Mapping TRF's AI Model Garden against key client challenges has been worthwhile.??

AI Model Garden is Google Vertex AI terminology we have picked up, internally; we think of it as off-the-shelf AI models, prompt pipelines, finetuning, and AI options to explore.

TLDR: ?? Break down client challenges into bite-sized steps! ?? Start with the leanest models and gradually amp up the horsepower ??, introducing specialized or larger models to see if you can increase efficiency. #HuggingFace & #OpenAI are your friend.

In the influencer marketing software industry, this could mean 40+ AI models and agents working in the garden ?? in harmony ?? to judge if a creator is truly theright.fit for your brand… or just one model with a series of pipelined prompts.

Let's dive in:

A Daily Challenge for our clients is: Vetting influencers for success! ?? #ROI #ShowMeTheMoney?

The standard operating procedure is two-fold for vetting:

1. ?? Best practice metrics filtering think #EngagementRate, #AvgView minimums can cut down an applicants list
2. ?? Qualitative nuances (human in the loop) & AI subjective decisions we make that could/should be automated with AI?

?? Best practice metrics filtering refers to the tangible, quantitative measures we can use to quickly filter and rank influencers. For instance, high average views per tok/reel/snap can indicate an influencer's content is well-received and engaging ??. Two other elements we slot under filtering are demographics (aligning your brand) to the viewers of an influencer's content (age, location, gender) and previous brand collaborations an influencer has made.?

So, an influencer might boast jaw-dropping metrics, but if their vibe doesn't fit your brand or one viral video skewed their metrics, you may not get the market penetration you need for a successful campaign.?

Enter: "Human In The Loop" assessment and a few nifty #opensource AI-models ??.

?? Human In The Loop assessment (Qualitative nuances) This is (was), all about individuals selecting influencers whose personal brand and content align with your company. It's subjective, it's deep, it's pure art ??. Everything from aesthetics, the editing finesse, and the composition to thematic vibes play a role; consistency, setting, and audio quality – all come into play when selecting an influencer to collaborate with your brand.

I am a helpful AI... enthusiast who is using AI models daily. Here is my list of qualitative elements of influencers' content you can assess with existing AI models/services…

Where AI can help today in a production/low code pipeline:

• Transcript analysis: allows for sentiment, controversial topic monitoring, and entity extraction (brand names/locations etc). OpenAI's Whisper & GCP Cloud Speech-to-Text are your go-to models; HuggingFace also has a few open-source versions
• OCR text extraction for overlays
• Object detection, colour schemes, backgrounds and settings can all be handled by Amazon Rekognition, GCP Cloud Vision, and Clarifai are leading the pack

Not available to public via API but has the highest likelihood of replacing your standard operating procedures is...

• ChatGPT and check out the new image upload feature????, use a prompt structure along the lines of: I want you to describe this image to catalogue it in an image library, respond in JSON format. For context, the person searching for the image is in the X industry.?

{
  "type": "video_thumbnail",
  "content": {
    "scene": {
      "location": "outdoor",
      "features": ["garden", "terrace"]
    },
    "subject": {
      "gender": "male",
      "apparel": {
        "type": "suit",
        "style": "evening gown",
        "color": "rose gold",
        "pattern": "glitter",
}        

An easy low-code, no-code pipeline you can create to make use of some of these AI models now:

• Drop all the links of potential influencers partners into a sheet/table.
• AirTable or n8n to manage triggers and API calls.
• Extract the data you wish to assess.
• Run the models?(prompts, processors). This is where you can break up each of the qualitative assessments you are currently making into small prompts that will increase efficiency in your business. For example, you could have a processor assess if the colour scheme fits your brand or if foul language is detected in their transcripts.
• Get all responses in JSON.
• Parse the JSON if required.
• Take action

#AI-washing #Rant Many of the metrics and filters mentioned above companies are claiming to be AI. They aren't; AI comes into play when assessing the visual aesthetics, distinctiveness, and audio of an influencer's personal brand.?

The Grand Finale: use ChatGPT more often to see what is possible; test in a low-code environment like AirTable to know if you can get repeatable results.?