AI Creator Workflow Wins Hackathon | Generated by AI

Home 2026.07

Question: How to win the AMD Radeon Hackathon Track 1 (Multimodal AI) with a strong project strategy?

Answer:

For Track 1, the judges are giving 80% to product value + innovation and only 20% to Radeon GPU performance. So don’t build “just another Stable Diffusion UI”. The winning strategy is:

Build a complete AI creator workflow that solves a real problem, then prove it runs efficiently on AMD Radeon.

Think AI-native product, not model demo.

Winning project idea: “AI Video Factory for Creators”

A pipeline:

Idea / Script
      |
      v
LLM Story Planner
      |
      v
Text -> Image (scene generation)
      |
      v
Image -> Video (motion generation)
      |
      v
Voice Generation
      |
      v
Auto Subtitle + Editing
      |
      v
Final Shorts/Reels/TikTok video

One click:

"Create a 30 second Tesla AI news video"

Output:

storyboard
images
animated video
narration
captions
thumbnail

This matches:

text-to-image ✅
image editing ✅
image-to-video ✅
audio generation ✅
creator application ✅

Why this can beat competitors

Most hackathon submissions will be:

Stable Diffusion wrapper
chatbot with image generation
simple ComfyUI workflow

Those are commodity.

Your differentiation:

“AI Content Factory”

Target:

YouTubers
TikTok creators
marketing teams
solo entrepreneurs

Problem:

Creating daily video content requires 3-5 hours.

Solution:

Generate 10 shorts/day from one idea.

Architecture

Frontend

Simple:

React
 |
 |
FastAPI
 |
 |
Workflow Engine

AI pipeline

1. LLM Planner

Input:

Topic:
"Explain AMD MI300X"

Output:

{
 "scenes":[
   {
    "duration":5,
    "image_prompt":
    "AMD MI300X GPU inside datacenter"
   }
 ]
}

Models:

Qwen
Llama
DeepSeek

2. Text -> Image

Use:

Stable Diffusion XL
FLUX
SD3

Example:

prompt
 |
v
diffusion model
 |
v
1024x1024 image

3. Image -> Video

This is where you win.

Use:

Stable Video Diffusion
AnimateDiff
CogVideoX
Wan2.1

Pipeline:

image
 |
encoder
 |
temporal transformer
 |
video frames

Generate:

1 image

↓

16-24 frames

↓

mp4

4. Audio

Use:

Whisper
XTTS
CosyVoice

Radeon optimization section (20 points)

This is important.

Don’t just say:

“Runs on AMD”

Show benchmarks.

Example:

Hardware:

RX 7900 XTX
24GB VRAM
ROCm

Measure:

Model              GPU       Time

SDXL               Radeon    18s/image

SDXL               CPU       4min/image


Video generation:
SVD
24 frames
45 seconds

Create dashboard:

AMD Radeon AI Performance

VRAM:
18.2GB / 24GB

GPU utilization:
96%

Generation:
2.1 sec/frame

Judges love visible engineering.

Even stronger: build an agent

Instead of fixed workflow:

Creator Agent

User:
"make a video about Nvidia vs AMD"


Agent:

1. research topic
2. write script
3. create images
4. animate
5. generate voice
6. edit video
7. publish

This connects with current AI trend.

Suggested tech stack

Frontend:
Next.js

Backend:
FastAPI

Agent:
LangGraph / custom

Image:
Diffusers

Video:
ComfyUI backend

Audio:
XTTS

Acceleration:
ROCm
PyTorch AMD build
ONNX Runtime

Demo scenario (very important)

Don’t demo:

“Here is my model.”

Demo:

Start timer:

Input:

"Create a 30 second AI news video:
AMD releases new GPU"

0 sec

Agent starts

20 sec

Storyboard ready

60 sec

Images generated

120 sec

Video generated

180 sec

Final MP4

Then play output.

Judges remember outcomes.

Extra points: use your own advantage

Given your background, I would make it:

“AI Research-to-Video Factory”

Because you already have:

thousands of AI notes
blog pipeline
video generation workflow
AI experiments

Demo:

Your blog markdown

        |
        v

AI reads article

        |
        v

Creates:

- YouTube short
- Bilibili video
- WeChat video
- thumbnail
- subtitles

This is a real product, not a hackathon toy.

48-hour implementation plan

Day 1 morning

Build:

Markdown
 |
LLM
 |
Storyboard JSON

Day 1 afternoon

Integrate:

SDXL
+
AnimateDiff

Day 1 night

Add:

voice
subtitle
video merge

Day 2

Polish:

UI
Radeon benchmark
demo video
README

Winning pitch

Title:

Radeon Creator AI Studio: An Autonomous Multimodal Agent for Generating Professional Short Videos

One sentence:

“We turn one idea into a complete publish-ready video using AMD Radeon accelerated multimodal AI models.”

References:

AMD ROCm documentation: https://rocm.docs.amd.com/
Hugging Face Diffusers: https://huggingface.co/docs/diffusers
ComfyUI: https://github.com/comfyanonymous/ComfyUI
Stable Video Diffusion: https://huggingface.co/docs/diffusers/en/using-diffusers/svd
Luma AI: https://lumalabs.ai/

Back Donate