo with AI-created settings, synthetic adult imagery and website hero mockups.
Fictional brand, AI-assisted. The website hero is a mockup, not a live site.
View case study on Contra >> -
Spec concept
Launch Films: six Product Hunt launches
Six short launch films for recent Product Hunt launches (Miso, Extrovert, tiun., Cosmic, Voiskey, Fuse). Every shot built in code and cut to the beat.
View case study on Contra >> -
Spec concept
Benoa — Short-form brand video
A self-initiated short-form brand video for Benoa, Corsica, presented as a spec concept.
View case study on Contra >> -
Spec concept
Daily Bread — Vertical café ad
A self-initiated vertical café ad for Daily Bread, presented as a spec concept.
View case study on Contra >> -
Personal experiments
AI Character Format Lab
Personal character and short-form video experiments on my own channels.
Clip shown: Rocky & Hardev — Yeh Dosti.
View case study on Contra >>
More experiments and builds
Characters, model tests, apps and sites.
03 · Services
What you can hire me for
Start with a focused pilot or scope a larger project. Every price is quoted before we start.
-
01
Product Visual Pilot
Two brand-ready photos of one product in new scenes, or four of your existing AI shots of one product fixed, with one creative direction and one round of changes. For a pack, the method is place, relight, lock: your own pack photo goes in, the AI relights the scene, and your original label goes back on top.
- Your real pack, not an AI redraw: your own pack photo goes into the scene, and its printing is put back from it
- Every word checked against your pack photo, and the barcode test-scanned where it ends up about 200 px wide or more
- Light, shadow and reflections that match the scene
- Skin fixed with the Un-Plastic method if a person is in the shot
- The brief written with you, so the direction is agreed before any work starts
- All tool costs included
Packs: tested on four packs made up for the test (a matte carton, a black can, a clear glass bottle and a pouch at an angle): 15 of 15 relights kept the pack where it was placed, and the test barcode scanned in 5 of 5 locked shots. Portraits: in Nano Banana Pro at 4K, 10 of 10 generated test portraits kept the face, the smile and the gaze.
- Timing
- Planning estimate: one week. Final dates agreed with the brief.
- Price
- Quoted before we start. Ask me and I'll tell you, or we can go through it on a 30-minute call.
-
02
Brand Characters & Short-Form Videos
Short-form brand videos, including a recurring AI character for your brand, so you can keep posting without a shoot. I cast the character once and build every clip from the same references, to keep the same face and look from clip to clip; my own duo, Rocky & Hardev, has reels with 81k and 49k likes. Direction, formats and revisions are agreed before production.
- Timing
- Planning estimate: two weeks. Final dates depend on the agreed scope.
- Price
- Quoted before we start. Ask me and I'll tell you, or we can go through it on a 30-minute call.
-
03
Ongoing Campaign Visuals & Brand Videos
The next step after a Pilot or a first video: an agreed number of product photos and brand videos in each four-week cycle, kept to the look you approved.
- Timing
- Planned in four-week cycles, with deliverables agreed for each cycle.
- Price
- Quoted before we start. Ask me and I'll tell you, or we can go through it on a 30-minute call.
04 · Products
Tools from my own workflow
Small digital products, each cut from something I run on my own work. One job each. Pay once, download, use it every day.
-
01
AvailableUn-Plastic
Turn waxy AI skin into real skin, in the photo you already made, without changing the face.
Paste-in edit blocks for Nano Banana Pro and GPT Image edit mode, plus a tested upscale step. Drop in the waxy render, paste one block, generate, upscale. Same face, same light. Only the skin, eyes and highlights change. Strongest in Nano Banana Pro; gentler in GPT Image (ChatGPT), where very smooth skin in small images may barely change.
- For
- Product-stills freelancers and small brands whose last delivery came back with “can you make it less AI?”
- Format
- A 12-page PDF guide and a plain text file
Inside
- Block A for Nano Banana Pro edit mode: brings back pores, sheen and real eyes without redrawing the face
- Block B for GPT Image and ChatGPT edit mode: a gentler fix, for when the face must not move at all
- Tested Magnific Precision V2 settings, because in testing the upscale on its own added no pores: one set for real photos, one for clean ads
- Two helper lines, for a photo that is still waxy or a face that drifts
- A 12-page PDF guide with tested before-and-after pairs, so you know what a good result looks like
- A plain text file, so the blocks paste in clean
Buy for $9 >>$9 USD · pay once
Details >> -
02
AvailableKeep the Pack
Your real pack in a new AI scene, with every word, the artwork and the barcode put back from your own pack shot.
A five-step method for AI product photos. Paste your real pack shot into a scene, relight it with one tested GPT Image prompt, then put your original printing back with Label Lock, a free browser tool. Real light and shadow from the AI; exact words, artwork and barcode from your own pack. In tests on one carton, its barcode scanned in all 5 finished shots, each about 200 px wide or more; shrunk to 154 px, it didn’t.
- For
- Freelancers, small brands and junior agency people who need lifestyle shots from one white-background pack photo, and can’t ship a redrawn label
- Format
- An 18-page PDF guide, a plain text file and the free Label Lock tool
Inside
- The relight block for GPT Image edit mode, so the light and shadow match the scene and the pack stays where you put it, with lines for card, metal, glass and pouches
- A fill-in scene prompt and four tested scenes, so the scene leaves an empty spot for your pack
- A two-pack version, for one relight with more than one product in the shot
- Label Lock, a free browser tool that puts back the words, artwork and barcode the AI redraws
- Placement rules, including how big the barcode must be to still scan
- An 18-page PDF guide with tested before-and-after results, so you can check yours against them
- A plain text file for clean copy-paste
Buy for $19 >>$19 USD · pay once
Details >> -
03
Coming soonOne Line to Scene
One typed line in. The full scene prompt for your character out.
A scene builder for a recurring AI character on Seedance 2.5 inside FLORA. Type “Hardev argues with a traffic cop on Mall Road” and get the whole prompt back: identity lock, timestamped shots, location, look and bans. Cast once; this writes every clip after that.
- For
- Creators who run a recurring AI character on Seedance 2.5 in FLORA and rewrite the whole scene prompt for every clip
- Format
- A Claude skill, plus a paste-in version for ChatGPT, Gemini or a Claude Project if you would rather not install one (tested in Claude so far)
Inside
- The Claude skill, with a one-time setup for your character
- A paste-in version for a ChatGPT Project, a Gemini Gem or a Claude Project
- The seven-part scene skeleton from the Rocky & Hardev archive
- Tested scenes, each with its prompt and its render (the prompts are tested; the renders are next)
Join the free waitlist >>$19 USD · pay once
Details >>
Want it done for you instead? See services >>
05 · Process
How we work
-
Step 01
Share the brief
Tell me about the product, where the work will be used, the assets available and your target date.
-
Step 02
Agree the direction
We confirm the visual direction, deliverables, revision scope, quote and schedule.
-
Step 03
Create and review
I develop the agreed assets and gather feedback at the agreed review points.
-
Step 04
Deliver
I prepare the approved files in the agreed formats.
Anirudh Thakur
Himachal Pradesh, India · working worldwide
06 · About
Engineer by training. Creator by obsession. Pahadi by birth.
I’m Anirudh, an AI visual creator based in Himachal Pradesh, India. I make product campaign visuals and short-form brand videos for brands and creative agencies. My engineering background helps me build repeatable production workflows.
- FLORA Expert
- Quick responder on Contra
- AI Fluency certified
- 8+ years engineering
Client reviews
From earlier AI image projects on Upwork.
Anirudh was exceptional and a great partner to work with in developing a gamified workshop with AI images. He understood the complexities of our project and used good initiative.
Anirudh has been an outstanding find. He is diligent, highly skilled in his domain, and consistently puts the client's interests first. What stands out is his reliability and patience.
He did everything we needed and was very responsive. We will work with him again.
Did a great job tackling difficult tasks, does not charge for his research!
The engineering side
Production workflows, automation and web builds.
-
YouTube production
Close to 200 videos for six YouTube channels, made with my own workflows for research, scripting, Kokoro voice-overs, Remotion edits and VidIQ-informed publishing.
-
Automation
An n8n pipeline for a client who publishes AI-generated stories and episodes on YouTube. It generated the clips and saved them to Google Drive; the client edited them into the movie.
-
Web apps
Small web apps and sites built with tools like Lovable, Base44 and Codex. A few are in More experiments and builds.
See the workflows
A UGC-style video tool
A tool I'm building: it reads the brand, casts an AI creator and drafts the script. Nothing renders until I've approved the script.
Brand homework
Finds the brand, reads its mission and offer.
AgentCast a creator
An AI influencer made for the brand — or yours.
AI avatarWrite the script
Hooks and lines in the brand's own voice.
LLMMy sign-off
Nothing renders until I've approved it.
HumanImages
The creator with the product.
Image modelsVideo
Clips cut from the approved script.
Video modelsThe YouTube engine
The workflow behind close to 200 videos for six YouTube channels. Each step is a workflow or a skill I built.
Research
Topic research before a word is written.
Own workflowScript
Written and structured for the channel's voice.
Script skillVoice-over
Open-source text-to-speech narration.
KokoroFootage
Created with AI or sourced, scene by scene.
AI + gatheredEdit
Assembled in code with Remotion.
RemotionThumbnail
Concepts from a dedicated skill.
Thumbnail skillPublish
Titles, tags and timing informed by VidIQ.
VidIQ6 channels
AI story clips in n8n
Built in n8n for a client who publishes AI-generated stories and episodes on YouTube. The pipeline generated the clips and saved them to Google Drive; the client edited them into the movie.
Trigger
Each run starts in n8n.
n8nGenerate clips
The clips for an episode.
AI videoSave to Drive
Every clip lands in a shared folder.
Google DriveGoogle Drive
Fashion, on-model
From a real garment to an editorial on a consistent model, using a try-on technique I've published on FLORA.
The garment
Start from the real piece.
Product shotThe model
One face across every look.
Consistent characterTry-on
The garment dressed onto the model.
FLORA techniqueEditorial
Campaign stills in new settings.
Image modelsMotion
The look, in motion.
Video modelsFLORA
Vehicle renders
Your vehicle in new settings, as stills and motion, starting from the photos you already have.
Your photos
Product shots of the real vehicle.
InputThe scene
Location, light and mood, agreed with you.
Art directionRender
The vehicle placed into the scene.
Image modelsDetail pass
Badges, wheels and reflections checked.
Upscale + fixStills
Sized for web, social and print.
FormatsMotion
Reveal shots from the stills.
Video modelsStills + motion
Create
- FLORA
- Kling
- Seedance
- Wan
- Nano Banana
- Midjourney
- ComfyUI
Produce
- Remotion
- Kokoro
- VidIQ
Automate
- n8n
- Claude skills
- APIs
Build
- Lovable
- Base44
- Supabase
- Vercel
New to FLORA, the canvas I build in? Start with my link >>
07 · Writing
Latest writing
-
· 11 min read
Rocky & Hardev: Every Prompt Behind Two AI Pahadi Superstars
Two AI characters from Manali and four reels, made in one day for $30.29 of generation. Every prompt, reference image, setting, cost and mistake, open-sourced so you can make your own.
Read >> -
· 13 min read
Fifteen Years of Government PDFs Told Me the Same Thing Fifteen Times
I built retrieval over fifteen years of Himachal Pradesh Economic Surveys. Pure vector search picked the right year 30% of the time — and that was the least surprising thing I found.
Read >> -
· 7 min read
I Turned My Portfolio Into an MCP Server
My website now answers to agents as well as browsers. Building it took an afternoon — and it immediately caught my own data telling two different stories about the same project.
Read >>
08 · Contact
Tell me what you’re making
Send a link to your product or brand, where the work will be used and when you need it. I’ll suggest a scope and quote, or ask a question first if I need to.
Email hello@anirudhthakur.work
IST · UTC+5:30