We just had the best year ever for mascots

A look back at 2025's breakthroughs and why they make mascots inevitable in the year ahead

Hey there, if you’re new here, welcome to Marcel’s Lab, where we break down how to build a character driven brand powered by generative-AI.

A year ago, generating a consistent mascot with gen-AI meant:

• Painstakingly spinning the slot machine to get multiple angles from a single image starting point

• Training multiple generations of LoRAs from a crappy dataset of images to get half-baked consistency

• Still getting 47 variations of “close but wrong" when taking that model for a spin

• Abandoning your creative vision for whatever worked

Today? We’re in a completely different game.

Let me show you with Marcel himself.

Same prompt, different model, 1 year difference

That’s the gap we closed in 12 months.

• Consistent character

• Extremely detailed

• World understanding (e.g. added the flashlight)

• Simple instructions (no crazy prompting architecture)

• 4K quality (just zoom to see the difference)

Not to mention the ability to render text consistently, combine multiple images, upscale, etc..

The Creativity Explosion

Sketch → Full Universe

Ink’s sketch-to-world example shows what’s now possible: a napkin doodle to complete a character world in one session.

Node-based workflows finally match how brains actually work. No more fighting with text prompts. Show it. Build it. Move on.

Models like Nano Banana Pro (or Gemini 3 Pro) understand the intent and context. The shift to multimodal input (image(s) + basic instructions) removed the weight of “prompt engineering.” You show it what you want. It gets it.

Combining a character with style inspiration to create a unique look & feel

You can now combine reference styles, locations & characters and blend them like an alchemist. The model does the heavy lifting.

Want to give it a try? Head over to Marcel’s sandbox and try it for free.

Just some of the latest creations of Marcel fans

Consistency: Actually Solved

This was THE blocker. Editing models like Nano Banana Pro and Seedream cracked it.

Same character. 50 different poses. Zero drift. No more custom-trained models. You can generate a single character sheet from one image and use that character sheet anywhere to generate consistent characters.

Build your character sheet from a single image - just ask for a character sheet based on the single image.

Video Control Got Surgical

First frame/last frame control changed everything. But hybrid performances and advanced video editing pushed it further.

We went from slot machine to smart planning:

• Storyboarding tools plan entire narratives

• Direct scenes like an actual filmmaker with first/last frame and video references

• Control character movement frame-by-frame

• Make your character speak, driven by voice samples (can be human-recorded or AI-generated) with models like Hedra, Kling Avatar or Fabric

See the full behind-the-scenes breakdown of my latest short film, ‘The Bubble’ to see this in action in the article below.

[Classified] The Bucke Up Briefing #10: Characters Are Eating the Internet (And Getting Zillow Listings)

Snippet from my latest work for AI sales assistant Donna. Full video here.

What is still broken?

• Audio quality on img to video models like Wan 2.6, Kling 2.6, and VEO 3.1 still sounds robotic. Currently, it’s better to use Lip Sync models as listed above.

• Character consistency in animation, especially facial features, is still an issue in video generations. Models often default to generic cartoon eyes and mouths.

Coming soon: As VEO 4 will have more understanding, like Nano Banana pro, I’m assuming it will be a lot better at solving both problems (audio and character consistency beyond the understanding of the img reference).

👀 Looking ahead to 2026

Here are some things I’m looking forward to in 2026 that will truly level up your mascot game.

World Building

This is not yet there yet, but I think it will be big in 2026. World models like Genie 3D (Google Deepmind) and World labs (Gaussian splatting) let you create navigable 3D spaces from 2D characters.

Your mascot will be able to exist in a consistent place, not just a flat image.

Character universes, not just character images.

Live Avatars

Real-time avatar solutions are showing serious signals. Genie, Avatar models, Runway experiments.

Zero-latency character performance isn’t solved (and mostly awkward, trust me). But we’re 6-12 months away, not years.

So What Does 2026 Look Like?

If you were subscribed to this newsletter, you might recognize the playbook for character-driven brands in the era of AI:

• 🐨 They will become more Distinctive

  • More unique, creative characters and worlds. The barrier to distinctive IP just collapsed.

• 🎪 They will become more Entertaining.

  • Unique stories driven by unique characters and IP. No infinite slop factory.

  • Longer form, higher quality, actually watchable content.

• 🕹️ They will become more Playable

  • Live avatar solutions: Real-time character interaction moves from experimental to standard.

  • Characters will cross more and more into physical space. Robotics is finally getting interesting. Not because the tech got perfect, but because designers stopped trying to make it perfect. Brands like Sunday and Disney are getting it right.

The AI companions are coming...

And of course, more slop (Merriam-Webster’s word of the year. We called it.)

Marcel: Physical me would absolutely have weird quirks. That’s not a bug, that’s branding.

---

Here’s the thing:

When everyone can make infinite content, creativity becomes the ONLY moat.

2025 proved that AI-generated characters can be:

• Consistent and qualitative enough for brand work

• Creative enough to stand out

• Efficient enough to scale

• blended into physical reality

The brands investing in distinctive character IP and universes right now will own 2026.

Everyone else will be making slop.

Marcel: And I’ll still be here, being handsome and correct about everything.

---

Ready to build your brand universe?

I help companies create distinctive brand universes and characters using Gen-AI. The kind that is actually used (not stuck in a Figma file somewhere). If you’re exploring this for your brand, book a 20-min call or check out some examples here.

And of course, you can subscribe to the newsletter to learn how to build a strong brand powered by Gen-AI.