add_nodes
ChatGPTAdds nodes to a workflow and connects them, to each other or to existing nodes. It is a single API operation that only edits the saved workflow graph: it does NOT run any node, call any model or spend credits. Nodes run only when start_run is called. Everything in one call is saved together: if any part or the resulting workflow is invalid, nothing is saved. A saved edit shows up live for anyone who has the workflow open and is kept as a restore point for revert_workflow.
cancel_run
ChatGPTCancels unstarted nodes in a run or batch, including downstream nodes of active runs. Nodes already generating finish and keep their results.
create_upload
ChatGPTReturns a pre-signed URL to PUT a local image file to (for example with curl -X PUT --upload-file). Use the returned key as { "key": ... } in start_run or node inputs. For images already online, pass { "url": ... } instead.
create_workflow
ChatGPTCreates a workflow in the workspace as a single API operation, optionally with its first nodes and connections (same fields as add_nodes). It only saves the workflow graph: it does NOT run any node, call any model or spend credits. Nodes run only when start_run is called.
estimate_cost
ChatGPTValidates a run without starting it and returns how many runs it creates, the estimated credits and whether the user can afford it. If some node can only be priced at run time, the result lists it in unpriced_nodes and the real total will be higher; the field is absent when everything is priced. node_notes says, per Crop & Expand node, whether each image crops instantly or expands with AI (minutes per image); tell the user before start_run.
get_authoring_guide
ChatGPTExplains how Caimera workflows are built and run: nodes, ports, Batch Inputs, prompt image tags, image inputs and editing rules. Read once before creating or editing a workflow.
get_prompt_guide
ChatGPTReturns one of Caimera's own prompt-writing guides, the same ones its studio agent and the Enhance Prompt node follow. Read the matching guide before writing an image prompt into a node: photorealism for on-model, lifestyle and product photography, catalog for consistent on-model catalog shots across a range, fabric_pattern for repeating prints, graphic_print for placement graphics, design_sketch for design sketches, model_portrait for model casting portraits.
get_run
ChatGPTStatus, credits charged, errors and result images (signed URLs valid 12 hours, plus keys) for a run or a whole batch. Pass previews: true to also see the images, all_outputs: true to get every node's outputs, not only the final ones.
get_workflow
ChatGPTA workflow with its nodes, stored input values, connections, the Batch Input nodes start_run expects (run_inputs), validation problems and recent versions.
list_node_types
ChatGPTRead-only catalog of the node types a workflow can contain, with each type's input and output port ids, value types, allowed values and defaults. Calling it runs nothing and changes nothing; it only tells you what add_nodes and update_nodes accept.
list_workflows
ChatGPTLists the workspace's workflows, most recently updated first.
remove_nodes
ChatGPTRemoves nodes from a workflow, with every connection to or from them. It is a single API operation that only edits the saved workflow graph: it does NOT run any node, call any model or spend credits. Nodes run only when start_run is called. Everything in one call is saved together: if any part or the resulting workflow is invalid, nothing is saved. A saved edit shows up live for anyone who has the workflow open and is kept as a restore point for revert_workflow.
revert_workflow
ChatGPTRestores a saved version (from recent_versions in get_workflow) as the current workflow. The graph it replaces is kept as a version too, so a restore can itself be undone.
start_run
ChatGPTRuns the saved workflow exactly as get_workflow shows it, as a single API operation that starts one batch. Every node runs with the type, stored inputs and connections it already has, in dependency order; this call cannot change or add steps (edit them first with add_nodes or update_nodes). Spends the user's credits, so quote estimate_cost first. Each Batch Input image becomes one run. Returns run and batch ids for wait_for_run. To rerun only part of a finished run (for example after changing one node), pass from_node_id and reuse_run_id: the nodes before it reuse their earlier outputs for free.
update_nodes
ChatGPTChanges stored input values, labels or positions of existing nodes, and removes connections. It is a single API operation that only edits the saved workflow graph: it does NOT run any node, call any model or spend credits. Nodes run only when start_run is called. Everything in one call is saved together: if any part or the resulting workflow is invalid, nothing is saved. A saved edit shows up live for anyone who has the workflow open and is kept as a restore point for revert_workflow.
wait_for_run
ChatGPTWaits up to timeout_seconds (max 30) for a run or batch to finish, then returns the same report as get_run. Call again if it is still running.