Raphael Guilhem builds AI-powered creative workflows for eCommerce advertising, including systems for generating UGC-style videos, product demos, motion ads, and fashion imagery.
Raphael Guilhem is most useful for building repeatable AI creative production systems for eCommerce ads, especially UGC-style video variants.
Established DTC eCommerce brands with one converting UGC ad and an internal creative or growth hire.
Wellness or beauty creator-partnership campaigns needing disclosed sponsored placements and consumer reach estimates.
Check first: Ask for commercial terms, availability, and measured cost, quality, and performance results.
Raphael Guilhem builds AI-assisted production processes for eCommerce advertising.
Raphael Guilhem designs production processes that let teams shape and review creative ideas before paying for final generation.
Raphael Guilhem’s posts treat AI advertising as a production system, not just a way to generate images or videos.
They emphasize brand-specific inputs, repeatable steps and checkpoints before costly generation. They also focus on making more versions of existing creative while preserving the elements that matter, and on solving practical visual problems such as flat posing or broken text.
Guilhem argues that models become useful when connected to research, brand context, concepts and production steps.
The process—not any one model—is the unit of work.
One proposed system maps a brand, studies competitor ads and uses the findings to generate static ads. Another uses brand context to create hooks and video concepts.
Guilhem presents storyboards and concepts as a low-cost checkpoint: teams can inspect an idea and approve it before committing to a more expensive video-generation step.
This puts control and iteration ahead of volume for its own sake.
A product image and storyboard reference feed a process that creates an ad concept, storyboard and prompts. In another example, a 12-frame storyboard keeps the same AI character across scenes before generating the video.
Rather than rebuilding an ad from scratch, Guilhem describes workflows that preserve its structure while changing a targeted element.
The aim is to expand creative options without losing the original sequence or brand assets.
One described process takes an ad that is already converting and a photo of a different person, maps the cuts and camera moves, and places the original audio over the generated version. Guilhem also describes animating static ads while aiming to preserve their text and product imagery.
Guilhem argues that technically convincing AI imagery can still fail creatively: a model may render the product, face and fabric well but leave the person in a dull pose.
They favor specific, reusable direction that can be tested and repeated.
Guilhem describes a pose system for fashion and e-commerce imagery, covering standing, seated, mid-motion and close-up views, and says it was tested across more than 50 brands. A separate post says the pose library was being open-sourced.
Answers “How does an end-to-end AI UGC video pipeline actually work?”
Start here for a clear summary of the tools and the cost argument.
The post lays out a four-step pipeline: upload a product, request a UGC-style video, use Gemini with web search to study the brand and write a campaign concept and influencer persona, then generate the video with Sora 2.
A DTC brand can test UGC video ad angles without booking creators, using a small budget per campaign to prototype scripts and personas before committing to live production.
Answers “How do I go from a product photo to a full ad concept without a creative team?”
Start here for a low-friction introduction to AI-assisted concepting.
The post describes a Claude Skill that takes a single product image and a storyboard reference, then writes the ad concept, storyboard, and generation prompts for a 15-second product demo.
A brand can produce a complete storyboard and prompt set from one product image, cutting the cost of early creative exploration before video generation.
Answers “How can I scale a single winning ad cheaply?”
Start here for a concrete character-replacement process.
The post describes a Claude Skill that takes an existing converting ad and a single photo of a new person, maps every cut, camera move, and lighting change, writes its own prompt, quotes the render cost, and renders on Seedance 2.5.
An eCommerce brand can extend the life of a winning ad by swapping in different creators or AI personas without reshooting or remixing.
Answers “How do I create an AI influencer with a consistent identity across scenes?”
Start here for a storyboard-first approach that controls cost.
The post describes a three-step process for creating a realistic AI UGC influencer in under two minutes: ask Claude to generate a detailed prompt using brand, audience, and tone context; generate a 12-frame storyboard showing the same character in different moments for consistent identity; then send the storyboard to Seedance 2.5 for the final video. Guilhem notes that you can use your own voice and suggests creating storyboards at low cost, reviewing them, then generating videos only from approved storyboards.
A brand can build a repeatable UGC factory: iterate cheaply on storyboards, get internal approval, then spend on final video generation only for approved concepts.
Answers “How do I animate a static ad without ruining the text or product?”
Start here for a technical fix to a common motion-design failure.
The post describes a Claude Skill that reads a static ad image, writes the full prompt, and automatically generates a motion video.
A brand with approved static ad assets can turn them into motion ads without recreating typography or product shots, expanding its video inventory.
Answers “Why do AI fashion images look flat, and how do I fix posing?”
Start here for a diagnosis and a tested fix.
The post argues that pose makes people stop scrolling, but AI has no clue what that means.
For fashion or apparel campaigns, a brand can use the pose system to make AI-generated model shots feel campaign-ready rather than obviously AI, with greater potential to stop the scroll.
Answers “How do I generate endless on-brand video hooks?”
Start here for a repeatable hook-generation system.
The post outlines a process that scrapes top-performing hooks with Claude and Social Growth Engineer, feeds them into Pletor with brand context, generates a concept and prompt, and produces a unique “infinite canvas” of on-brand video hooks. The author says the challenge is fitting AI models into brand-specific processes and offers the full setup to those who like and comment.
A brand can build a system that continuously generates fresh, on-brand video hooks from top-performing examples, reducing hook fatigue in paid social testing.
Answers “How can I automate hundreds of static ad variations for Amazon?”
Start here for a high-volume static-ad factory.
The post outlines a process that maps brand DNA, scrapes competitor ads, runs sentiment analysis, and generates 100+ hyper-targeted static ads, 24/7.
An Amazon seller or eCommerce brand can automate static ad variations at scale, using competitor ad scraping and sentiment analysis to fill creative pipelines without manual research.
Several posts describe connecting AI tools into a branded production process: research or brand context feeds a concept, prompts, and then a static or video ad.
Examples include generating hooks from competitor ads and brand information, producing static ads, and building an end-to-end UGC ad production process. The emphasis is on fitting models into a defined process, rather than using a model in isolation.
Ask for independent results and enough detail to verify the stated savings and output scale.
A brand evaluating Raphael for process training could ask for a process map that starts with its brand inputs and ends with usable ad assets. Test the handoffs and brand checks before assuming the proposed output volume or cost claims will hold for that brand.
Raphael presents storyboards and concepts as a checkpoint before rendering.
One process starts with a product image and reference storyboard, then produces a concept, storyboard, and generation prompts. Another uses a consistent 12-frame character storyboard before sending it to a video model.
This approach may suit a team that needs to compare concepts or get client approval before committing to video generation. A useful training deliverable would be a storyboard template and approval criteria tested against the brand’s products and claims.
One described process takes an already-converting ad and replaces its featured person while retaining the original sequence, captions, graphics, product shots, and audio.
Another turns static ads into motion versions, with attention to preserving text and bringing elements into frame over time.
The examples focus on making a targeted change to an existing asset rather than developing a new concept from scratch.
For a brand with proven creative, assess whether these workflows can create controlled variations without changing the parts that made the ad work. Review generated versions for visual fidelity, legible text, and changes to the product or message before using them.
Raphael argues that technically convincing product, face, and fabric imagery can still feel flat if the pose is weak.
They describe a pose library for standing, seated, mid-motion, and close-up shots, intended for editorial and e-commerce use, and say it was tested across more than 50 brands. A separate post says the library was open-sourced.
Fashion and e-commerce teams could evaluate the pose system as a way to brief and vary product imagery. Compare selected poses against the brand’s visual direction and check which work for its products.
Raphael Guilhem's posts document a recurring commercial product relationship with Pletor, the AI agent platform behind the ad-production workflows, alongside the AI tools they use to build those workflows: Anthropic's Claude and Claude Code, Seedance 2.5, Gemini, Sora 2, ElevenLabs, and Nanobanana Pro.
Ask for their relationship with Pletor and any sponsorship, affiliate, client, or investment terms with any brand; the posts do not specify these.
Disclosed partnership
Pletor is the AI agent platform behind every workflow in these posts.
Guilhem describes rebuilding a full UGC ad factory with Pletor agents, generating AI influencer videos through a Pletor workflow, and open-sourcing Pletor's library of poses with the phrase "we built out a full pose system."
Mention
Claude and Claude Code are the orchestration layer in most workflows shown: generating ad prompts from brand context, writing storyboards, mapping an existing ad's cuts and lighting before a render, and producing static ad variants.
One post also notes Anthropic's Loop feature, which lets agents run autonomously. These posts show product use and platform news, not a stated partnership.
Mention
Seedance 2.5 is the video model used for final renders in several workflows: swapping a new person into an already-converting ad while keeping audio and cuts, generating an AI UGC video from a 12-frame storyboard, and animating static ads into motion design without breaking typography.
Mention
In the UGC ad factory workflow, Gemini uses web search to study a brand and write a creative brief and matching influencer persona before video generation.
Mention
Sora 2 generates the final UGC video from the Gemini-written creative brief and persona; the post frames the combination as the full AI marketing stack.
Mention
Guilhem used a random ElevenLabs voice for the AI UGC influencer demo and noted that viewers can substitute their own voice.
Mention
Listed alongside Claude Code and Pletor as components of a workflow that generates 100-plus static Amazon ad variants.
Raphael Guilhem writes practical breakdowns of AI-assisted advertising production.
The recurring subject is not a single model but the production process around it: gather brand or competitor inputs, write a concept and prompts, review a storyboard or still frame, then generate the final ad.
Both write for DTC brands and agencies paying for photoshoots that fatigue quickly, and both propose AI generation as the way out.
Mike Futia's posts take one product photo and describe producing dozens of campaign-ready images through virtual photoshoots, persona swaps, text-in-image hooks, and relighting. Their teaching approaches are similar: a named playbook of protocols, a short list of what the tools change, and an offer to send the full guide.
Mike Futia focuses almost entirely on static image generation and the mechanics of a single tool family, while Raphael Guilhem covers the full production chain — brief, storyboard, video generation, and variation of an existing ad — and builds custom skills to run it. Choose Mike Futia for still-image creative and persona testing; choose Raphael Guilhem for video and UGC pipelines.
Both demonstrate AI production by showing the finished artifact and walking through the steps that produced it.
Josh Cavalier's post on building a 48-second promo with Claude describes writing the script, cloning a voice, storyboarding animations, and coding the scenes, including where the process broke and had to be worked around. That candor about friction and the hands-on tool chain matches Raphael Guilhem's approach.
Josh Cavalier focuses on enterprise learning, talent development, and workforce AI strategy — routing costs, token budgets, and upskilling.
Both write for DTC creative teams about how AI changes the pace of ad production.
Roop Gill argues that the real shift is speed of testing rather than realism, and that the constraint moves from making content to knowing what to test — a point that aligns with Raphael Guilhem's case for reviewing a concept before paying for a render. Both organize posts around a numbered method readers can apply.
Roop Gill focuses on audience psychology — emotional jobs, trigger chains, and the motivators underneath a hook — using mostly prompts and strategy questions.
Raphael Guilhem's measured audience reach sits on LinkedIn, not on a consumer audience: the LinkedIn profile listed 43,982 followers as of 9 October 2026.
For a DTC brand scaling AI-generated UGC video ads, that signals a professional placement — the people following this work are most plausibly eCommerce, agency, and creative-production operators rather than shoppers, so the same budget that reaches 43,982 followers here reaches practitioners rather than buyers of the advertised product.
43982
LinkedIn profile for Raphael Guilhem · 2026-10-09
The nine posts describe Raphael Guilhem’s own systems — Claude skills, Pletor agent setups and video-model pipelines — rather than a sponsoring or brand-partnership offer.
The closest thing to a call to action is comment-gated lead capture: like a post, comment a keyword such as "STORYBOARD", "AGENT" or "library", and connect on LinkedIn to receive the file.
A need: an established DTC eCommerce brand already has one UGC video ad that converts and wants more versions, plus a way to keep producing variants after the engagement ends.
The brand owns its product photography and brand guidelines, and has one internal creative or growth hire who could run a system once it is built.
A need: a wellness or beauty brand wants a creator-partnership campaign: a named and disclosed sponsored placement in which a real person speaks to their own consumer following about the product, with agreed deliverables, a usage window and a reach estimate against a consumer audience.
Raphael Guilhem's posts do not support this need on its three requirements.
A worked angle for the first need: one winner, ten versions, then hand it over.
Objective: turn one proven UGC ad into a versioned campaign and leave the brand able to keep producing variants unaided.
Phase 1, week 1.
Phase 2, week 2.
Phase 3, week 3.
Phase 4, weeks 4 to 6.
What this angle buys: version volume from an asset that already works, a review checkpoint before expensive renders, and a process the brand keeps.
Next step: commission a paid pilot of one ad and three variants with usage rights defined in writing, request one result from a comparable brand, and only then extend to the wider catalogue.
Mike Futia is the better choice when the campaign decision is to multiply static ad creative rather than video.
Raphael Guilhem’s posts show a process that turns one product photo into 50+ campaign-ready images using Nano Banana Pro and Nano Banana 2: virtual photoshoots, instant model swaps for persona testing, text rendered inside the image, relighting to fight fatigue, and localization.
The tradeoff is format.
Josh Cavalier is the better choice when the campaign decision is to produce one polished AI-assisted promotional video from scratch, or to work through how AI tasks are allocated inside a larger organization.
The posts include a 48-second promo built with Claude Opus 5.5, an ElevenLabs voice clone, coded animation in HyperFrames, and browser-controlled final assembly, plus a breakdown of AT&T’s AI gateway, token budgets and human-machine task routing. If the brand wants a scripted, narrated, animated explainer or course promo rather than a set of UGC ad variants, this background is closer to the task.
The tradeoff is the need to scale a winning UGC ad into ten character variants and support the performance-testing loop a DTC brand uses to fight ad fatigue.
Roop Gill is the better choice when the campaign decision is to fix the strategy before generating anything: defining who the ad is for, what emotion it sells, and what triggers make someone stop.
The posts lay out a persona method built on emotional job-to-be-done, external and internal triggers, and three core motivators, plus a Sora 2 argument that the bottleneck is clarity, not content volume.
The tradeoff is that the posts provide frameworks and prompts, not a production pipeline.
Editable Markdown templates for planning a campaign with this creator.
