Checklist IOAI 2024 Practical Round · Task 2
AI-Generated Music Video Challenge: Love by Maria Ilieva
Produce a 47-second AI-generated music video for Maria Ilieva's 'Love' in Neural Frames, featuring a stylised portrayal of the singer.
The task
The second practical-round task asks teams for an AI-generated music video covering a 47-second segment of Maria Ilieva's song 'Love', made with the Neural Frames platform.
Teams received the audio segment, a lyrics video of that segment in English for context, five photos of the singer, and the full lyrics in Bulgarian and English. The video must show a stylised Maria Ilieva, keep a consistent visual style, include at least 5 objects from the same list as the cover task, and match its visuals to the music.
Teams could train up to 3 custom Neural Frames models, which takes 15 to 20 minutes each and requires a fresh project, and could use images of the singer from the web. The final video had to be made and rendered entirely inside Neural Frames with no further editing. Other free AI tools, LLMs and the provided Midjourney account could help with concepts, assets and prompts.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
- A 47-second song segment (WAV), a lyrics video of the segment in English (MP4), five images of Maria Ilieva, and the full song lyrics in Bulgarian and English, all in the practical-round zip, plus Neural Frames access with credits for up to 3 custom models.
- You submit
- The completed 47-second video downloaded from Neural Frames without upscaling, a description of the chosen visual style and why it was chosen, the list of included objects, a list of all additional tools and how they were used, and screenshots of the final Neural Frames timeline.
- Scoring
- Jury score out of 40 points: artist likeness (1–5), predefined object inclusion (up to 7: 1 point for each of the first five objects, 0.25 for each additional one), homogeneous art style (1–5), music synchronization (1–5), generation accuracy (1–5), prompting strategies (1–5, based on the described strategy), video duration (1–3; the full 47 seconds earns 3) and effective use of Neural Frames (1–5). The image and video tasks are weighted 40 to 60 in the practical round's final ranking.
- Rules
- Four hours for both practical-round tasks together
- The final video must be produced and rendered entirely within Neural Frames; no further video editing is allowed
- Credits to train up to 3 custom Neural Frames models; unlimited rendering in one tab at a time
- Other free AI tools, LLMs and the provided Midjourney account may be used for seeding concepts and assets, prompt engineering or workflow enhancement
- Images of Maria Ilieva from the web may be used
- Must include a stylised portrayal of Maria Ilieva and at least 5 objects from the given list
- Organisers strongly suggested avoiding XL-size models because of longer rendering times (a recommendation, not a rule)
- Format
- Practical round, on-site (4 hours for both practical tasks)