Dataset edits do not update an existing Element
Leonardo's Elements FAQ directly states that adding or removing images from a Dataset does not change Elements already trained from it. Training a new Element is required for those dataset changes to affect the trained artifact. This is an explicit provider rule, not an inference from a hypothetical failed generation. A merchant who changes bottle packaging photos but continues selecting the old Element is still using the old training artifact.
Track four independent states
Consider a fictional skincare bottle whose cap changes from silver to matte black. The proposed version register distinguishes image evidence, training, completed artifacts and generation eligibility.
| Record | Hypothetical state | What changes | What does not follow |
|---|---|---|---|
| Dataset D1 | Old silver-cap images | Source collection exists | No new training implied |
| Element E1 | Trained from D1, completed | Reusable trained artifact | Editing D1 later does not update E1 |
| Dataset D2 | Revised approved matte-cap images | Training inputs changed | E1 remains unchanged |
| Training job T2 | Planned or running from D2 | New training process | Not a usable completed Element yet |
| Element E2 | Only after T2 completes | New trained artifact to select | Does not guarantee correct cap or lettering |
| Generation G2 | Eligible model plus E2 | Candidate image produced if supported | Not an approved product photograph |
These identifiers are fictional bookkeeping examples. No training job was started.
Check whether training is currently an option
The provider places this documentation under advanced legacy features and describes Elements as usable with certain older models. Its default Element training base is Flux Dev; an Element trained on Flux Dev is not compatible with Flux Schnell. Other bases have their own compatibility restrictions. Before allocating work, confirm that the account still exposes the legacy training interface, that the intended generation model accepts custom Elements, and that a reusable trained appearance is necessary. Do not assume a newer Omni image-reference workflow accepts the same Element.
| Symptom | Likely decision boundary | Next action |
|---|---|---|
| Updated dataset, unchanged output | Still using E1 | Plan new training if justified |
| Training job still running | No completed E2 | Wait for actual job completion before selecting it |
| E2 absent in selected generation model | Model compatibility or availability | Inspect training base and current model support |
| E2 selectable, cap still wrong | Image fidelity, not version identity | Compare with real SKU evidence; reject unsuitable output |
Choose a proportionate repair
If only one campaign image needs a new cap, an approved photograph and manual composite may be more appropriate than training a reusable Element. If many concept images require the revised product appearance, assess a fresh dataset and compatible training route. The vendor training-policy guide addresses permissions and data use, not this version rule; model identity versus output differences explains another reason results may diverge. Record dataset snapshot, Element identity, training base, job completion, runtime model and physical-reference review before releasing any image.