Connecting AI, humans, and the physical world. TTS|ASR|VL |Audio|Video|Contextual Intelligence ⚡️Vibe with us: https://xitter.cf/t.co/Lmk4Kd65JJ

Joined January 2026
👀 want EARLY ACCESS to our next models? join the MOSS Discord for upcoming early access opportunities, help with MOSS-Image, and ideas from fellow creators. love seeing the community's LoRAs already 🫶 come show us yours join MOSS discord 👉 discord.gg/z7TD5ehYts
17
12
1
74
4,726
MOSS retweeted
best OSS image model
MOSS-Image hit 673K downloads on Hugging Face's Comfy-Org repo just 3 days after release. 🚀 🥇 MOSS-Image-Base hit #1 on Hugging Face's Text-to-Image Base Trending list.@huggingface Both Base and SFT weights are open, with Day-0 support for @sgl_project @vllm_project @ModelScope2022 Try it ↓
1
1
29
My new favorite image model! True open source, base and finetuned models, uncensored, LoRA-friendly... A perfect release!
🤗 MOSS-Image is now open source on @huggingface . Built on a 5B Diffusion Transformer (DiT), MOSS-Image brings Chinese & English text rendering and controllable layouts to image generation — from posters and infographics to creative visuals.@Open_MOSS What it offers: 📝 Long prompts & text rendering — up to 4K-token prompts, including multi-line text 📐 Experimental JSON layout control — specify text, objects, and their positions using bounding boxes 🖼️ Flexible generation — multiple aspect ratios, resolutions, and customizable inference settings 🛠️ Developer workflows — Python (Diffusers), Gradio, ComfyUI, and a Prompt Enhancer Skill 📦 Base & SFT checkpoints are both available, with open model weights and Apache-2.0 licensed inference code. 🙌 Huge thanks to @sgl_project @vllm_project @ModelScope2022 for the Day-0 support!
1
2
1
10
532
I made this entire film with Codex and MOSS Image. @MosiAI_Official @Open_MOSS
1
2
1
3
164
MOSS-Image hit 673K downloads on Hugging Face's Comfy-Org repo just 3 days after release. 🚀 🥇 MOSS-Image-Base hit #1 on Hugging Face's Text-to-Image Base Trending list.@huggingface Both Base and SFT weights are open, with Day-0 support for @sgl_project @vllm_project @ModelScope2022 Try it ↓
3
1
1
45
757
MOSS-Image API is coming soon in 24hrs Join MOSS discord for FREE early access 👉 discord.gg/Wjku5AYyxx
4
3
1
45
1,087
MOSS Image prompts: Mounds of raw pigment form a courtyard stair in ochre, cobalt, and vine black, the powder shaped into broad steps. An ochre mound fills the near corner, bright and slightly dusty at the edges. A few loose grains have spilled onto the stone in front of that lowest mound. Cobalt and black mounds climb behind it toward a plain cream wall. Open daylight falls from above, even and clear, and a soft shadow sits in each tread. Ochre, cobalt, vine black, and cream keep the picture bold. A 35mm lens makes this an art photograph, graphic and calm. Powdered pigment hangs in a loose circle, lifted by a soft current of air and frozen against black. The lifted dust is the only thing visible, a thin ring with a darker core. The nearest grains are sharp specks of vermilion and yellow, separate and bright. The circle thins as it moves back into the black and finally disappears. A hard light from the left catches the powder and leaves the ground dark. Vermilion, ultramarine, yellow, and white spark against that black. An 85mm lens makes this a color-burst photograph, stopped in midair. Rose and turquoise ink bloom through a tank of clear water, caught while the plumes are still separate. A rose plume curls nearest, the edge feathered and sharp at once. Turquoise rises behind it, and the two colors meet in a thin indigo seam. The far side of the tank fades to clean transparency, with nothing else in the water. Side light from the left makes the inks glow without muddying them. Rose, turquoise, indigo, and silver-white highlights keep the abstract elegant. An 85mm lens gives this the precision of a high-speed photograph, luminous and exact. Indigo and coral silk, patterned in shibori, drapes over a plain plaster block in a quiet studio. The block is simple and square, and the cloth lies still across the top and one side. A coral fold crosses the near corner, with resisted white lines easy to see in the weave. A seamless pale ground sits behind the block and keeps the attention on the cloth. Soft studio light comes from the left and shapes both silk and plaster. Indigo, coral, plaster white, and pale gray shadow define this textile photograph. A 50mm lens keeps the drape and the block clear. #MOSSImage
1
3
320
We've got Midjourney at home. This is crazy
MOSS-Image Base vs SFT both Seed 42 prompts below
1
3
194
NGL, I really like the new Moss-image model. @MosiAI_Official
1
5
177
Testing out the MOSS by @MosiAI_Official Mainly checking out style flexibility, which is what I care about. Styles here are: 3D Render Pixar Animation Fortnite Style Modern Super Mario (Nintendo) Style Will keep posting. Thoughts and prompt in feedback.
28
1
36
1,842
Sprawdziłem u siebie w domu MOSS-Image, który pojawił się w piątek, i porównałem go z modelem Z-Image-Turbo, którego używam na co dzień: te same prompty, te same seedy i ten sam komputer (ASUS Ascent GX10 z układem NVIDIA GB10). Każdy model działał na ustawieniach zalecanych przez twórców (Z-Image-Turbo w 9 krokach, MOSS-Image w 50), więc różnica w czasie jest duża: średnio 12 sekund na obraz wobec 135, czyli mniej więcej 11 razy dłużej. Na slajdach są obrazy dla seeda 42: po lewej zawsze Z-Image-Turbo, po prawej MOSS-Image. Co zauważyłem na 10 promptach: MOSS-Image dokładniej oddaje szczegóły z polecenia (np. okulary zsunięte na czubek głowy, a nie na nosie, albo błyski na kolczykach), ale dorysowuje też rzeczy, o które nie prosiłem (w tle portretu kobiety stoi bus, który pewnie wziął się z frazy „vintage van mural” w opisie stylu). Z polskimi znakami oba poradziły sobie dobrze, choć na plakacie z wyprzedażą każdy z nich zrobił jeden błąd (Z-Image-Turbo napisał „zl” zamiast „zł”, a MOSS-Image „wyprzedaź” przez „ź”). Licencja nie przesądza o wyborze: według kart modeli na Hugging Face oba są udostępnione na licencji Apache 2.0, która pozwala także na użycie komercyjne. Na razie nie mam faworyta (wydaje mi się, że każdy z tych modeli ma swoje mocne i słabe strony), więc pewnie popracuję jeszcze trochę z oboma, żeby się przekonać, który bardziej mi odpowiada. Oczywistą zaletą Z-Image-Turbo jest czas generowania.
🤗 MOSS-Image is now open source on @huggingface . Built on a 5B Diffusion Transformer (DiT), MOSS-Image brings Chinese & English text rendering and controllable layouts to image generation — from posters and infographics to creative visuals.@Open_MOSS What it offers: 📝 Long prompts & text rendering — up to 4K-token prompts, including multi-line text 📐 Experimental JSON layout control — specify text, objects, and their positions using bounding boxes 🖼️ Flexible generation — multiple aspect ratios, resolutions, and customizable inference settings 🛠️ Developer workflows — Python (Diffusers), Gradio, ComfyUI, and a Prompt Enhancer Skill 📦 Base & SFT checkpoints are both available, with open model weights and Apache-2.0 licensed inference code. 🙌 Huge thanks to @sgl_project @vllm_project @ModelScope2022 for the Day-0 support!
1
1
4
295
MOSS Image prompt: Saudek style - 24 50 vs 8 dmd steps Strawberry-blonde waves and freckles belong to a beautiful young woman in a simple ivory slip, a branch of orange blossom held across her body. Her fingers stay loose and calm, her look quietly deadpan. A Mediterranean wall stands as a flat theatrical canvas, sun-faded color painted on, a side batten in view. Pale floorboards and the corner of a low stool sit at the bottom of the frame. Warm key light from above and slightly left gilds her hair while the far cheek falls soft. Citrus orange, blossom white, and ivory are brushed on in saturated hand tint. Warm analog grain and an 80mm medium-format lens keep the portrait intimate. #MOSSImage
1
1
8
444
MOSS-Image 5B parameters of power and beauty. Thanks @Open_MOSS for this amazing model! On a single DGX Spark 1024x1024 images: - ~48 secs for 50 steps (MOSS official) - ~30 secs for 25 steps (ComfyUI recipe) I love the official recipe more. You?
6
3
78
6,687
Moss-Image is pretty good at complex prompts (Prompt below)
MOSS Image Base vs SFT - seed 42 - 50 steps hometown theme 🧋😋 I prefer SFT with left side element and layout, but the human I prefer Base - looks sooo natural, looks really alike to my high school classmates in general, i think Base has the best natural girls than any other image gen models, variety of faces, and natural skin tone - even closed models like seedream, gpt etc has a very obvious AI girl look & skin texture if you also like natural human gen, try MOSS Image Base as usual, prompts below 👇
3
3
1
68
2,310
▼話題 画像生成モデル「MOSS-Image」の開発チームによる音声系モデル(MOSS-Transcribe-Diarize等)の高性能ぶりと、MOSS-Image自体の評価を巡る議論。 ▼主な反応 ・従来上位モデルの約1割のサイズで高精度な音声認識モデルや、MOSS-TTSなど同チームの音声系モデルを評価する声 ・MOSS-Imageの表現制限のない仕様やシード多様性、10〜16GBのVRAM環境で扱える実用性に期待する意見 ・MOSS-Imageはサイズ相応の並の品質であり、過熱気味の評判はボットや宣伝工作によるものだと疑う見方 ▼受け止め 音声系モデルが好意的に受け止められる一方、MOSS-Imageは中容量VRAM層からの期待と過大評価への懐疑論で賛否両論。 #MOSSImage #audiocpp URLはリプ⬇️
2
10
66
4,880
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***
1
21
5
226
31,461