Pipelines
A pipeline is a structured sequence of processing stages used to turn input data into one or more output artifacts. Fluxara pipelines are designed to be simple, repeatable, and operationally predictable.
Pipeline model
Each pipeline defines an ordered set of stages. A stage performs one specific task, such as preparing input, generating an image, applying an upscale step, or exporting the final artifact.
Pipelines in the current beta are intentionally constrained. The goal is not unlimited graph complexity, but stable execution and clear behavior across repeated runs.
Typical pipeline structure
Not every workflow uses every stage, but most beta pipelines follow a similar structure. This makes output behavior easier to compare across presets and projects.
Common stages
Input preparation
Collects prompt text, source images, or structured parameters and validates them before execution starts.
Generation
Produces the initial output using the selected preset and model configuration.
Selection
Narrows the result set when multiple candidates are produced in a single run.
Upscale
Applies a resolution increase or refinement stage to improve final output quality.
Cleanup
Applies optional post-processing steps used to normalize or refine visual output.
Export
Produces the final artifact, metadata, and output record for later retrieval.
Preset interaction
A pipeline does not usually define every low-level parameter itself. Instead, it works together with a preset. The preset supplies stable model and execution settings, while the pipeline defines the logical flow of stages.
This separation helps keep workflows consistent while still allowing teams to compare different preset configurations under the same processing structure.
Job execution
When a pipeline is executed, it creates a job. A job is a concrete runtime instance that binds together:
- the selected pipeline
- the chosen preset
- the provided input
- execution options such as count or resolution
Jobs move through the queue independently and produce their own output artifacts and metadata.
Batch workflows
Pipelines can be used in batch-oriented scenarios where the same preset and stage sequence are applied to multiple prompts or source images.
Prompt batches
Apply a single preset across multiple prompt variations for comparative evaluation.
Source image batches
Reuse the same transformation pipeline across a set of related input images.
Preset comparison
Run multiple presets against the same structured input to compare consistency and visual behavior.
Review-oriented runs
Generate output groups intended for internal review and selection rather than direct publication.
Current beta constraints
- Pipelines are intentionally linear and limited in complexity.
- Branch-heavy workflow graphs are not part of the current beta.
- Real-time collaborative editing is out of scope.
- Stage behavior may evolve during the beta period.
Design goals
Repeatability
The same workflow structure should behave consistently across multiple runs.
Operational clarity
Pipeline stages should be easy to understand and inspect without hidden complexity.
Controlled variability
Variation is introduced deliberately through presets and inputs, not through unstable workflow structure.
Compact scope
The system is optimized for internal workflows, not consumer-scale experimentation.