One product photo, a whole campaign: the graph behind the Flowgen film
A launch needs dozens of shots of one product. Studio, lifestyle, someone wearing it. The film below follows one handbag photo through a single Flowgen graph, and this post is the graph, node by node.
The photo becomes a node
Drop the product photo on the canvas and it becomes an Image Source node. Everything after this reads from that one node, which is the whole trick: the bag is defined once, so every shot downstream gets the same bag, with the same brass rings, instead of a model redrawing it from a description.
If you have a Brand Kit, the photo can live there instead and be tagged into any prompt with @.
Relight it first
The first generate node takes the photo and one line of prompt: relight it on a clean studio sweep. This is the hero shot, and it is also the reference the rest of the graph uses, because a cleanly lit product is easier for a model to place into a scene than a phone photo on a kitchen table.
Branch it into scenes
From the hero shot, four branches, each a prompt node wired into its own generate node: on a desk, on linen, under one warm spotlight, under a Christmas tree. They run in parallel when you press Run all, and each one shows its cost before it runs.
A branch that almost works gets a Variation Lab node after it, which tries four directions from that result rather than from scratch.
Put it on a person
The last branch wires in a second image, a person, and asks for the bag on her shoulder. Two references into one node: the product and the model. This is where a single graph earns its keep, because the person and the bag are both fixed inputs and only the scene changes.
What it ran on
The images in the film ran on GPT Image 2 through my ChatGPT plan, so they cost nothing past the subscription I already had. The same graph runs on Nano Banana Pro or Seedream with a fal, Google or Kinovi key by changing the model on each node. What each of those costs per image is on the prices page, and it differs by more than double between hosts.
Next week, swap the photo
This is the part I built Flowgen for. The graph is saved as a file. Next week a new product comes in, I replace the photo in the first node and press Run, and the same studio shot, the same four scenes and the same on-model shot come out for the new product.
If you would rather not wire it yourself, Flowgen is an MCP server. Ask Claude or Codex to build the graph and you can watch the nodes appear on the canvas.
Download Flowgen. The free tier has every node and five saved workflows, which is enough to build this one.