Workflow for MOSS Image
T2I + BBox, you can use the bounding box to control where the character is placed, or just ignore it and use T2I as usual.
Before use it, update your comfyui to the latest.
Example how to use it :
You describe what's in there and here, boom.
Too lazy? Just ignore the BBox and write what you want.
There's a LoRA node inside the subgraph. You can enable it and plug your LoRA in there. To access it, just click "Enter subgraph" in the MOSS-Image node.
To use bounding boxes feature, you need enable Node 2.0 in your menu settings.
The model link is already included in the workflow note. You can download it from there.
If it's not found, you can use visit the official Comfyui link to get the MOSS model here
If you encounter any errors or issues, feel free to leave a comment!
Description
- Fix 1:1 BBox ratio issue
- Fix dead links
- Add more LoRA slots
- Add BBox text preview
- More compact layout
FAQ
Comments (4)
I just noticed that when I change the size, the grid doesn't change. Even though I'm using the 9:16 aspect ratio, it keeps giving me square results. At first, I thought the placement of each sector would be based on hierarchical order, like layers, but when I select the back box (number 1), if box 2 is on top of box 1, it ends up going to the back, and to bring it back up, I have to shrink the shape of box 1, which ends up messing up the composition. If you could add tabs below the grid to organize each box and assign different colors to them, it would improve things a lot. It reminds me of the regional prompt that was in Forge, but getting it to work properly was a real headache. Do you also plan to add a mode for painting by layer on a canvas?
PS: I'd had an idea for a while, and that was to use up to a third mode—one in which you create a 3D environment with stick figures to better express body movement and gestures, with the option to add a 2D background, and then use a camera to properly adjust the perspective. If you took this further, you'd end up creating the ultimate tool for compositing, if you integrate this into a workflow with an image-to-video generator, you’ll definitely be able to make feature-length films— you just need to composite using each of your tools. If you add a fourth mode where you use the third mode but to generate keyframes, you can make the video generators look less artificial, resulting in a hybrid approach between traditional and artificial methods that can be even more valuable—since using only video generators limits you to a very artificial look.
If you're using the custom_node in the newest update, this is fixed! It auto-syncs with the ResolutionSelector. (If it doesn't change right away, try reloading the page). Or maybe you can use manual workflow which is you can change it directly.
Yeah, I feel you. If only the bbox had proper layer tabs/z-index controls, it would be so much easier. For now, the best is drawing the larger background box first and foreground elements last.
That 3D figure idea is wild! It sounds pretty close to a 3D ControlNet / OpenPose workflow.
@HissingCat I have Contronet and OpenPose on Illustrious; they're resource-intensive, take up space, and don't work well with styles and concepts, However, if you manage to get the other modes I mentioned working, it’s very likely you’ll become the go-to solution for this generative model without needing to use pre-trained models. Find a nice logo, and you’re sure to hit half million downloads if you can connect with other models to create the composition.
Brutal workflow my dude, prompting using the boxes is way easier, thank you.