AI Video Generation Tutorial: (There is some demonstration speech in the tutorial, if you want to turn the audio on)

We get asked a lot about how to make the AI videos, and what AI programs/apps/websites are used?

Hope this tutorial helps.

Need to remember:

Some of these are free, some of these are expensive, some of these are higher resolution, and some are generated low resolution - for cost.

None of these are edits (except for the Wan 2.5 example, for modesty), some simple editing techniques are shown at the end.

In these examples Kling, Wan and Grok can do elements of voiceovers, with varying degrees of functionality, permissions, tolerance, moderation and language.

All videos are 5 seconds (LUMA, KLING), 6 seconds(GROK), 8 seconds (Pixverse) or 10 seconds (KLING, WAN) generated depending on the service used, what's possible and what's paid.

Sometimes we use a Service directly e.g. Kling or sometimes through an aggretator website like POLLA or HIGGSFIELD. I don't do local on-machine generations or animation or use LORAs or anything like that.

Each individual service usually have different options and versions available, and multiple different subscription tiers.

Depending on the prompts, and a random chaos/idea generator concept called the "seed", (that you have none or little control over) basically you're putting the image into a "magic box: and waiting to see what is churned out.

Then you look at it in disappointment and wonder how to salvage it. But like any addiction every so often there is a reward.....

All except for the Wan 2.5 example were generated with the same prompt, (Wan 2.5 tolerates certain other more anatomic "functionality".)

What are the best?

- - - well Kling (usually 2.1) and Wan 2.5 for permissiveness- at least in my experience.

We don't use Veo as yet, and Sora is as shown.....

TyrannoMax Trading Card & Sticker Album, Stikks™ Company, Inc. Selections from Set 1: Fear the Roar!

The Stikks Company was best known for making sports cards, but they branched out in the late 70s by getting the contract to make trading cards for Cocytus Comics' Nightmare Pit and Tomorrownauts titles (a mint Nightmare Pit series 1 Farrah Fyendlyne vs Glodie Redeemer (37) is $300+ on the secondary market, yow!).

This got their foot in the door for the 80s when the Cocytus / Buzby Spurlock Animation merger put the Cocytus characters on TV sets and toy shelves (semi)worldwide. Buzby-Spurlock needed merch liscenees and Stikks got most of the trading cards and sticker albums.

As with most things, Stikks was acquired by Hasbro in the 90s.

EXTENSIVE Process/Tutorial under the fold.

In the interest of community and education, here is an example.

These four images were the result of one submission of one prompt - I didn’t have to wrestle the machine for them at all. The prompt is:

underexposed Polaroid, side view from far away, two Icelandic bodybuilding bros facing each other submerged near a hot spring, enormously muscular, golden light, loving embrace, buzzed blond hair, relaxed, unbelievably enormous muscles, muscle morph, leg muscles like enormous heavy water balloons, enormous muscular arms, high body fat, leaning against each other

Now be warned, this is a bit of a jenga tower. Moving things around too much may break it. I’d recommend writing your own from scratch but stealing specific key phrases, modifying and evolving those, see what works best for you.

Thanks to @thespacewerewolf for the “near a hot spring” trick to get them into a hot spring, and to @zangtangimpersonator for the water balloon / weather balloon comparison trick, which is a Swiss Army knife of a prompt for anyone who likes big round shapes.

Changed up a lot over time.

The thing that worked best most recently was building a moodboard of screencaps from a potluck of shows, but for straight prompting I tended to have a format along the lines of:

(description of scene), vintage cartoon screencap, 1985 (series real or imaginary), by (pick two or more of Filmation, TOEI, AKOM, TMS, Sunbow Productions, Toei, etc, etc.), vhscore. vintage cel animation still.

Usually at --ar 4:3, under niji if you want it to look high budget, under the standard model if you want it to be a low budget ep. But moodboards and style prompting are your friends.

Disclosure: I am in the Vidu artists program, pretty sure this is technically sponsored content. This is a fixed version of the original post.

Of course, it's another Fauxstalgia Fake-Em-Up from yours truly, this time demonstrating the beta-access potential of the new Vidu Q2 reference-to-video mode (get 100 free credits if you click here to try it out for free).

So, lets talk about how it was done (an overview)

GIFs are unedited wherever possible to show raw results.

The big problem with using AI for anything actually productive is inconsistency. While there's always wobble and bad gens, Vidu's reference-to-video option is the best solution I've seen thus far.

Essentially, you build a little profile with up to three reference pics and a basic prompt for your characters, props, etc. From then on out you can use the reference in your prompts to generate new scenes. It decreases confusion about what character is what, and it lets you establish styles of movement and other effects characters should always have going without having to use prompt-space in every scene.

As you can see with Tilly Tepesh's jacket and the ghost of her uncle Drac's translucency, and with the fact that I can have four different kinds of near-identical blokroid in the same scene without them morphing into each other or making weird hybrids.

Images can also be used with ref-to-video without making a full reference profile. For the various Drac-and-Tilly playing scenes the robot kept making filly play the back of the keyboard and spaced them to far apart, so I composited a demonstration image together to help guide the process.

Which helped a lot.

I used a variation on my typical asset-creation process on this (tutorials), utilizing a lot of early gens and ref-to-images in Vidu's system (tutorial) to bulk up the process, along with Midjourney, Sora, and Civit.

For the audio, I recorded the lines myself, then loaded them into Suno to modify my voice into something more announcer-like and bring the music in under it.

All the editing, compositing, logos, etc were done the old fashioned way.

Quality Improvements (also an overview)

Basics:

Up to 8 second gens.

4:3 and 3:4 aspect ratios (in addition to 1:1, 9:16 and 16:9)

General improvements to quality, coherence, and prompt understanding.

Extend clips multiple times up to 5 mins (launched after the video above was finished)

While I'll get into the details of the improvements to quality as well as a number of other new features (up to 8 second generations, clip extending, and sound (very early on that one)) in other posts, those who have followed TyrannoMax may remember the issue with boulders.

Well, after a lot of coaching and getting Max a new trailer, he's finally stopped screwing around on set.

In short, a lot of things that just weren't doable before now work, and the general quality of gens is higher and sharper than ever before.

What I was not expecting was for it to emulate Gerry Anderson-style puppetry well. Ironically, the kind of jank you get from pre-digital media is hard for AI to duplicate. If you want something to look polished, smooth, and modern, that's easy.

It did take a lot of prompting for non-moving faces/immobile dolls, and editing is always needed, but the differences between Q1 and Q2 is apparent when you compare the same prompt in both:

The deadly starmantis gingerly setting the Queen Seltza prop down is adorable, but not what the director called for.

All in all, it took me about a week to go from concept to video, mostly on the back of my needing to make base character, prop, and set assets for everything that wasn't TyrannoMax more or less from scratch. If you want to give it a try yourself, here's that link again, and I'm posting a bunch of tutorials, old and (hopefully) new this week.