<div data-youtube-video="https://youtu.be/09qEF5oo8Gc"><iframe src="https://www.youtube.com/embed/09qEF5oo8Gc" width="853" height="480" allowfullscreen></iframe></div> 
 
Compare MOSS-Image Base and SFT under matching generation settings in one ComfyUI graph. Both active branches lead to their own SaveImage output so the results can be reviewed side by side. 
 
Main features: 
- Separate Base INT8 ConvRot and SFT INT8 ConvRot model loaders; shared Qwen3.5 9B BF16 text encoder and FLUX.2 VAE. 
- Each output-reachable branch uses Euler, 50 steps, CFG 3, simple scheduler, and fixed seed 777. 
- One sample ceramic teapot prompt and a shared negative prompt. No LoRA is connected. 
 
Suggested workflow: 
1. Install both MOSS-Image model files plus the text encoder and VAE, then load the JSON in ComfyUI. 
2. Keep the prompt, seed, steps, and CFG aligned while generating both branches. 
3. Compare the two saved images for detail, composition, and prompt response; repeat with your own prompts. 
 
Required model filenames: moss_image_base_int8_convrot.safetensors; moss_image_sft_int8_convrot.safetensors; qwen3.5_9b_bf16.safetensors; flux2-vae.safetensors. This is a comparison workflow, not a measured guarantee that either model wins. 
 
RunningHub AI workflow: https://www.runninghub.ai/zh-cn/post/2113906846670061569?inviteCode=rh-v1111 
RunningHub signup: https://www.runninghub.ai/?inviteCode=rh-v1111 
Bilibili video: https://www.bilibili.com/video/BV1ZHHP6FEVZ/ 
Related model files: https://pan.quark.cn/s/20e19307e8ef

MOSS-Image Base vs SFT Comparison

<div data-youtube-video="https://youtu.be/09qEF5oo8Gc"></div> 
 
Compare MOSS-Image Base and SFT under matching generation settings in one ComfyUI graph. Both active branches lead to their own SaveImage output so the results can be reviewed side by side. 
 
Main features: 
- Separate Base INT8 ConvRot and SFT INT8 ConvRot model loaders; shared Qwen3.5 9B BF16 text encoder and FLUX.2 VAE. 
- Each output-reachable branch uses Euler, 50 steps, CFG 3, simple scheduler, and fixed seed 777. 
- One sample ceramic teapot prompt and a shared negative prompt. No LoRA is connected. 
 
Suggested workflow: 
1. Install both MOSS-Image model files plus the text encoder and VAE, then load the JSON in ComfyUI. 
2. Keep the prompt, seed, steps, and CFG aligned while generating both branches. 
3. Compare the two saved images for detail, composition, and prompt response; repeat with your own prompts. 
 
Required model filenames: moss_image_base_int8_convrot.safetensors; moss_image_sft_int8_convrot.safetensors; qwen3.5_9b_bf16.safetensors; flux2-vae.safetensors. This is a comparison workflow, not a measured guarantee that either model wins. 
 
RunningHub AI workflow: https://www.runninghub.ai/zh-cn/post/2113906846670061569?inviteCode=rh-v1111 
RunningHub signup: https://www.runninghub.ai/?inviteCode=rh-v1111 
Bilibili video: https://www.bilibili.com/video/BV1ZHHP6FEVZ/ 
Related model files: https://pan.quark.cn/s/20e19307e8ef

Files with SHA256: fb96316fcc57e5e871bd414d31c0fa02d890bef6942bc88019538f70db99b737