09:41
AI video generation
The video demonstrates that Higgsfield AI's new Canvas feature lets you chain Claude AI (for UGC actor concepting and script writing), GPT image 2 (for actor image generation), and Cedents 2.0 (for video generation) into one reusable node-based workflow that turns a single product photo and URL into a hyperrealistic AI UGC ad, without switching between separate tools or subscriptions.
Higgsfield AI Canvas combines multiple leading AI models (text, image, video) into a single infinite-canvas workflow, eliminating the need to jump between different tools or subscriptions.
The workflow starts by right-clicking on the canvas to upload a product photo, e.g. an Adidas squatting shoe.
An 'LLM assistant' node running Claude Sonnet 4.6 is prompted with the product URL and product image to analyze the product and generate an authentic, audience-accurate UGC actor description plus a prompt for the image generator.
The prompt is deliberately written to be reusable across any product rather than hardcoded to specifics like 'a man performing squats' — the user just swaps in a new product URL and image.
The actor image is generated with the GPT image 2 model (called 'the best right now'), at 9:16 aspect ratio and 4K resolution, with the product photo also attached to the image port so the product itself renders accurately.
A second Claude Sonnet 4.6 LLM node, fed the generated actor image, writes a short (~15-second) spoken script plus audio instructions for the actor to read.
The actor image and script are connected into a video generator node using the Cedents 2.0 model to produce the final video, set to 12 seconds, 480p, 9:16 aspect ratio, with sound explicitly toggled on (otherwise the clip is muted).
The whole connected chain of nodes can be executed in one action via 'run pipeline', letting the same workflow be rerun for a different product just by swapping the input photo and URL.
The finished canvas workflow can be made public and shared via link with role-based access (default role 'viewer'), letting others view and clone it without editing the original.
Higgsfield's raw website/model list is described as 'a total mess', and Canvas is positioned as the fix by consolidating everything into one buildable app.
The presenter provides a direct link in the description to clone the exact prompts and workflow shown.
Higgsfield AI Canvas — A node-based, infinite-canvas workflow builder inside Higgsfield AI that combines multiple leading text, image, and video AI models in one place instead of requiring separate tools or subscriptions. Apply: Right-click on the canvas to add and connect nodes (upload, LLM assistant, image generator, video generator) into a repeatable ad-generation pipeline.
LLM assistant node (Claude Sonnet 4.6) — A canvas node that runs a full Claude conversation to analyze a product URL and product photo and generate an authentic, audience-accurate UGC actor description plus a matching image-generator prompt. Apply: Right-click → 'LLM assistant', select Claude Sonnet 4.6, feed it the product URL and product photo, and instruct it to output a generalized UGC actor prompt reusable for any product.
Image generator node (GPT image 2) — A canvas node that converts the LLM's actor prompt into a photorealistic, non-polished UGC-style image of the actor holding the product. Apply: Connect the LLM node's output and the original product photo into an image generator node, choose GPT image 2, set 9:16 aspect ratio and 4K resolution, then run it to produce the actor image.
Video generator node (Cedents 2.0) — A canvas node that animates the generated actor image into a talking UGC ad video using a script and audio instructions, though it lacks explicit start/end frame control. Apply: Connect the actor image and the LLM-generated script into a video generator node, select Cedents 2.0, set duration, 480p resolution, 9:16 aspect ratio, and toggle sound on before generating.
Run pipeline — A canvas feature that executes an entire connected chain of nodes in a single action rather than running each node individually. Apply: Select the full node chain and click 'run pipeline' to regenerate a complete new ad end-to-end simply by swapping the input product photo and URL.
Public sharing with role-based access — A collaboration feature that publishes a canvas workflow via a shareable link with a configurable default access role, such as viewer. Apply: Go public on the canvas, set the default role to viewer, then copy and send the link so others can view and clone the workflow without editing the original.
The Cedents 2.0 video node currently has no clear start or end frame control (unlike competing tools), so the presenter says you need to 'do some gymnastics in the prompt' to get the AI to understand what's going on — an explicit limitation flagged in the video.
The reusability of the workflow hinges on a specific prompt-design choice: parameterizing the Claude prompt on product URL + product image rather than hardcoding product specifics, which is what lets the same node chain work for any product category.
Attaching the actual product photo to the image port (not just describing the product in text) is called out as necessary to make the product itself — not just the actor — render accurately and consistently across the generated image and video.
Even AI-authored output needs manual correction in the workflow: the generated script ran shorter than the requested 15 seconds, so the presenter shortened the video duration and deleted the '15 seconds' text from the prompt to avoid confusing the AI.
The presenter frames Higgsfield's Canvas move as part of a broader competitive trend, explicitly naming Freepik Spaces and Magnific as tools that already have the missing start/end-frame capability.
«I just built a custom AI UGC ad generator workflow with which I created this insane-looking AI UGC ad.»
— 00:00
«So, no longer do you need to jump between these different features or models, but everything happens inside this one canvas.»
— 00:34
«Just finished the hardest session of my week and these held up perfectly.»
— 00:08
«Flat sole, locked-in fit, zero slip.»
— 00:13
«If you're serious about lifting, you need a pair.»
— 00:17
«So, my goal is to make this prompt so that you can use it with any kind of a product you might be promoting.»
— 03:04
«I mean, that is just so fantastic. It sounds like he's in a gym. It doesn't have a robotic voice or stare.»
— 07:35
«So, that is something that is still missing here in Higgsfield AI canvas, which is present in other similar tools, for example, like Freepik Spaces or Magnific these days.»
— 09:13
Reception
Viewers are engaged and generally positive, actively asking questions and comparing features with competing tools like Magnific and Weavy.
This is a promotional tutorial/case study showcasing a specific, cloneable Claude + GPT image 2 + Cedents 2.0 prompt chain inside Higgsfield AI Canvas, valuable as a concrete reusable workflow rather than a general treatment of AI ad-generation theory.
09:41