"How To:" (make an amazing song w/ music video) + some OpenAI speculation, news and rumors

I′ve recently begun generating and sharing cartoons & comics about AI. generated by AI (driven by my own creativity, time & talents as well - they′re certainly ′human-in-the-loop′) as well as producing music videos & songs that I use AI to help generate as well, & am having so much fun that I decided to publish an issue of Tech For Good dedicated to how I′m generating them so that anyone else can make them too if they so desire, but also just to help more of you fully embrace & adopt AI into your lives and workflows, & to have a better understanding of how to use and leverage existing tools for things like marketing creative, branding, mixed media for your & your businesses′ social media, prototyping & much more.

To see my most recent music video "It′s Taking Our Tasks (It′s Taking Our Jobs)" check it out here or in my recent LinkedIn post about it here: https://www.dhirubhai.net/embed/feed/update/urn:li:ugcPost:7258505403288670208; to see the cartoons & comics, check out the new page Techitoons (my goal is to make this the most fun & collaborative business page on the platform - to this end, I welcome submissions to feature on the page for those creating cartoons or comics about AI using AI, & if you don′t already know how to do so, just keep reading).

But before I go into the tutorials, I will share some of my predictions & speculation with regards to OpenAI

& the state of AI as we move into the final months of 2024, & in anticipation of accelerated developments moving into 2025so that those who already know how to make music, videos, & how to use AI proficiently, or simply who don′t have interest in the same, can get what they came here for too (this is for everybody, & I must clarify that these are not factual, but speculations and predictions that may or may not be wholly accurate, & in no way is officially affiliated with or informed by anyone at OpenAI ):

• when OpenAI announced Sora 263 days ago, they did make it clear that they did not intend to release it prior to the 2024 U.S. Presidential Elections, to ensure the tech wasn′t used to make deep fakes or otherwise risk potentially impacting the election. The elections are tomorrow (more accurately, conclude tomorrow), and while OpenAI has seen departures from several key personnel recently, including co-lead on Sora, Tim Brookes (who left to go to Google Deepmind), it has been my belief (& hope!!) that Sora will be released shortly after the elections - possibly as soon as this week. While other video generation models have emerged & indeed got better, we haven′t seen any on the level of Sora yet, & this would be the time for OpenAI to reassert dominance. Also, the fact that Google has struggled developing their own text-to-video models, their recent addition of Mr. Brookes will almost assuredly put more fire under OpenAI to release Sora (although the company claims to be in dire need of more compute)
• this weekend, some of us got a brief glimpse of & access to what is likely ChatGPT 4-full, with superior reasoning, full integrations of Canvas, Voice, & Browser. realtime API & again, I believe it will likely have Sora and potentially Dall-E 4 next generation text-to-image/image generator. You may be wondering - but wait, hasn′t ChatGPT had browser & voice for a while now?? & you wouldn′t be wrong, but voice is now available on desktop & across all models, search has been upgraded & is now available in all models as well (including free ones I believe) & Canvas is a pretty big deal, whilst it′s also worth noting that many other AI platforms & companies are also launching canvases, so this seems to be a pervasive theme in AI moving forward
• Sam Altman has already teased o2, which would logically follow the current o1 naming mechanism, so there may never be a ′ChatGPT5′ but whatever the next model is will assuredly be a next step closer to AGI (artificial general intelligence, but you must already know that). OpenAI is seeking to substantially up their compute power, which often happens before they begin testing & releasing a new model. That said, there′s speculation that the next model will not be available to all users, but rather it will likely be licensed to companies to use & hopefully embed & provide we the consumers access to it & it′s capabilities.

Please do bear in mind that all of the above is speculative, not verified, and subject to changing or being incorrect. These are informed hypotheses, no more, no less.

And now, how I made this amazing song & accompanying decent video

& how you can too.

First there′s the song generation. There are two equally amazing music generation platforms I′ve highlighted here several times previously: namely, Suno & Udio , & while I genuinely like them both, I find myself using Suno more often, & it′s what I′ve used to make virtually every song I′ve shared publicly. To use Suno to it′s fullest capability, I do find myself wanting to be the creative, & so I typically provide the lyrics, rather than having it generate lyrics (or having Claude or ChatGPT do it) - but I am a lyricist, songwriter, freestyle rapper & have a passion for it, so I know that I have a human advantage there. That said, when you have it generate it′s own lyrics, it sometimes knows how to time them & deliver them better, so it′s worth seeing what it can come up with for you, particularly if you′re not inherently a wordsmith.

The secret to a great song generated is the prompt that you give it for the style.

There are certain genres it′s less familiar with (& some, like Amapiano that it doesn′t even know, yet), so don′t ′dig in the crates′ too deeply - give it what it will understand. Be as creative & descriptive as the number of characters allows, but do document or recall things it doesn′t seem to ′get right′, as these will trip it up. For example, I really want a beat made of ′angelic female choir vocals looped′ but it never gets any of that right, so although I take that as a challenge & keep trying (the definition of insanity, lol) I know that when I′m composing a generation I will want to use, I need to stick to descriptions it can process. Also, I′m having some success asking it to sound a bit like certain artists or songs, but often that rightfully triggers it to not generate that song for copyright purposes et al., so it seems much more effective to escribe the artist, sound, or song as succinctly as possible & let it make inferences.

Example of a good prompt: A soulful 70s inspired R&B song about gratitude and being smooth; sing-along catchy chorus, funky bass groove, mention (your name/partner/company)

OR

the prompt I used to create the song in the video was: "Hip-hop top-40 mashup with samples from the news/debates, wicked turntablism and a danceable beat with sick drops"

(PRO TIP - to see what prompt you used in a previous generation, click the three dots (...) & click "reuse style")

Once you have a song that you love, you may want to create a video for it. A great way to do this easily is to tell ChatGPT what the song is, ask it to use Dall-e to create some images for a music video for the song, then upload those into an image-to-video generator like Runway, and/or just give Runway (or others like Kling) prompts pertaining to the theme of the song & let it generate video clips for the music video. Once you have the song file (download it as an mp3/mp4 locally to your files) & the video snippets for the music video, you can open your preferred movie editor (Adobe is good, or I use iMovie) & align the video clips & song, adding cut-aways, transitions, captions (I like the app ′Captions′ but there are others too) & you're ready to go.

It can be a bit laborious, but is a labor of love, & the more ′say′ you have in the creative process from the lyrics to the video content etc., the more connected to it you′ll feel. These songs & music videos are fun & sharable, but can also be done for corporate videos, commercials, fun activity with your kids & much more.

Have a cool use-case, idea or creation? Please share it in the comments, or with us directly.

For the cartoons I′m creating, I′m using the same method as above. For the cartoons (inspired by The Far Side, Calvin & Hobbes, Dunesberry et al.) I use Dall-E 3 to generate them with my creative direction, & then since the text is never usable (hence the lyric in the song above ¨But how can it work, when it can′t even spell¨) I go to Canva, find a shape the color and size of the text in the speech bubbles that needs to be edited, lock those in, find a similar font to the words/letters there I′m keeping, & then manually make them say what I wanted to, & then I share them on Techitoons (I really wanted the name Techtoons but it′s moderately taken).

A few days late for Halloween, here′s an unedited GPT cartoon prompted with the following:

"Think like the original Adams Family artist and make the most hilarious cartoon comic ever about Tech For Good newsletter, the weekly Linkedin newsletter about AI. Use Dall-E 3 and be confident that you′re a talented artist, are very funny, ensure you spell everything properly, and aspire to make the reader laugh out loud with this one. Cheers!" which does showcase several things:

• ChatGPT clearly infringes upon copyrights with cartoons - it′s way to similar to The Adams Family here, and when I′ve requested The Far Side, Calvin & Hobbes, & even The Simpsons in the past all only as frames of reference, it′s done the same. Thankfully Suno & the other music platforms do NOT do this to the recording artists, but these cartoonists deserve the same respect
• Dall-E definitely does not have a world model - in other words it doesn't understand "how things work" as evidenced by the book cover facing away from the reader (unless she's reading the back cover, I guess?), the "tech for good" paper is turned upside, obviously the spelling horrendous & also there shouldn't be two people attributed to speaking the same thing, & especially especially not when only one of their mouths are open!
• The model struggles with humor, but is clearly going for a pun with "eerious" for a spooky take on ′serious′

So in this case, if I wanted to salvage this for Techitoons, I′d probably get a beige rounded square to cover all the speech in the bubble & find a font that is cartoonish and simply write something like "When even The Ghoulish Family gets into AI you know it must be getting EEERIOUS"... or in this case, more than likely scrap it. With good prompts, often they′ll only need a few letters prompted, although sometimes it may take a few attempts to get there. We′re definitely at a point in the AI adoption where patience & perseverance pay off!

That′s all for this week - thanks for reading it! I hope you enjoyed it & have fun utilizing some of the strategies I′ve shared here if you didn′t already know & use them. If you really loved this, tell some friends (invite them by tagging them in the comments), make sure you′re subscribed, & feel free to leave feedback in the comments.