A New Video Engine

Introducing MiniMax H3

This week’s video almost didn’t come together because I don’t have that many credits left for the month after blowing through a remarkable number of them thanks to the uninformative inefficiencies of OpenArt’s Director, which spends your credits like water without telling you about what things are going to cost you before they’re spent.

However, the addition of MiniMax H3, which is priced halfway between the discounted Kling Omni 3.0 and the discounted Seedance 2.0 rate, happened to come with an offer of unlimited free generations for a few days, so long as the generations were limited to 5 seconds or less. And since 5 seconds is about the perfect length for a basic music video scene, which can either be shortened or lengthened a bit as needed in post, I was able to complete this video for the Cradle to Calvary song Book of Ephesians while spending about a quarter of the credits I had left.

The song has been around for about a year and was originally recorded in Suno 4.5, but I’d been intending to remake it with 5.5 for a while. The video was inspired by the image at the top of the post; it just seemed like it would thematically pair very well with the Christian nu metal of the song; as to how well it worked, I leave to the reader to decide.

BOOK OF EPHESIANS by Cradle to Cavalry

The music itself deserves some explanation, as this is the first time I’ve managed to produce a full two-vocal duet with Suno. However, I didn’t do the mix within Suno, although that is theoretically possible. Instead, I utilized the Cover feature to reproduce the same song with a different vocalist, extracted the vocal stems, and then mixed them in Reaper. This vocal extraction was going to be necessary anyhow, for the Clarify lipsyncing, and it turned out to be about one level more complicated than it ideally would have been, because the two vocal parts didn’t quite line up perfectly and required a bit of cutting here and moving a little bit forward or backward in time, depending upon the vocal timing.

The male vocal is the original, so I just copied each of the primary vocal tracks, then cut or pasted as required, first for the original song render, then again for the lipsync cuts. The Reaper screenshot shows how the process was a little less seamless than the theory suggested it might be.

The primary expense of the video was the lipsyncing, which uses the Clarify Aurora engine and is an absolute necessity despite its expense and its flaws. None of the other options, which include OpenArt Lip Sync, Kling Avata, Kling Lipsync V1, Byte Plus Omni Human 1.5, or Hedra perform anywhere nearly as well. It costs about 10 cents per second, more than three times more than the alternatives, and it’s definitely worth it.

Aurora is not perfect, as the frozen backgrounds and occasionally weird, out-of-control hand gestures indicate, and there isn’t any way of telling it what you want the character to be doing besides lipsyncing the audio provided. But it does exactly what it is supposed to do, which is provide a visually accurate lipsync of the lyrics being vocalized. The best way to get results is cut the lines from the extracted vocal stem so there isn’t any music surrounding the vocal in Audacity or some other audio editor, then upload that in addition to providing the lyrics and the starting image.

Perhaps the most impressive thing is that I didn’t require any second takes on any of the lipsync clips on the basis of Aurora’s performance. Although I will admit to being tempted to redo the clips of the duo in the mountains, since I neglected to notice that the male singer’s leather jacket had somehow been converted to a vest without my noticing in time.

As a general rule, you’ll want to have at least a close-up, and often an extreme close-up will work best. And I was pleasantly surprised that when I tried it with two characters singing in harmony, it worked pretty well; I wasn’t bold enough to try it when the two different singers are singing two different things and since there is no way to indicate that in the text box, I’m assuming it wouldn’t have worked anyhow.

So this entire video was produced for less than one-sixth the cost of the video for Soulsigma’s Perfection, under $15, even though it has more lip-syncing and required less post-production sleight of hand. And one thing that I have learned to utilize profitably is the Enhance Prompt option provided by MiniMax H3 and several of the other video engines, since it will take your basic description of the clip you want and put it into proper filming terms and provide various details that will not only improve the clip being produced, but also help you learn the terminology and the functionality involved.

So, for example this: As an ominous fiery light appears in the dark sky, the monsters look up, terrified, and then flee in every direction.

Becomes this enhanced prompt:

[Cinematography]: Wide cinematic shot opening directly from the uploaded start frame. The camera remains mostly stable with a slight, ominous tilt upwards to emphasize the sky, followed by a subtle, dynamic handheld shake as the creatures scatter.

[Subject]: The monstrous horde from the start frame, including the towering winged demons with glowing red eyes, the pale, gaunt ghoul in the foreground, and the silhouetted undead figures. Maintain their exact anatomy, tattered clothing, and terrifying features throughout the clip.

[Action / Transition]: The creatures begin in their initial menacing stances. Suddenly, they snap their heads upward in unified alarm, their expressions shifting to panic. They then break formation and flee frantically in every direction—some scrambling into the shadows, others taking flight, and the ghouls sprinting off-screen with erratic, terrified movements.

[Context]: The desolate, foggy wasteland remains the setting. The heavy clouds above dramatically shift as the cool moonlight is rapidly overtaken by a sweeping, ominous fiery gold light that washes over the entire scene, casting long, stark shadows and harsh, warm highlights on the ground and the monsters.

[Style & Ambiance]: Dark fantasy cinematic aesthetic. High-quality visual effects focusing on the dramatic lighting transition from cold blues to burning golds, realistic creature animations, dense atmospheric fog, and an intense, terrifying mood with perfect temporal consistency.

You can see the results in the video.

The video still required an amount of post-production editing, but in this case it was usually a choice between something that worked all right and something that worked better rather than needing to correct fatal flaws. The flaws are still there, such as the female singer having green eyes in one clip and blue eyes in another, the inconsistency in the appearance of the little girl across video engines, and an attempt to accurately recreate playing the guitar by utilizing guitar tabs to generate prompts failed comprehensively, but regardless, it is clear that both the technology as well as ability to make use of it appropriately are improving.

The tools utilized:

• Suno 5.5

• Bandlabs

• Reaper

• Audacity

• CapCut

• Seedream 4.5

• Aurora Clarify

• MiniMax H3

• Kling 3.0 Omni

You can’t run away from your demons. But you can absolutely turn them over to someone who knows exactly how to deal with them.

Ephesians 6:12