Create a skincare brand by defining a visual system first, generating a coordinated set of product and campaign images, reviewing them as a group, and then turning approved stills into short product videos and channel-specific crops.
Behind this: 13 build steps · 2 tools and how each is used · how to validate demand · 6 things the video never answers.
Searchable transcript of This Claude MCP Turns One Video Into Ads, Shorts & Viral Content — ManuAGI - AutoGPT Tutorials (09:30). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by ManuAGI - AutoGPT Tutorials. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 Most AI creative workflows start with a prompt and end with one image. That is useful, but it is not really a brand workflow. A real brand needs a visual system. It needs a product world, a consistent color language, packaging, campaign images, social assets, and finally, motion that all feels like it belongs together. So, in this video, I'm going to build a fictional skin care brand called Luma Botanica from scratch using Claude connected to Higgsfield MCP.
00:30 I'm not going to hand Claude a giant creative brief or manually write every image prompt. I'll give it the business goal, ask it to make a few focused decisions, and then let Claude orchestrate the Higgsfield generation workflow. The final result will be mostly image generation followed by a smaller video section at the end. We'll create the brand identity through stills first, then use those approved assets to build short product videos.
00:56 That order matters. The images establish consistency. The videos add movement after the visual system already exists. First, I'll connect Higgsfield MCP to Claude. In Claude, open settings, go to connectors, click the plus button, choose a custom connector, enter the Higgsfield MCP endpoint, and add it. Once the connection is active, Claude can access the available Higgsfield creative tools from the conversation.
01:24 The important thing is that I'm not asking Claude to generate immediately. I'll first ask it to act as a creative director. I'll give it a simple brief. Luma Botanica is a modern botanical skin care brand for people who want calm, considered daily rituals. The hero product is a glass facial serum bottle. The visual direction should feel warm, premium, natural, and editorial, not clinical and not overly rustic.
01:51 Then I'll ask Claude to make the decisions we need before generation. Define the audience, propose a pallet, choose the photography style, define the recurring materials, and recommend the first set of brand assets. At this point, Claude is helping me make decisions. Higgsfield is the production layer. Once Claude returns the direction, I'll ask it to convert those decisions into exact Higgsfield prompts.
02:16 I'm still reviewing the choices, but I'm not manually translating every idea into generation language. Now we start the image-heavy part of the workflow. The first asset is the hero product image. This gives us the anchor for everything else. The bottle, the label treatment, the material quality, the lighting, and the tone of the campaign. I'll ask Claude to generate a clean studio image of the serum bottle with warm ivory stone, soft shadows, muted sage accents, and a small botanical detail.
02:48 The composition should have enough negative space for a headline, because this needs to work as a website hero or a paid social asset. The next step is variation. I'll ask Claude to keep the product world consistent while generating a top-down ritual image. The serum, a folded linen cloth, a ceramic tray, botanical leaves, and a small dropper mark on the stone.
03:12 This is where the connected workflow is valuable. I'm not asking for a completely new style. I'm asking for a new composition inside the same system. Now I'll build supporting images for the launch. A close-up of the serum texture, a bathroom shelf scene, and a hand applying the product in soft window light. The texture image gives us a detail asset.
03:34 The shelf scene makes the brand feel lived in. The application image shows the product in use. These are different jobs, but they should still share the same light, pallet, and restraint. The final image generation step is a social campaign set. Claude can ask Higgsfield for three crop-friendly compositions: a centered square, a vertical portrait, and a wide banner.
03:58 The visual identity stays constant, but the framing changes for each channel. This is the difference between generating random images and building a brand system. Every asset has a role, and every asset inherits the same decisions. Before moving into video, I'll ask Claude to review the generated images as a set. The prompt is not "make them better."
04:21 It is more specific. Compare the images for consistency in palette, product shape, lighting, background materials, and perceived brand quality. Then recommend which assets should be the hero, supporting, and detail images. This is where I make the human decision. I'll approve the strongest assets and ask Claude to revise only the weak ones. For example, if the bottle shape drifts, I'll ask for a corrected product image.
04:48 If one image feels too green, I'll ask Claude to bring it back toward the warm ivory and sage palette. If the lighting feels too dramatic, I'll ask for softer morning light. Again, Claude orchestrates the decisions, and Higgsfield produces the assets. I'm not generating endlessly. I'm reviewing, selecting, and directing the next action. The goal is not to invent a new visual language.
05:12 The goal is to add controlled motion to the approved brand assets. I'll start with a short product reveal. Use the hero bottle image as the reference and ask for a slow camera push in, a gentle highlight moving across the glass, and a subtle botanical shadow shifting in the background. The second video is a ritual shot. I'll use the flat lay and ask for a slow overhead drift with a small serum drop moving into frame.
05:38 The final video is a short social cut using the hand application image. The movement should be simple: a natural hand motion, a soft focus shift from skin to bottle, and a clean end frame. At the end, Claude can ask Higgsfield to create the three outputs in a consistent aspect ratio set. Vertical for short form, square for social feeds, and wide for a website or presentation.
06:03 We started with a brand decision, not a random image prompt. Claude researched the creative direction lightly, made the visual choices, converted those choices into exact generation prompts, and orchestrated the Higgsfield MCP workflow. Then we spent most of the process creating images. The hero product, Ritual flat lay, texture detail, bathroom lifestyle scene, application image, and campaign variations.
06:30 Only after the still system was approved, did we move into video. The video generation was focused and controlled because the brand language was already established. That is the practical advantage of connecting Claude to Higgsfield MCP. Claude gives you the planning and orchestration layer. Higgsfield gives the agent the ability to generate the actual media and move the assets into the working directory.
06:54 You are not just asking AI to make a picture. You are building a usable brand system from one conversation. So after Claude generates the first group of assets, I'm going to ask it to compare them instead of blindly requesting more images. I want a short decision report. Which image is strongest? Which image has a product consistency issue? Which image has the best negative space?
07:14 And which image is most useful for a landing page, an email, or a social post? That gives me a clear next step. Maybe the hero image is already approved, but the lifestyle image needs a softer background. Maybe the detail shot has the right serum texture, but too much contrast. Maybe the application image has a realistic hand, but the bottle label is inconsistent.
07:38 I can say, "Keep the approved Luma Botanica bottle, palette, stone surface, and lighting direction. Only soften the shadow, remove the extra leaf, and create more negative space on the right for a headline." Claude can turn that into a targeted Higgs field request. The same approach works for crops. A website hero needs room for copy. A social post can use a tighter product composition.
08:05 A vertical story needs the product positioned inside a tall frame without losing the focal point. These are not new brands. They are different applications of the same visual system. Now, I'll ask Claude to prepare those applications. It can decide which image is the source for each crop, what needs to remain fixed, and where the composition should leave room for text.
08:27 I'll still review the output because layout is a creative decision, but I'm not manually rewriting every prompt. This is also where the workflow becomes useful for a small team. A designer can approve the direction. A marketer can request a campaign variation. And the agent can carry the same visual rules through the generation steps. The brand does not depend on one person remembering every detail.
08:53 Let's also look at how the assets could be used in a real launch. The hero image can open the website. The ritual flat lay can support an announcement post. The bathroom scene can explain the daily use case. The texture close-up can appear in a product detail section. The application image can make the product feel approachable. Each image answers a different question.
09:15 Start with one brand, define the visual rules, generate the stills first, review them as a set, and only then move into video. The more consistent your decisions are at the beginning, the less time you spend fixing disconnected outputs later.