The student actively searched for technical resources and materials regarding 3D character animation methods. He utilized several essential relevant sources, while less methodical references, such as documentation and software manuals for the generation tools, are appropriately cited in the footnotes.

Work on system design, character generation, and the preparation of the test data set progressed continuously. While there was some room for improvement in completing the thesis with greater lead time before the deadline, the final content and the results of the evaluation experiments were consulted, and the main supervisor's recommendations were incorporated.

The technical report has some issues that make the work harder to understand. For a significant part of the text, the goal of the thesis is described only very generally, and it becomes clear rather late whether the work focuses on developing or evaluating machine learning models, or on integrating an existing model into a user-facing system.

The text also makes several claims that are not sufficiently supported by experiments. For example, showing that AI can simplify and speed up animation creation would require comparison with a traditional workflow, and showing that the choice of AI model is crucial would require comparing multiple models. The thesis meets the assignment, but some of the presented conclusions are stronger than what the evaluation supports.

The literature review would benefit from a broader overview of machine-learning-based animation generation methods. Section 2.6 describes only the method used in the thesis, which makes the choice of this method seem somewhat convenient rather than well justified. The description of the proposed solution, technical details, and implementation is generally good. Remaining weaknesses are the less readable system diagrams in Figures 3.1 and 4.1, and the fact that testing could have been separated more clearly from the implementation chapter.

The practical output of the thesis is a custom system for generating animations from text prompts for a specific type of data: 3D human characters. The system consists of several components: a ML backend server handling prompts from the client application, an intermediate layer using SMPL for Blender to map animation parameters to GLB 3D objects, and an experimental Unity scene with interactive elements that communicates with the server. Although part of the work is the creation of a user interface in which users can interactively specify and modify 3D animations in a 3D scene, the user testing focuses more on the perceived quality and plausibility of the generated animations than on the usability of the application itself. Nevertheless, the resulting implementation is appropriate for a bachelor’s thesis.

Evaluation level: zadání splněno

The assignment has been fulfilled. The student created a framework that combines an available pre-trained machine learning model for text-based animation generation with an experimental prototype of a 3D scene. Within this scene, the user can generate several animations of 3D characters by specifying a text prompt and other parameters. The system also allows the user to edit existing animation sequences. The technical report includes the required system testing in the form of user testing and verification of the generated 3D animations.

The work with literature is below average. One of the most technical parts of the theoretical section, dealing with generative AI for animations, relies in several places on footnote references to blogs and popular science websites. A broader selection of sources related to methods similar to MDM [7] is missing. The bibliography contains only nine items in total. Moreover, some references, such as [7] and [8], are cited as arXiv preprints, even though peer-reviewed versions, for example, conference or journal publications, are available. Reference [1] is incomplete.