Generate Prompt
Generated Prompt
Preset Examples
Anime Portrait
Turbo mode, anime style, portrait composition
Product Photo
Full model, photorealistic + soft studio lighting
Text Sign Poster
Full model, render text "OPEN", negative "blurry, watermark"
How to Write Better Kromoss Prompts
- Put rendered text in quotes. Use exact wording you want to appear, e.g., "KROMOSS" so the model can focus on rendering it legibly.
- Negative prompts need CFG above 1. Turbo LoRA (CFG 1) ignores negative branches — switch to Full model to use them.
- Turbo LoRA = 8 steps CFG 1 for speed. Use for quick iterations; Full model gives higher quality with negatives.
- Native 1024px — keep aspect multiples of 16. The model prefers dimensions divisible by 16 for cleaner generation.
- English prompts recommended. Clear, descriptive language produces more reliable results.
- Drop-in ComfyUI replacement for moss_image_sft_bf16.safetensors. Same loader files — just swap in kromoss_bf16.safetensors; with the LoRA use KSampler euler/simple, 8 steps, CFG 1.
FAQ
What is Kromoss?
A full fine-tune of MOSS-Image-Base by lodestones, producing sharper images. Apache-2.0 licensed. Unofficial fan-made tool.
Full model vs Turbo LoRA?
Full model: 50 steps, CFG 3, with negative-branch layer skipping. Turbo LoRA: 8 steps, CFG 1, no negative branch — negatives are ignored.
Why is my negative prompt ignored?
Turbo LoRA mode (CFG 1) has no negative branch. Use Full model to apply negative prompts.
Where can I run Kromoss?
Hugging Face diffusers pipeline, native ComfyUI support, and Civitai. Use kromoss_bf16.safetensors in place of moss_image_sft_bf16.safetensors.
Can Kromoss render text in images?
Yes — put the text you want rendered inside quotes, e.g., "KROMOSS" on a sign.
Is this tool affiliated with lodestones?
No. This is an unofficial fan-made tool and is not affiliated with lodestones.
NewModelHub