Home Creators Posts Import Register Favorites Logout

Downloads

Content

Video : https://youtu.be/VtA9vfOVWnU

Hey everyone!

New video is up, and it's a head-to-head between the two image models everyone is downloading right now: Flux 3 Image Klein and Qwen Image 2.1.

I ran both raw. Same prompts, same seeds, side by side. No LoRAs, no extra samplers, nothing dressed up. We test short prompts against long ones, rain and water physics, dancing figures, hands and faces, text on signs and menus, and image editing with single and multiple reference images.

One honest finding: Flux 3 Klein was the more stylish text renderer, it even got the neon sign and scoreboard clean where Qwen slipped. But the water drops looked plasticky on it, and human anatomy is genuinely hard for it at this size. Qwen Image 2.1 held clothing detail better in the two-reference outfit swap, and its faces were just... there.

Best part of the whole video is watching two different models melt the exact same card tower in completely different ways.

Links to everything posted below. See you in the next one.

— Benji

Resources:

black-forest-labs/flux-3-image-klein-comfy

https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy

black-forest-labs/flux-3-image-klein

https://huggingface.co/black-forest-labs/flux-3-image-klein/

Files

Flux 3 Image Klein In ComfyUI - Setup And Test Against Qwen Image 2.1

Flux 3 Image Klein vs Qwen Image 2.1 tested raw in ComfyUI — same prompts, same seeds, no LoRAs. Full setup, the new prompt upsampler, and honest side-by-side results. Black Forest Labs just shipped FLUX 3 Image Klein, the 5.6B distilled image model with the new Qwen-based prompt upsampler, and I put it straight into a head-to-head against Qwen Image 2.1. In this video we load the official ComfyUI repo files, pick between the base and distilled checkpoints, and run group tests: short prompts versus long-form captions, rain and water physics, dancing figures and hands, text rendering on signs and menus, and image editing with single and multi-reference inputs. This is for anyone running ComfyUI locally who wants to know what these models actually produce out of the box — no curated showcase images, no sampler tricks, just same seed comparisons you can reproduce. If you followed my Qwen Image 2.1 tutorial, this is the direct follow-up; if you are new here, the setup section covers every file you need from the Hugging Face repo. Why it matters: FLUX 3 Klein is the first mainstream image model where a second language model (the prompt upsampler) sits between you and the generator, and BFL says not to run it without one. Testing both approaches side by side shows where the upsampling helps, where the small DiT still struggles with anatomy and water, and which model keeps clothing detail across multi-reference edits. Resources: - FLUX 3 Image Klein ComfyUI repo: https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy - Original FLUX 3 Klein repo: https://huggingface.co/black-forest-labs/flux-3-image-klein/ - Prompt upsampler (PUP): https://huggingface.co/black-forest-labs/flux-3-image-klein-pup - Black Forest Labs announcement: https://bfl.ai/blog/flux-3-image-klein Calliope - Local-first Agentic Idea-to-video Studio https://github.com/benjiyaya/Calliope If You Like tutorial like this, You Can Support Our Work In Patreon: https://www.patreon.com/c/aifuturetech #Flux3 #ComfyUI #QwenImage #ImageGeneration #BlackForestLabs

Comments

No comments found for this post.