MOSSing around in 4K
I've been doing some testing to check the creative variety of MOSS image.
Proof1 community thread
Key points
- It has got quite some cinematic looks in it.
- Nothing fancy about the workflow, just straight from the comfy template.
- diffusion model: mossimagesftint8convrot.safetensors text econder: qwen3.59bint8convrot.safetensors vae: flux2-vae steps: 50 cfg: 3 sampler: euler scheduler: sgmuniform / simple With these settings 4k res is rendering in about 55 sec on a GTX5090 with peak VRAM 19,5 GB.
Sources (1)
- [1]MOSSing around in 4Kr/StableDiffusion (top, day) · Oct 17, 01:49 PM
I've been doing some testing to check the creative variety of MOSS image.
It has got quite some cinematic looks in it.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 17, 2026Nemotron 3.5 Bolt, Muse Spark Open, and a Fast LFM2.5 24B
- Oct 16, 2026VaeZip: store VAE-encoded latent datasets in 5-10x fewer space with almost no quality loss
- Oct 16, 2026MOSS-Image (5B, Apache 2.0) on a 12 GB RTX 3060: INT8 quant + lodestones' 8-step DMD turbo in ComfyUI - 35 hard prompts, one seed, no cherry-picking
- Oct 16, 2026MOSS-Image Technical Report
- Oct 15, 2026QUASAR 4-bit QAT checkpoints for Qwen3.5, Gemma 4 and Muse Glimmer: closer to BF16 than Google’s and NVIDIA’s 4-bit releases
- Oct 15, 2026huggingface/transformers v5.20.0: Release v5.20.0
