ComfyUI Image Generation Training: Z Image, Flux, Qwen and Krea

Our ComfyUI image generation training teaches adults, company teams and enterprise creative departments to produce campaign-quality images on their own hardware with the current open models: Z-Image Turbo, Flux 2 Klein, Qwen-Image 2.1 and Krea 2 Turbo. Taught by a creative director with over 20 years of advertising experience in motion and 3D, the course is built around the decisions a professional makes: which model suits the brief, which settings control the look, and which licence allows client work.

Here is one of the benchmark exercises attendees run in class, using the same prompt across models so the differences are easy to see. Click any image to view it larger.

Exercise 1: The Beauty Portrait

The prompt reads like a photographer’s brief: a close-up of a young East Asian woman, chin lifted, three-quarter view, gaze off camera, dewy skin with visible texture, a loose updo with flyaway strands, soft frontal light, a warm beige backdrop and a 100mm lens at f/4.

Z-Image Turbo gives a flawless, retouched beauty look, with a neat updo, glossy lips and smooth, dewy skin. It is close to a finished skincare key visual as it comes.

Qwen-Image 2.1 follows the loose hair and flyaway strands more literally, with a messier, more editorial hairline and visible texture across the cheeks.

Krea 2 Turbo looks the most like a real photograph, with fine flyaways, natural highlights on the skin and a slightly filmic tone.

For East Asian faces, Z-Image Turbo and Krea 2 Turbo were the strongest, so the choice comes down to whether the brief wants idealised beauty or natural, believable people. Flux 2 Klein struggled with East Asian faces, which is worth knowing before you choose a model for Singapore work.

Mind Theory is Singapore’s pioneering AI and creative-tech education provider, established March 2023, with hundreds of students across primary, secondary, and adult programmes. Read about Us.

The same prompt with a male model

Z-Image Turbo gave a sharp, idol-style portrait with short hair, and ignored the updo in the prompt.

Krea 2 Turbo was the only model that tied the hair back as the prompt asked, and it added the most skin detail. Attendees learn how to fix a missed detail like Z-Image’s through wording, because at the settings these fast models use, a negative prompt has no effect.

Z-Image Turbo at its full 2048 pixel output. Open it larger and look at the lashes, brows and lip texture.

Exercise 2: The Fashion Catalogue Shot

The second prompt is a full-body catalogue brief: a model in a minimalist studio, one hand in pocket, an oversized neutral knit and wide-leg trousers, a key light from the left and a seamless beige backdrop. It tests anatomy, how clothing drapes and whether the light comes from the right direction.

Flux 2 Klein 4B handled the light direction best, with a hard key from the left throwing a clear shadow across the backdrop. It is also the fastest of the four, ideal for quick concept rounds.

Krea 2 Turbo looks the most like a real catalogue photograph, with visible knit texture, film grain and a natural stance.

Z-Image Turbo gives a clean, evenly lit image that is ready to drop straight into an online store.

Want to create images like these yourself? Email us or WhatsApp us about the next ComfyUI class.

The Settings You Will Learn to Control

CFG sets how strongly the model follows the prompt. Fast distilled models run at 1.0, and pushing it higher makes colours oversaturate and skin look plastic.

Steps decide how long the model refines the image. Distilled models finish in 4 to 9 steps, while Qwen-Image 2.1 needs around 25.

Sampler and scheduler. We use Euler with the simple scheduler as a baseline because the same seed always gives the same image, which makes revisions controllable.

Resolution and seeds. Each model has a pixel budget it works best at, and recording seeds lets you recreate an approved image exactly.

Licences: What You Can Use Commercially

Open weights do not always mean free commercial use, and we cover this in class. Z-Image Turbo and Flux 2 Klein 4B are Apache 2.0, so commercial use is allowed. Flux 2 Klein 9B and Qwen-Image 2.1 are non-commercial unless you obtain a commercial licence. Krea 2 is free for commercial use only for smaller organisations, with larger companies needing an enterprise licence. Always check the current terms on each model’s official page.

Ready to start? See the full lesson plan on our ComfyUI course page, or ask about ComfyUI corporate training for your team.

More from this ComfyUI course series: turning images into video and image editing.

Established in March 2023, Mind Theory is Singapore’s pioneering AI education provider, offering Gen AI courses and corporate training for adults, working professionals and marketers.

Attendees build web apps through vibe coding and create production-ready AI visuals using prompt frameworks and AI-powered tools. To find out more, visit our Courses page.

Contact us to book a class by email or Whatsapp..