Two different mechanisms share one name
No: a FASHN Element is a curated reference image, while Leonardo documents trained Elements as LoRAs for compatible models. A library selection in FASHN does not start Leonardo-style training. Confirm the provider and the operation before looking for a training status or expecting a reusable learned identity. This distinction matters when a merchant has selected an attractive reference but the generated product still has the wrong construction.
Diagnose the failed reference choice
Consider a fictional jacket campaign. The merchant selects a FASHN Background Element showing a minimal studio and expects it to teach the system the jacket’s six-pocket construction for future campaigns. The selection supplies a scene reference, not jacket evidence or trained weights. A correct-looking studio therefore does not resolve an incorrect pocket count. The repair begins by attaching the actual jacket photograph and explicitly identifying which input establishes the merchandise. Keep the background image for scene guidance if it remains useful. Review the resulting jacket against the source; adding the right input still does not guarantee exact reproduction.
| Wanted outcome | Appropriate responsibility | What does not establish it |
|---|---|---|
| Specific jacket construction | Merchant’s approved product source | Background or style reference |
| Person’s facial appearance | Portrait reference in a supported field | Garment photograph alone |
| Stance or gesture | Pose reference | Assuming a portrait preserves a full pose |
| Scene and visual treatment | Background and photography-style references | Proof of product identity |
| Reusable trained Leonardo Element | Separate supported training workflow | Selecting a FASHN library image |
Check where the image was attached
FASHN offers three routes with different interface contexts. Selecting an image in the Elements library opens a new Agent conversation with it attached. Within Agent, Add images → Browse Elements adds a chosen image to the request. Studio reference fields offer an Elements picker that opens a relevant category, with category switching available. Before running, inspect the selected field or attached images. Merely browsing a category is not proof that its image is attached to the intended task.
For the fictional jacket, record “product photograph: garment construction; Background Element: studio environment; Portrait Element: facial guidance,” rather than labeling every input “reference.” Avoid conflicting scene references unless the brief explains the intended relationship. Product, identity, pose and style remain separate review responsibilities. The FASHN face-reference guide explains why a face cue is not an entire original scene.
Use the training route only for a training requirement
Leonardo’s official guide describes building a dataset, training an Element and selecting that Element with a compatible base model. It is a separate workflow, not an alternate button inside FASHN. Its documented model compatibility also prevents assuming that a trained Element works with every generator. For dataset changes and retraining, use the Leonardo dataset and Element compatibility guide. Close the reference-choice issue only after the intended images are attached in their correct roles; close the merchandise-quality issue only after inspecting the generated product. This example is a diagnostic walkthrough, not an executed generation or measured fidelity test.