Make pictures with MOSS-Image, the new open model that writes the words you give it. Signs, menus, posters and
labels come out spelled right when you put the words in "double quotes". Three switches help: Auto prompt writes a
full prompt from a few words, Start from a picture describes a picture of yours and adds your changes, and
Layout boxes puts things and words where you draw them. A Steps pill picks 50 steps, or 8 or 4 with a speed LoRA for a picture in about two seconds.
<h3>How to use it</h3>
<ol>
<li>Type your prompt. Put every word you want in the picture in "double quotes".</li>
<li>Set the size in Latent Size.</li>
<li>Run.</li>
</ol>
<h3>The switches</h3>
<ul>
<li>
Auto prompt. The text encoder MOSS already uses turns a few words of yours into a full prompt, and writes out
 every word that belongs on a sign, a menu or a poster. What it wrote shows in Show Text. A second or two more, no
 extra download.
</li>
<li>
Start from a picture. Load a picture of yours. The encoder describes it and adds what you typed ("make it a rainy
 night", "the sign says LAST TRAIN"), then MOSS draws a new picture from that description. A slider shows yours beside
 the new one. MOSS draws the scene again rather than editing your picture, so faces and small things change.
</li>
<li>
Layout boxes. Draw a box for each thing or line of words and say what goes in it. Your prompt still describes
 the whole picture. The box canvas draws in the Nodes 2.0 view (the toggle at the top left of ComfyUI).
</li>
</ul>
All three are off as the file opens.
<h3>Steps</h3>
<ul>
<li>50 steps: the settings OpenMOSS gives, the most detail. About 18 seconds for 1.5 megapixels on an RTX 5090.</li>
<li>8 steps or 4 steps: a speed LoRA makes the picture in about 2 seconds or 1.5. It comes out a little softer than 50 steps.</li>
<li>Only one can be picked. The pill sets the steps and CFG and turns the matching row on in LoRA Loader. If its LoRA is missing, the run stops and names the file.</li>
</ul>
50 steps is picked as the file opens.
<h3>What you need</h3>
<ul>
<li>
ComfyUI's development version from 2026-10-16 or newer. MOSS-Image support is merged but not in a release yet.
 Windows portable: run <code>update/update_comfyui.bat</code>, not the <code>_stable</code> one. A git install: <code>git pull</code> on master, then
 <code>pip install -r requirements.txt</code>. The Desktop app gets it with ComfyUI's next release.
</li>
<li>
AusBoss nodes 2.13.0 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first.
</li>
<li>The MOSS-Image files from <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image">Comfy-Org/MOSS-Image</a>:<ul>
<li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/diffusion_models/moss_image_sft_int8_convrot.safetensors">moss_image_sft_int8_convrot.safetensors</a> → <code>models/diffusion_models</code> · 5.41 GB</li>
<li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_int8_convrot.safetensors">qwen3.5_9b_int8_convrot.safetensors</a> → <code>models/text_encoders</code> · 9.75 GB</li>
<li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/vae/flux2-vae.safetensors">flux2-vae.safetensors</a> → <code>models/vae</code> · 336 MB</li>
</ul>
</li>
<li>For 8 and 4 steps, the speed LoRAs → <code>models/loras</code> · 1.19 GB each:<ul>
<li><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/resolve/f95df83b5e90ae91ea412434b3f9b7f1c8c6314f/lora_latest.safetensors">lora_latest.safetensors</a> by lodestones, for 8 steps</li>
<li><a target="_blank" rel="ugc" href="https://huggingface.co/akhaliq/MOSS-Image-SFT-4-steps/resolve/5c6b125d1a1ae8847f0b44e4cb1cb10d38354ffd/lora.safetensors">MOSS-Image-SFT-4-steps</a> by akhaliq, for 4 steps. It downloads as <code>lora.safetensors</code>: rename it <code>MOSS-Image-SFT-4-steps.safetensors</code>.</li>
</ul>
</li>
</ul>
The INT8 file gives the same picture as the full-size one (<a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/diffusion_models/moss_image_sft_bf16.safetensors">moss_image_sft_bf16</a>, 10.2 GB), about 2.4 times faster. For a smaller card there is a smaller text encoder, <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_w4a8_gptq.safetensors">qwen3.5_9b_w4a8_gptq</a> (7.34 GB): pick it in Text encoder.
<h3>Tips</h3>
<ul>
<li>
Spell out every word. MOSS writes exactly the words you give it. Words it has to make up come out as gibberish,
 so put them in quotes, or let Auto prompt write them. Up to about 40 words came out right in my tests; a full page of
 text is too much.
</li>
<li>
Write "a young woman in her twenties" rather than "girl". MOSS draws a child for "girl". Auto prompt does this
 for you.
</li>
<li>
Start from a picture draws the person again from a description. If they come out looking different, say how
 they look in your words: "she has pale skin and freckles".
</li>
<li>
Faint vertical stripes on a smooth sky or wall are a known quirk of this model. MATLOWAI's two
 <a target="_blank" rel="ugc" href="https://huggingface.co/MATLOWAI/MOSS-Image-SFT-DeBand">DeBand LoRAs</a> together, both at strength 1 in LoRA
 Loader, removed them in my tests; one alone barely helps.
</li>
<li>
Settings: 50 steps at CFG 3 with layer 22 skipped on the negative, the settings OpenMOSS gives. On an RTX 5090 a
 1.5 megapixel picture takes about 18 seconds, about 20 with Auto prompt or a picture. 8 and 4 steps run at CFG 1, as the speed LoRAs ask.
</li>
<li>
Size: 1 to 2 megapixels works best. The picture keeps the size you set in Latent Size, also with Start
 from a picture.
</li>
</ul>
<h3>What it does not do</h3>
<ul>
<li>
No editing. MOSS-Image draws new pictures from words. Start from a picture describes yours and draws a new
 one; it does not change your picture in place.
</li>
<li>
No upscale in this version.
</li>
</ul>
<h3>Credits</h3>
MOSS-Image is by the OpenMOSS team, under the Apache 2.0 license. The ComfyUI files are Comfy-Org's. The speed LoRAs are by lodestones and akhaliq, under Apache 2.0.
v1.1 (2026-10-18): the Steps pill, 50, 8 or 4 steps. v1.0 (2026-10-18): first release.
Questions or bugs: <a target="_blank" rel="ugc" href="https://github.com/ausboss/ComfyUI-AusBoss/issues">GitHub issues</a> or the comments here. Videos on <a target="_blank" rel="ugc" href="https://www.youtube.com/@ausbossdev">YouTube @ausbossdev</a>. More from me on <a target="_blank" rel="ugc" href="https://x.com/Zanzibased">X @Zanzibased</a> and <a target="_blank" rel="ugc" href="https://github.com/ausboss">GitHub</a>.

MOSS-Image Text to Image

Steps pill: 50, 8 or 4 steps. 8 and 4 use a speed LoRA and make a picture in about 2 seconds or 1.5 on an RTX 5090.<ul><li>One is picked at a time. The pill sets the steps and CFG and turns the matching LoRA row on in LoRA Loader.</li><li>Get the speed LoRAs from the note: lodestones' <code>lora_latest.safetensors</code> for 8 steps, and akhaliq's 4-step LoRA for 4 steps. The 4-step file downloads as <code>lora.safetensors</code>: rename it <code>MOSS-Image-SFT-4-steps.safetensors</code>.</li><li>A missing speed LoRA stops the run and names the file.</li></ul>Needs ComfyUI's development version from 2026-10-16 or newer and AusBoss nodes 2.13.0 or newer.

Files with SHA256: 34af5279bff4ff59bde31bfc592087134285f520bfbc197157c37acabcc150bc