​

Welcome back to Aituts, your weekly dose of creative AI tuts and news:

Today's programming:

• 3 things to know about Midjourney V6
• The best way to get consistent faces
• 16x image upscaling
• No, that AI-generated influencer was not making 10k/mo

HEADLINE

3 things to know about Midjourney V6

Midjourney V6 Alpha was released on December 21, 2023. A wonderful early Christmas present.

The reception has been overwhelmingly positive. Now that the dust has settled a bit, here are 3 very cool things to know about V6:

1/ To make it more realistic, make it uglier:

V6 is the most realistic model to date.

But there's realism and then there's "Midjourney realism". You know, that glossy, over-saturated Midjourney look.

User SandwichTough9171 says: "OMG. Im having literally issues because V6 makes everything extremelly cinematic and unreal hahah".

To avoid this and get photos that look like real photos:

• Mention styles of raw, unedited photos, eg. "phone photo", "posted to reddit"
• Add "--style raw", which is an alternate style that has less beautification applied and is more accurate to your prompt
• Set the stylize to 0 with "--s 0". Stylize is what "beautifies" the image and gives you the Midjourney aesthetic.

2/ Clutter may make your results worse:

David Holz, Midjourney founder, says in the V6 announcement: "Avoid 'junk' like 'award winning, photorealistic, 4k, 8k'" and "You will need to 'relearn' how to prompt."

V6 interprets things much more literally than previous models.

That's great if you want to include a lot of details and object relationships like "above" and "inside", but not great if you were used to stuffing your prompts full of strange words.

3/ "Traditional" work is SO good:

It took 6 versions, but paintings really look like paintings now:

Just look at the brushstrokes on these images:

And this 'terrible' child's illustration:

The big theme with V6?

Before I could always tell which images were generated by Midjourney.

Now I'm not so sure.

​Midjourney V6 prompt ideas​

FEATURED TOOL

AI-generated flashcards

Wisdolia uses AI to generate smart flashcards from any article, PDF or Youtube video.

Any flashcard question you answer, you get smart feedback on:

It has generous free use limits, and it generates your flashcards almost instantly.

I was surprised at how easy it was.

​Try it for free​

TO TRY

IP-Adapter is incredible at face consistency

Here's a problem we've all had when generating with Midjourney or Stable Diffusion:

The images look great, but how do you keep the face the same across multuiple generations?

I've discovered that using IP-Adapter is one of the fastest and most effective ways of keeping your face consistent.

A practical way to describe it is "single image fine-tuning".

Instead of training models with 20-1000 images, all you have to do is input a single image.

​How to use IP-Adapter for Stable Diffusion​

TO TRY

Magnific AI's newest upscaler is magical

Magnific AI is many peoples' choice for upscaling Midjourney images, and for good reason.

Magnific's newest tool can increase image resolution by 16x (sadly, this feature is only available to early beta users at the moment).

​Try Magnific AI​

Now technically, Magnific AI isn't "upscaling". The app is recreating the image at a very high resolution using img2img in the background.

Since Magnific is a paid tool, the open source community has tried to get the same results with ComfyUI workflows.

Some people have come close, but it's harder than it looks.

One redditor concludes: "I thought this was going to be way way easier than what in reality has been, a nightmare." It looks like Magnific has some magic after all:

​Open source recreation of Magnific AI​

SHOWCASE

This is the best AnimateDiff video I've seen.

The creator @AIWarper hasn't said much about the process other than it's a lot of ComfyUI workflows strung together. [AnimateDiff in ComfyUI tutorial]

ROUNDUP

1/ Midjourney Alpha: Midjourney finally has a web app and the alpha version is available for anybody who has generated over 10k images to try. Better get generating!

2/ ComfyUI 3D Pose Editor is a new extension that lets you move a 3D skeleton around, and then use that pose in the images you generate. A big win for customizability!

3/ Comfy SVD Tools is another ComfyUI extension that lets you extend SVD videos (Stable Video Diffusion) to 4x the default length.

4/ The best ComfyUI workflows: OpenArt is an online Stable Diffusion generation service; they held a contest for users to submit the best ComfyUI workflow, and here are the results. [ComfyUI guide]

5/ No, that AI-generated influencer isn't making 10k Euros/mo. Guy finds that an extremely viral story about Aitana Lopez, an AI-generated model, was completely made up. Looks like influencers are here to stay, for now.

That's all folks!

You can reply directly to this email with feedback or comments. We read every reply.

Till next time,

Yubin & Crew

1223 Cleveland Ave #200, San Diego, CA 92103
​Unsubscribe · Preferences​