I typed one sentence into 5 AI tools

StepAudio 3 Musica sentence becomes a whole song

Describe a song and it writes the lyrics, the melody and sung vocals.

My sentence: "A cheerful pop-punk song about my cat stealing my lunch, female vocals". It took 120 seconds and gave me a 2 minute 50 second track. One line: "Now he's licking his paws like he's won the show".
Free with no account for now. StepFun's own docs say it's "Free (limited time)" and that the preview will be retired for a paid version. Busy times can mean a queue.
Who owns the song isn't clear. StepFun asks for a copyright review before public or commercial use, so treat songs as personal for now.

Type what a person should do, and it generates the body animation, which you can orbit and export.

My sentence: "a person sneaks forward on tiptoes, then jumps back in fright". "Motion generation finished" came back in 3.4 seconds.
It's motion only, on a grey mannequin, and you get a 5-minute demo session. If playback looks choppy, set the speed to 2x (the page streams the poses).
NVIDIA's model licence says the models are commercially usable and NVIDIA doesn't claim your outputs.

Google's Gemma 4 (E2B) runs on your own GPU through WebGPU, with no server, and draws an editable Excalidraw diagram.

My sentence: "How a pizza order travels from the customer's phone to the delivery driver". It drew a sequence diagram in 47.8 seconds.
The first visit downloads a 3.1 GB model, which is cached after that. You need desktop Chrome with WebGPU and about 3 GB of free memory.

freetexttocad.com · free, no account · STL, STEP, OBJ

An AI plans the part, and a CAD engine builds it in your browser, with sliders to resize it.

My sentence: "A desk phone stand that holds a phone at 60 degrees, 80mm wide, with a slot for the charging cable". It gave an 80 × 100 × 80 mm stand with a 60° backrest and a cable slot about two minutes after I pressed Generate, and its own check passed.
Limits: 12 parts a day per device, and the full-strength AI once every 2 hours. It checks its own work, and if it warns "this may not match", press Retry.
Its terms say you own the models you make.

A 4B model writes a Three.js scene inside your browser.

My sentence: "A tiny floating island with a red lighthouse, a waterfall pouring off the edge, and a hot air balloon drifting past". On my M1 Air it wrote at about 3.7 tokens a second, and after 25 minutes it still hadn't drawn the island. A faster GPU should do much better.

More free guides: pranavinnovates.com