Chris Green
banner
chrisukgreen.bsky.social
Chris Green
@chrisukgreen.bsky.social
Fractional COO and CTO in the daytime. Co-founder & MD of Etherlive for many years delivering technology to live events. Lifelong tech enthusiast, hobbyist GenAI and long term music DAW user.

https://substack.com/@chrisukgreen
I came across this finetuned VAE for Wan2.1 and Qwen Image almost by accident. It can do 2x upscaling in the latent space removing the need for UltimateSD and saving a lot of processing time producing similar if not better results.
Wan2.1/Qwen Finetuned VAE 2x Upscale Model
Upscaling Wan2.1 and Qwen Image in the latent Space using a custom VAE in ComfyUI reducing processing time.
open.substack.com
November 10, 2025 at 5:53 PM
This week’s topic is not often talked about openly: the stresses and mental impact of running a small business. Nearly all start-ups have lows at some point and making sure you survive personally requires resolve, coping strategies and support.
Lonely at the Top
The mental toll of being the one in charge
open.substack.com
October 31, 2025 at 5:37 PM
ByteDance's new USO Flux Model is not a full model but a model patch, a LoRA and two custom nodes creating a cross between Flux.1 Kontext, an IPAdapter and a ControlNet.
USO (Unified Style-Subject Optimized) Model
ByteDance's approach to style transfer and subject consistency
open.substack.com
September 22, 2025 at 3:49 PM
The addition of ControlNets for Qwen Image extends the model's flexibility and power. Three different approaches are available providing many of the normal features such as depth, canny and pose, all of which are covered in my latest post.
Qwen Image ControlNets
Introduction to extending the power of Qwen Image with ControlNets
open.substack.com
September 16, 2025 at 10:48 AM
Flux.1 Kontext has two new challengers in the form of Qwen Image Edit and Google’s Nano Banana. Does Flux.1 Kontext remain the standout leader for image editing?
Qwen Image Edit vs Flux.1 Kontext vs Nano Banana
Is there a standout image editing leader?
open.substack.com
September 9, 2025 at 10:22 PM
Your path to success may not be what you originally planned. Assessing, evolving and even pivoting entirely are all part of the business journey.

Those decisions can be hard when you are so deep in the business but the discussions and decisions need to be objective to ensure you take the best path.
Starts Fade, Finishes Last
In business it's OK to change direction or focus
open.substack.com
September 2, 2025 at 9:14 AM
After reading of people raving about how good the Wan2.2 video model was at creating normal images I thought I should give it a try. Can it really outshine Flux.1 Dev, Flux.1 Krea and HiDream I1?
Wan2.2 - Text 2 Image Generation?
Using the Video Model for Image Generation
open.substack.com
August 9, 2025 at 11:06 AM
Flux.1 Krea Dev - A model with attitude, well more just a distinct style, aiming to bring increased realism and less AI feel. Does it succeed? Not time for a full review but some early testing.
Flux.1 Krea (Dev) - 'Opinionated' Model
A model with a distinctive aesthetic approach
open.substack.com
August 5, 2025 at 10:24 AM
Every small business owner hates cash flow management but it is one of the most critical tasks in the early years. There is no magic solution but there are some approaches to keep on top of it and to ensure you do not end up with a nasty surprise.
When Are You Paying Me?
Balancing cash flow in a small business
open.substack.com
July 30, 2025 at 10:25 AM
A review of the OmniGen2 model in comparison with Flux.1 Kontext (Dev). Is it a true Jack of all trades or does it end up being a master of none?
OmniGen2: Unified Multimodal Model
Are Dual Distinct Decoding Pathways the route to the perfect model?
open.substack.com
July 16, 2025 at 6:18 PM
Getting a business off the ground takes funding, getting the right amount from a good source can be a big hurdle. My latest post has some thoughts and stories around initial funding.
Funding: Too Much, Too Little, Just Enough
Finding the Right Start-Up Funding
open.substack.com
July 14, 2025 at 8:46 PM
Diffusion Doodles: WAN 2.1 VACE FusioniX

Short post covering some limited testing of the Wan2.1 refined VACE FusioniX model for video generation. As a VACE (Video All-in-One Creation and Editing) model it covers i2v, t2v, v2v, flf2v, control nets, scene consistency and more.
Wan2.1 VACE FusioniX
Extending Video While Maintaining Consistency
open.substack.com
July 8, 2025 at 3:57 PM
After the release of Flux.1 Kontext (Dev) earlier this week, a quick run through and test of some of its features. It doesn’t disappoint.
Flux.1 Kontext (Dev) Multimodal Image Editing
Unleashing AI image manipulation locally on your computer
open.substack.com
June 28, 2025 at 1:10 PM
Managing Python environments, packages and dependencies can become a big time sink when working with AI generation, especially as requirements become more and more complex. A short piece looking at a few of the options and my experience so far of using environment managers.
Conda vs. Pinokio: Choosing the Right Tool for Managing Python Environments
Reducing the Stress of Dependencies & Versions
open.substack.com
June 19, 2025 at 12:50 PM
If you work with multiple AI models, tools and workflow managers with dependencies and version conflicts frustrating you then Pinokio, a 'light virtualisation' app, is worth a look.

Pinokio can create isolated environments, automating installation and updates, letting you focus on using the tools.
Pinokio: A One-Click Playground for Running AI Models Locally
Isolating Your AI Sandboxes Simply and Effectively
diffusiondoodles.substack.com
June 8, 2025 at 2:48 PM
One of the hardest periods whilst running a company was when it had challenges and I felt powerless to make changes because of mistakes in our original structure.

Early decisions can live with you for a long time. Remove the emotion and take time to ensure clear, sensible, business decisions.
Company Structure: Mistakes Can Haunt You for Years
Why Founding Agreements Matter More Than You Think
notesfromasmallbusiness.substack.com
May 29, 2025 at 2:03 PM
Why Your AI Model Doesn’t Know How to Spell.

Diffusion models learn from images. Not from dictionaries.

They see that “text-like stuff” appears on signs, labels, shirts… But they don’t learn spelling, grammar, or symbolic rules.

It’s not stupidity, it’s just not what they were trained to do. Yet.
Dear AI: What Even Is ‘B1ork!’ Supposed to Mean?
Why AI can create a masterpiece but not write ‘Hello'
diffusiondoodles.substack.com
May 23, 2025 at 7:21 PM
You don’t need a brilliant idea to build a great business. You just need to do it better than the rest.

Stop chasing the perfect idea. Start practicing relentless improvement.

Successful businesses aren’t fixed, they evolve. Your best idea today might not be what makes you successful tomorrow.
Starting a Business: The Big Idea
Why Execution Matters More Than Inspiration
open.substack.com
May 20, 2025 at 1:07 PM
What is Your Business?

Not defining or losing sight of your objectives can seriously hamper a small business. It’s tough when you are in the early phases and all revenue looks good, but are you going in the direction you should be? A look at how my business evolved with the benefit of hindsight.
What is Your Business?
Understanding Your Direction and Goals
open.substack.com
May 16, 2025 at 3:27 PM
New substack post taking a look at some of the current video diffusion models available in ComfyUI and online

diffusiondoodles.substack.com/p/big-steps-...

#ai #comfyui #video #imagegeneration
May 16, 2025 at 8:59 AM
My second article in 'Notes from a Small Business' reflects on reasons for starting a business, motivation, character traits, founder age and the importance of understanding the journey, not just the end goal.
Why Would You Start a Business?
Reflections from a Founder
notesfromasmallbusiness.substack.com
May 9, 2025 at 10:23 AM
Some quick thoughts on the HiDream I1 and E1 diffusion models. Amazing to have another open source model of such a high quality. HiDream is big though - 17 billion parameters and 32GB in its full form, so it requires punchy hardware to run.
Quick Thoughts on HiDream-I1 & E1
Amazing output but quite a beast to work with
diffusiondoodles.substack.com
May 7, 2025 at 6:40 PM
HiDream-I1 isn't perfect and can fail in similar ways to Flux. Its prompt adherence is strong. Images look different to Flux but hard to say what is different! Created with no LoRAs or refined models. HiDream-I1 (Q8 GGUF), multi-stage sampling & Ultimate SD.
#GenAI #diffusionmodels #comfyUI #hidream
May 1, 2025 at 8:57 PM
Alongside my substack on image generation I am also starting a substack publication on small business based on my journey over the last 15 years. #entrepreneur #smallbusiness #eventtech
Notes from a Small Business: An Introduction
Reflections, Observations and Thoughts for those on a Start-Up Journey
open.substack.com
May 1, 2025 at 3:40 PM
Experimenting with the Q8 quantised version of Hi-Dream. Performance/quality trade-off is good.

#GenAI #diffusionmodels #imagegeneration #comfyUI #hidream
May 1, 2025 at 12:31 PM