MOSS-Image’s layout/bbox feature only changes the prompt. <code>CreateBoundingBoxes</code> turns the boxes into a layout JSON, but nothing scopes a LoRA to a region. I wanted a two-character scene where each character LoRA stays inside its own box and a style LoRA covers the whole image, so I built a workflow for it.How it works<ol><li>Pass 1: the stock MOSS-Image subgraph draws the whole scene from the layout prompt, with the style LoRA applied globally.</li><li>Pass 2: the area inside box 1 is re-sampled with character LoRA 1 and its own prompt, using a latent noise mask.</li><li>Pass 3: same for box 2 with LoRA 2.</li><li>Pass 4: repeat and so on..</li></ol>Each pass only touches its own masked region, so the LoRAs can’t bleed into each other or into the background.The part that makes it dynamic A small custom node, <code>moss_bboxes_to_masks.py</code>, reads the <code>bboxes</code> output of <code>CreateBoundingBoxes</code> and builds the masks automatically. Drag a box or change the resolution and the masks follow, with no typing coordinates. It has:<ul><li><code>box_index_1</code> / <code>box_index_2</code>: which box feeds which LoRA pass</li><li><code>padding</code>: grow or shrink the box in pixels</li><li><code>feather</code>: soft mask edges to hide seams</li><li>an info output that lists the boxes it read, so you can check them</li></ul>Why not hook LoRAs? My first attempt used ComfyUI’s hook LoRAs (<code>CreateHookLoraModelOnly</code> + conditioning masks), which would let you do it in a single pass. On the int8-convrot checkpoint it crashes in the sampler with <code>'Linear' object has no attribute 'weight_scale'</code>. Normal LoRA loaders work fine on that checkpoint, so the two-pass approach avoids the problem. I haven’t tested the hook version on the bf16 checkpoint.Setup<ol><li>Put <code>moss_bboxes_to_masks.py</code> in <code>ComfyUI/custom_nodes/</code> and restart.</li><li>Load the workflow JSON.</li><li>Pick your LoRAs, fill in the two character prompts, and draw your boxes in <code>CreateBoundingBoxes</code>.</li><li>Check that <code>box_index_1</code> / <code>box_index_2</code> match your box order (index 0 is the first box in the node).</li></ol>Tips<ul><li>Put the style trigger word in the character prompts too, since the passes still run the style LoRA.</li><li>Don’t stack the turbo LoRA twice. The subgraph’s model output already includes it.</li><li>Passes 2 and 3 start at denoise ~0.5 to 0.6. Raise it for stronger character identity, lower it to keep the pass 1 composition.</li><li>A little <code>padding</code> (20 to 40 px) plus some <code>feather</code> helps with subjects that spill outside their box.</li></ul>Limitations<ul><li>It’s built for up to 5 characters. If you need less, just bypass some nodes, if you need more, you'll have to hook more from the custom node yourself.</li><li>I’m guessing at the format of the <code>bboxes</code> output, so the node accepts several formats. It works on my setup (ComfyUI 0.42.5), but let me know if it can’t read your boxes.</li><li>Character LoRAs need to be trained on MOSS-Image to apply properly.</li><li>Due to the nature of MOSS's bbox node, this only works with Nodes 2.0</li></ul>Download: [workflow JSON] / [custom node .py] extract the moss_bboxes_to_masks.py and drop the file itself into your custom nodes folder.Feedback welcome, especially if you get it working with three or more characters.

Files with SHA256: 904cc824c3dcf49950923d625ca10ced8d3f7aac7a5e8d2fe6bd759cd8970cfd