Artificial_Monkey

Midjourney DALL·E Ideogram Flux SDXL Turbo Stability Diffusion Runway Gen-3 Alpha Runway Gen-3 Pika 1.0 Pika Labs Luma Dream Machine Sora Veo Krea Kaiber Leonardo AI Kinetix Synthesia HeyGen Scenario Inworld AI Playground AI Kandinsky Lightricks LeiaPix PromeAI Visions of Chaos DeepFloyd IF InstantID IP-Adapter ControlNet ComfyUI

OpenAI Anthropic Google DeepMind Meta AI Microsoft Azure AI Amazon Bedrock xAI Mistral AI Cohere Hugging Face Stability AI Runway Perplexity Replicate ElevenLabs Jasper Notion AI Canva Magic Studio DataRobot NVIDIA NIM

Beer Workflow // 12h // krea.ai / Midjourney / Elevenlabs

In this workflow test I needed to see if current ai platforms can deliver a proper ad. It was a challange to use a bear and force it to do stuff they normally wouldn´t. The last shot had to be veo3, because other video models lack emotional depth in their renderings.

While it´s far away from being production ready, I got it to 80% in only 12h. The last 20% would be spent in generating more refined shots, better physics and color grading. All in all i would say this could be done in 40h.

WORKFLOW

Old-school storyboarding
Midjourney for frame creation
Chat GPT for narrator script
Elevenlabs to turn skript into speech
Udio for music
Different Video Models to generate frames
PremierPro to glue it all together

Guinness Workflow // 16h // krea.ai / Midjourney /

I always LOVED the old Guinness ads. Always reminds me of the good old days when ads were allowed to be art in the first place. For this I wanted to test out Midjourneys moodboards and also try to recreate the old look. I used screenshots from the original ad to train the AI and then created the 3 firends, only 30 years later. The bioggest challanged I faced here, was the restrictions of creating old people in possibly dangerous situations. This was a huge time consumer because I had to dance around my prompt to force the ai to produce what I wanted. Another problem where the screenshots I used to train the model. Since they where super old in 720p max, maybe even 320p, I had a badly trained model on the one hand, but also a perfect one to capture the old low resolution look.

This was made in September 2025, before Nano Banana! I´m confident that the same thing would take me half the time today and look way better.

WORKFLOW

Trained Midjourney moodboard with screenshots of the old ad
In Midjourney I then created my desired Frames
In Krea I used multiple video models to get those dynamic shots
Stiched it all together in Premier Pro

Ski Workflow // 8h // krea.ai / Midjourney / Elevenlabs

This was a “quick” one. I wanted to test out Midjourneys style reference feature and came up with a short ski ad. For this I also used midjourneys moodboard feature to lock in the look I wanted.

I realise that the more mainstream you go the easier it gets. Obviously, because the models are trained on “mainstream” media that should have an abundance of ski clips. Now try to have a bear riding those snowboards and watch the struggle unfold 🙂

The music track surprised me though. I did it in elevenlabs and was super pleased with its quality. Wear your best headphones and enjoy the bassss

WORKFLOW

Generated 20 ski images in the style I wanted
Trained Midjourney moodboard with those
Generated roughly my frames – no storyboard here
Made a banger track on elevenlabs
Stiched it together in After Effects and added a couple transitions

Crunch Workflow // 2h // Higgsfield / Krea

A really quick one to test higgsfield images and nano banana. Simply made a woman and a human monkey, designed quickly a chips packaging with nano banana, placed it in situ and let it run through Wan 2.1

It´s not perfect obviously, but as a rather quick shot it´s usefull to visualize an idea. But one problem I had, was the transition and it failed to do it the way I wanted. This isn´t perfect, but it was the best outcome without burning through my credits.

WORKFLOW

Higgsfield for Image
Nano Banana for product placement
Wan 2.1 for transition