A custom tool for building controlled fashion image generation workflows from visual references and production choices. La Factory combines reference management, production controls and modular prompt construction within a dedicated interface.
La Factory is organized around two modes that share the same underlying prompt architecture while addressing different types of fashion image production.
A controlled workflow built around Still Life or On-Model references, automatic outfit completion, optional garment control and a defined set of production controls.
A more flexible workflow based on role-specific references, broader visual direction and iterative refinement of the active image.
The interface switches between Classic and Editorial while preserving a common logic for references, parameters and prompt construction.
La Factory does not rely on a single static prompt. A structured base is extended through conditional instruction modules generated from the active references and selected production parameters, producing the final prompt in real time.
Changes to model, framing, reference structure, styling and other production choices update the relevant instruction blocks without rebuilding the workflow manually.
Classic is the more structured production mode. The workflow progressively defines the source reference, outfit control and production parameters before generation.
The first selection identifies the type of source reference. The system then applies the corresponding instruction logic to the image and to the prompt structure used for generation.
The Main reference defines the primary garment to preserve. When no additional garments are supplied, La Factory completes the outfit autonomously according to the active styling and production settings. Optional garment references can be added when specific pieces need to be preserved or controlled.
Model profile, shot type and framing, outfit behaviour, output size and additional production constraints can be controlled directly from the interface. Each selection feeds the modular prompt system.
Once the setup is complete, the active visual references and structured prompt are sent to the generation backend. Results are saved and remain accessible through the output viewer and session archive.
Editorial uses the same modular foundation but reorganizes the process around a broader visual direction. References are assigned specific roles, and the active output can become the starting point for subsequent refinement.
Pose Reference, Main Garment, Optional Garments and, when needed, Environment Reference are treated as distinct inputs. Each reference contributes a specific type of information to the generated image.
Model, framing, styling, background and additional notes can be combined to define the first image while maintaining the role and hierarchy of the supplied references.
The selected references, parameters and prompt modules produce the first controlled output, which becomes the visual base for the following stages.
Instead of restarting from the original references, the current output can become the source image for a new refinement step. Specific aspects can be changed while preserving the established visual base.
A small selection of results from both workflows. At this point the images can stand on their own: the previous sections already explain the logic used to produce them.
Independent tool designed and developed by Luca Muliedda for controlled AI-assisted fashion image production and workflow experimentation.