A creative production line for agencies that serve many brands

One team space per client

Give every client its own team space. The briefs table, the Library folder, the workflows and the run history all live inside it, so a designer working on one brand cannot pick up another brand’s product photos by mistake.

Separate spaces also make billing simple. When generations run on the platform key, usage is metered to the workspace that ran them. If a client brings their own model provider key, add it under Settings → Model API keys for that client. Each generated file records its prompt, model and cost, so a line on the invoice can be traced back to the assets behind it.

The briefs table is the work queue

Create a briefs table in Tables with one row per asset: product, format, channel, key message, due date, status and reviewer. Account managers fill it in, or paste a client email into Chat and ask it to add the rows.

Status drives everything. Each workflow starts from a table trigger that fires when a row’s status changes. A row marked ready for images starts the image workflow. An approved image starts the video workflow. An approved asset starts the publish step.

Workflow 1: on-brand images

The image workflow runs once per brief. An Agent block reads the brief and the client’s brand guide, then writes a generation prompt. The Image Generator block takes that prompt plus a reference product photo from the client’s Library folder in Files, so the product in the image is the client’s product, not an invented one.

Pick models per client. Access policies set default image and video models for a group, so the fashion client can default to one model and the furniture client to another. The live catalogue includes models such as Nano Banana 2.1, Seedream 5.0, FLUX.2 and GPT Image 2, and it changes over time. When someone explores ideas in Chat first, they can pin a media model for that chat with + → Media models, separate from the reasoning model.

Workflow 2: video from an approved frame

Video costs more than images, so start it only from an approved key frame. The Video Generator block uses the approved image as the first frame and a short motion prompt, with a model such as Seedance 2.5, Veo 3.1, Veo 3.1 Fast or Kling v3.0.

Generations estimated above the cost confirmation threshold ask before running. The default is US$1, and admins set their own figure in Access policies. A running generation shows a progress card with a cancel button. To produce several variants, loop over a list of motion prompts with a Loop or Parallel block, one run per variant.

Workflow 3: review, then publish

The third workflow starts with a Human block. The run pauses and sends the creative lead the asset, the brief and the checklist. They approve, reject, or return it with notes that go back into the briefs table.

Only after approval does the workflow deliver:

• Instagram. The Instagram block publishes an image, video, reel, story or carousel to the client’s professional account.
• TikTok. The TikTok block uploads the video as a draft and checks its status. A person posts it.

Every generated file lands in the Library under Generated/Images or Generated/Videos with its provenance. Every run is traced block by block in Logs, and workflows run from a deployed version, so a prompt change is reviewed before it reaches live briefs.

Where to start

Pick one client with a clear brand guide and good product photos. Run twenty briefs through the image workflow only, and have the creative lead score each output against the checklist. Fix the prompt writer’s instructions first and the model choice second. Add video once most images pass first time. For what happens after the post goes live, see Instagram monitoring, and for weekly results, client reporting.