Building first for our own NFL channel

Your NFL takes.
Your AI studio.
Your final say.

ShareFuze turns NFL stats, game film and a creator's own takes into finished graphics, clips and posts. A human approves every post. We build it first for our own NFL channel, then open it to other creators.

Get early access See the ML pipeline

Working tools today. Video models in testing. AWS plans below.

Illustration of the production dashboard

Product illustration

77

Graphics templates

32

NFL teams

6

Export platforms

Inside the studio

How we use machine learning

Machine learning (ML) helps turn research into media. The War Room is our own working app for this studio. We test video jobs on Runpod GPUs (graphics processors) and fal.ai.

Live: works in our own app. In testing: a trial, not a finished feature. Planned on AWS: future work.

  1. Research
  2. Script
  3. Video
  4. Voice
  5. Review
  6. Publish By a human
In testing

Masks and tracking

SAM 3 tracks a selected person on Runpod GPUs so we can cut them out of a clip. SAM 2 is a planned alternative on AWS.

In testing

Person swap in video

Wan 2.1 and Wan 2.2 support person-swap trials through Runpod GPUs and fal.ai. Wan 2.1 VACE is a planned alternative on AWS.

Planned on AWS

Lip sync

We plan GPU jobs with LatentSync and MuseTalk to match mouth movements to speech. The app already has a local MuseTalk engine on the Mac.

Planned on AWS

Labelled parody voices

We plan to use Chatterbox, CosyVoice and RVC for parody voices on AWS GPUs. Every parody voice must carry a clear label.

Planned on AWS

Captions

We plan to use Whisper to turn speech into timed captions. AWS GPUs will run this job, with the Mac as a local option.

Live

Scripts, angles and fact checks

Claude, GPT and Gemini models help draft scripts and story angles. The app compares numerical claims with saved nflverse evidence before human review.

Our next steps

Our ML roadmap on AWS

These phases describe future work. We have not moved this pipeline to Amazon Web Services (AWS).

Planned on AWS

Phase 1: foundation

Amazon S3 will store all media. Claude on Amazon Bedrock will help write scripts and check facts.

Bedrock embeddings will turn clips and stats into searchable representations, so creators can find related material.

Planned on AWS

Phase 2: GPU pipeline

Amazon EC2 G6e and G6 instances, with NVIDIA L40S and L4 GPUs, will run mask, swap, lip-sync, voice and caption jobs.

AWS Batch will queue jobs. AWS Step Functions will chain the steps. We plan to use Spot capacity to reduce costs.

Planned on AWS

Phase 3: our own models

Amazon SageMaker training jobs will train small LoRA adapters, which adapt existing models, and a model that ranks clips.

The ranking model will learn from nflverse play-by-play data, including expected points added (EPA) and win probability, to find strong clips.

Why AWS

AWS would put storage, GPUs and models in one place, with Claude on Bedrock next to the media in S3. Managed job queues would help a one-person team avoid running servers. Credits would let us move the GPU work that runs on Runpod today.

Creator in control

Responsible AI

These rules guide the working app and every planned feature.

Working in the War Room

What you get

These tools work in our own app today. Early access for other creators comes next.

Live

NFL stats

Search and compare nflverse stats. Build your take from the numbers.

Live

Graphics studio

Choose from 77 graphics templates. Prepare graphics for OBS Studio, the recording app, or export them.

Live

Studio and live scores

Control OBS Studio and recordings from the app. Follow live scores while you work.

Live

Game prep

Bring matchup research into one place before you record.

Live

Ask and check

Ask Claude, GPT or Gemini for help. Check numerical claims against your saved evidence before you approve a post.

Live

Post kit

Prepare files and captions for X, YouTube, TikTok, Instagram, Facebook and Threads. You review the kit and post it yourself.

Built for our own channel first

One workspace, human judgement

Research, make the media, then review it. The final publishing step belongs to the creator.

Illustration of research, graphics and preparation for publishing

Workflow illustration

Get early access

Get news as we prepare ShareFuze for other NFL creators.

We store your email only to send early-access news. No spam.

The form opens an email draft until our sign-up service is ready. Send it to ceo@sharefuze.com.