If you know basic HTML/CSS & web design, Ming-Image for ComfyUI is a game-changer!

#3
by LabMike3D - opened

Ming-Image for ComfyUI is a pretty powerful tool. If you have some experience with web design and a basic knowledge of HTML/CSS, this is the right tool for you. Instead of writing a vague, standard prompt, this model lets you feed it absolutely exact json instructions regarding the placement of individual elements. Whether you are building an app UI panel, a website preview, or a poster layout, the structure remains perfectly preserved across subsequent generations. Only the specific changes you request are applied.

While it is still not 100% perfect and text generation can still hit occasional errors, in layout structure and text placement are far less frequent. It is a massive step up from the frustrating experience of traditional image models that randomly scramble your composition with every single click. For professionals with years of experience in Photoshop or web development, this level of control is a total game-changer.

I tested Ming-Image in ComfyUI for several hours today, and I'm thrilled that we finally have a tool, or rather, a model/text-encoder combo inside ComfyUI that actually listens to its master.

By the way, I didn't rely solely on ComfyUI for this workflow. To generate the precise prompt structure, or rather, the required JSON file I used Llama.cpp running Qwen 3.6. This is actually my go-to local combination now, and I regularly use it to write text and craft highly structured scripts and prompts for video models like LTX 2.5 and MiniMax H3. Having this kind of local pipeline gives you unmatched flexibility.

Have fun and keep rocking!
1

wow cool now I can generate image of flexbox

Ming-Image for ComfyUI is a game-changer!

Have fun and keep rocking!

https://x.com/LabMike3D
https://www.youtube.com/@ComfyUIPlayground

Cheers!
LabMike3D

LabMike3D changed discussion status to closed

If you have some web design experience and a basic knowledge of HTML/CSS, the Ming-Image model is great for you. It allows you to forget about the endless guesswork and trial-and-error typical of using similar AI image-generation models in ComfyUI when creating layouts or posters and you can ditch Photoshop, too.

By the way, I didn't rely solely on ComfyUI for this workflow. To generate the precise prompt structure (or rather, the required JSON file), I used Llama.cpp running Qwen 3.6. This is actually my go-to local combination now, and I regularly use it to write text and craft highly structured scripts and prompts for video models like LTX 2.5 and MiniMax H3. Having this kind of local pipeline gives you unmatched flexibility.

Best of all, I run this entire setup on a completely standard RTX 3060 12GB, and I can get around 50 t/s for the Qwen 3.6 35B MoE quant.

1
2
4
3

Cheers!

LabMike3D changed discussion status to open

Sign up or log in to comment