Scenario.

When you need your references

Scenario combines the generation of assets, work with references and training of user models. For a one-time picture, it is enough to check how the selected model uses the sample. For a series in one style, it can be useful to train a separate model on agreed materials. The selection should show the desired style and do not contain random details that should not go into the result.

The documentation describes different ways of learning, including LoRA for visual tasks. The input data and requirements depend on the method. Prepare several test tasks that do not literally repeat the original pictures.

Check on a series of items

For example, for the game you need illustrations of tools in one design. Collect your approved samples, disassemble them by topic and remove pictures with unwanted signatures. After training, try to create a new subject that was not in the set. Check the thickness of the lines, palette and readability of the silhouette when reduced.

If the model transfers a repeating background to each item, correct the set and compare the new series with the original sample. Do not consider the visual similarity as a confirmation of the rights to use.

Access and Computing

In Scenario, you can choose a general or your own model, and some tasks can be automated through the API. Before training and a long series, check the calculation cost for the current plan conditions. If the materials belong to the customer, specify whether it is possible to upload them to the platform and use them for training.

Suitable tasks

• Prepare a series of illustrations in one style
• Train the model on permitted materials