#6161st of 120 in Image and videoFor image, video and design, image and design
moss-image
5B text-to-image diffusion model with layout control and text rendering
- Stars
- 3k
- License
- Apache-2.0
- Last commit
- Nov 2026
- Language
- Python
37 on 14 Oct 20263,006 on 18 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
MOSS-Image provides the Base and SFT text-to-image models plus a Diffusers pipeline, a Gradio app and a ComfyUI plugin. It accepts prompts up to 4K tokens and JSON layouts with bounding boxes to place text and objects. Memory modes allow generation with about 3.65 GiB of GPU memory at 1024×1024, at the cost of 14 GiB system RAM.
Who it is for: Self-hosters generating images with layout-controlled text on local GPUs
Strengths
- JSON layout input places text and objects by bounding box
- Low-memory profile peaks at 3.65 GiB GPU for 1024×1024, 50 steps
- Day-0 support in SGLang, vLLM-Omni and native ComfyUI
- Community LoRA training, DMD distillation and a C++/CUDA engine exist
Weaknesses
- Low-memory mode needs about 14 GiB system RAM and an SSD
- Default sampling is 50 steps; the 8-step distillation is community-made
- Apple Silicon support has documented limitations, tested only on M1 Max 32GB
- Compile DiT is unavailable in Sequential and Low memory modes
What it needs
- RAM ≥ 14 GB
- GPU optional
- Needs PyTorch, Diffusers, Gradio, ComfyUI
- Models: MOSS-Image-Base, MOSS-Image-SFT, Diffusers, ComfyUI format
Known problems
Open now, reported in the last 90 days
- Caption Generation Pipeline in MOSS-Image ↗ (opens in a new tab)3 reactions
- SFT always blurs the background: we traced it to the attention outputs of blocks 21–41 ↗ (opens in a new tab)1 reactions
- A Related Work on Layout-Controlled Image Generation ↗ (opens in a new tab)1 reactions
Titles are the reporters' own words, taken from the project's GitHub issues and ranked by reactions. We have not tried to reproduce any of them.
Also in Image and video
See all 120| Rank | Project | Score |
|---|---|---|
| 1 | frigateLocal NVR with realtime object detection for IP cameras | 94 out of 100 |
| 2 | toonflow-appInfinite-canvas tool for AI short drama and comic video production | 89 out of 100 |
| 3 | MoneyPrinterTurboGenerates short videos from a topic, with script, footage, voiceover and subtitles | 88 out of 100 |
| #4 | InvokeAIWeb UI for diffusion image generation with inpainting canvas and node workflows | 88 out of 100 |
| #5 | autoclipTurns a video link into short clips with covers and captions | 88 out of 100 |